Professional Documents
Culture Documents
NUEL BELNAP
MICHAEL PERLOFF
M I N G XU
With Contributions by
Paul Bartha
Mitchell Green
John Horty
OXPORD
UNIVERSITY PRESS
2001
OXFORD
U N I V E R S I T Y PRESS
Oxford New York
Athens Auckland Bangkok Bogota Buenos Aires Cape Town
Chennai Dar es Salaam Delhi Florence Hong Kong Istanbul Karachi
Kolkata Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi
Pans Sao Paulo Shanghai Singapore Taipei Tokyo Toronto Warsaw
and associated companies in
Berlin Ibadan
9 8 7 6 5 4 3 2 1
Printed in the United States of America
on acid-free paper
Preface
This is a book about the causal structure of agency and action. It frames a
rigorous theory by using techniques and ideas from philosophical logic, philos-
ophy of language, and metaphysics with a small "m." This theory, which we
sometimes call "the theory of agents and choices in branching time," describes
agents as facing a future replete with real possibilities, some of which various
agents realize by making choices. It is central to our theory that choices and
the actions that they ground are radically indeterministic: Before an event of
choosing, there are multiple alternatives open to the agent. Furthermore, since
the choice is real, so must be the alternatives, and each alternative must be as
real as any other. All we can say before the moment of choice is that the agent
will make one of the open choices, leaving behind the unchosen alternatives.
After the choice, it is correct to say that they were once possible, but are no
longer possible. None of the possible choices is a mental or linguistic figment,
nor is any a mere ghost image of "the actual choice." Given that the possibil-
ities relevant for action are always possibilities for our future, the theory also
refrains from appealing to "possible worlds" other than the one and only world
that we all inhabit.
These ideas are in some part rooted in common sense. Without help, however,
common sense cannot seem to pull them together into a coherent whole. One of
our principal aims is to carry out that job by articulating them in a completely
intelligible exact theory. The resultant theory of agents and choices in branching
time pictures the causal structure of our world as made up of alternative courses
of events branching tree-like toward the future. Each branch point represents
a choice event or chance event. On the one hand, each continuation from a
branch point is individually possible; on the other hand, it is impossible that
more than one of these continuations should be realized. If that sounds obscure,
we agree: It is, we think, almost impossible to speak clearly and accurately about
indeterminism except in the framework of a rigorously fashioned theory such as
the one we propose.
The theory of agents and choices in branching time is of real but limited
interest without its application to understanding the language of action. We
propose a certain linguistic form, a "modal connective," as being unusually
helpful to anyone who wishes to think deeply about agents and their actions.1
1
Our usage here is common but not universal. A connective is any grammatical construe-
vi Preface
The form is "a sees to it that Q," where a names an agent and Q holds the
place of a sentence; for example, we think of "Ahab sailed the seven seas" as
"Ahab saw to it that he sailed the seven seas." The form is so important to our
enterpriseall but three of our eighteen chapters are devoted to its study or
usethat we give it an abbreviation, "[a stit: Q]" with "stit" as an acronym
for "sees to it that." Stit theory explains the meaning of [a stit: Q] in the idiom
of philosophical logic. In doing so, stit theory invokes a certain melding of the
Prior-Thomason indeterministic semantics with Kaplan's indexical semantics.
The combined semantics make [a stit: Q] roughly equivalent to "a prior choice
by a guaranteed that Q."
The stit idea is many-sided. We explore its grammar, semantics, and proof
theory as logicians do. We delight in the fact that stit does not treat actions pri-
marily as "things" to be counted or named. We explore some of the linguistics
of stit (especially how and why its status as a modal connective lends itself to
usefully complicated constructions), we consider some applications to difficult
conceptual problems, and we argue the ability of stit to illuminate agency in a
variety of ways. We look at some ways in which stit might be modified or gen-
eralized. In all of this, however, we try never to forget the central constraining
thought: There is neither action nor agency nor doings without real choices,
choices that find their place not merely in the agent's mind, but within the
(indeterminist) causal order of our world. To see to it that Q, an agent must
make a real choice among objectively incompatible future alternatives.
When we say that an event may have many possible but incompatible out-
comes, we thereby come down on the side of "hard" indeterminism as against
determinism. There is no consensus on these ideas. Since the eighteenth cen-
tury became understandably awed by the success of Newtonian science, the
presumption of determinism has guided most of the philosophical and scientific
explorations of both agency and nature. Taking determinism to be delivered
by science as an unquestioned "fact," philosophers since Hume and Kant have
worked at developing "compatibilist" theories that hold agency, in the guise of
moral responsibility, to be compatible with what James called the block uni-
verse. Such theories have often taken possibilities as unreal: as arising from the
mind, or from social practices, or from languagefor example from consistency
with the bits of language called "scientific laws." Our contrasting indeterminis-
tic presumption is eloquently expressed by the eminent paleontologist Stephen
Jay Gould.
liberating truth that tiny inputs, virtually invisible and risibly impo-
tent in appearance at the outset, can cause history to cascade down
any route in a vast array of entirely different pathways. (Gould 1999,
p. 30)
This book neither argues for indeterminism nor tries to pick holes in argu-
ments for compatibilism. Our project assumes the indeterminism of the causal
order in which agency is embedded, it assumes that actions are based on real
choices, and it assumes that choices are therefore not predetermined. Our goal
is not to persuade, but to make these ideas intelligible. Although numerous
philosophers share our general point of view, not many exact theories share these
assumptions and aims. Our strategy is to concentrate almost exclusively on the
objectively causal side of indeterminism and agency, which already presents
enough difficulties without bringing in noncausal concepts. We therefore lay
aside many deeply important aspects of agency and choice that involve inten-
tions, propositional attitudes, or other mental phenomena.
We look for ways in which applications of stit theory can engender a better
understanding of agency. Seven examples: (i) an analysis of refraining that clar-
ifies how it can be both a doing and a not-doing; (ii) an analysis of imperatives
that emphasizes their agentive content; (iii) an extended treatment of deontic
logic that insists that obligations and permissions (a) are directed to agents
capable of making choices, and (b) are embedded in the indeterministic causal
order of our world; (iv) fresh analyses of promising and of assertion, analyses
that argue the unwisdom of the doctrine that among all the objective possibili-
ties, a unique course of events constitutes the one and only "actual future"; (v)
an exploration of the causal side of the requirement on action that the agent
"could have done otherwise"; (vi) the causal structure of joint agency; and (vii)
a generalization of stit theory to strategies considered from a causal point of
view.
In sum: Holding the extra-mental and extra-linguistic status of incompatible
possibilities as given, and supposing that the future sometimes depends upon
an agent's choices among incompatible options, we offer a tense-modal theory
intended to describe some causal aspects of agency in our indeterministic world.
of the deliberative stit from the achievement stit. Largely concentrating on the
achievement stit, the various chapters of part I suggest applications with the
help of many pictures, make comparisons with some other work on agency, and,
beginning with Anselm's work in 1100, give a little history of the modal logic of
agency. This part also contains applications of stit theory to imperatives and
to promising.
Part II supplies precisely and in detail the nuts and boltsor, more aptly,
roots and branchesof the theoretical structure that supports our account of
agents, actions, and our indeterminist world. The three chapters of this part
are foundational in character, and involve substantially more rigor, though not
much more mathematics. They stress conceptual analysis rather than theorem-
proving. This part more than any other focuses on the problems faced by
any indeterministic theory. Here we argue against the beguiling but harmful
doctrine of "the actual future," which says that among the many courses of
events that might come to pass, there now exists a privileged such course that
will actually do so. This foundational part examines, postulate by postulate,
the theory of agents and choices in branching time, and explains in detail the
semantic subtleties required of a language spoken in an indeterminist world.
Part III offers two applications of the achievement stit: One chapter aims to
illuminate the dark idea of "could have done otherwise," and another considers
the causal aspects of joint agency. These chapters are a little more technical.
Part IV is of the same level of technicality as part III: It offers applications
of the deliberative stit, chiefly to help in elaborating such deontic concepts as
obligation and permission, which, we believe, are in much need of a theory of
agency.
Part V uses the already-established theoretical structure of agents and choices
in branching time in order to develop an austere (causal, not normative) account
of strategies as a kind of generalization of stit. One chapter in this part connects
our theory of strategies to Thomason's deontic kinematics. Part V proves a
theorem or two, though much of it is, again, conceptual analysis.
Part VI provides the technical backbone of stit theory, including proofs of de-
cidability, soundness, and completeness. The chapters of this part are required
reading for those who wish to investigate or develop the logical and mathemat-
ical properties of theories of agency similar to ours.
The appendix gathers, for easy reference, most of the various theses, struc-
tures, postulates, definitions, semantic ideas, and systems that are introduced
elsewhere and are employed throughout the book. We use boldface to refer to
certain of these items: Look in the appendix for its sections l-9, for stit the-
ses Thesis 1-Thesis 6, for postulates Post. 1-Post. 10, for definitions Def.
1-Def. 20, and for axiomatic concepts Ax. Conc. 1-Ax. Conc. 3.
Although in each section of this book we feel free to refer to any other section,
it may be useful to indicate the following dependency-structure among the vari-
ous parts. Parts I (introduction to stit), II (foundations of indeterminism), and
VI (proofs and models) are almost entirely independent. The reader's primary
interests may therefore be allowed to determine with which of parts I, II, or VI
it is best to begin. Part III (applications of the achievement stit) and part IV
Preface ix
I Introduction to stit
1 Stit: A canonical form for agentives 3
1A Agentives . 5
1B Stit: Simple cases . 9
1C Grammar of the modal logic of agency . 14
1D Mini-history of the modal logic of agency . 18
1E Conclusion and summary . 26
II Foundations of indeterminism
6 Indeterminism and the Thin Red Line 133
6A Preliminary considerations . 134
6B Parameters of truth . 141
6C The assertion problem . . 156
6D The Thin Red Line . . 160
6E Time's winged chariot hurries near , . 170
V Strategies
13 An austere theory of strategies 341
13A Nature of austere strategics . 342
13B Review of choices in branching histories . 344
13C Elementary theory of strategies . 345
13D Favoring . 356
13E Application to finding a strategy for inaction . 359
14 Deontic kinematics and austere strategics 364
14A Basic concepts . 365
14B From Thomason's deontic kinematics to austere strategics . 368
14C From austere strategics to Thomason's deontic kinematics . 370
14D Remarks . 376
Index 483
Part I
Introduction to stit
This page intentionally left blank
1
Stit: A canonical form
for agent ives
Among the topics of discussion in this world, none are more common than
those concerning the achievements and refrainings, obligations and prohibitions,
successes and failures of the agents with whom we share a common space.*
"What happened when so-and-so did that?" we ask, or "What should have
been done?" Biographies and narrative histories, which form a sizable segment
of our reading materials, have as their central concern the doings of agents,
their obligations and prohibitions, the outcomes of their choices, and the range
of things from which they refrain.
Philosophers have long sought to find a distinction between those sentences
which attribute agency and those that do not in the verbal configurations we
commonly use to talk about such matters. If such a distinction existed it should
be a relatively simple matter to uncover some general principle in everyday
speech to differentiate
Joshua fit the battle of Jericho
from
Joshua survived the battle of Jericho.
We find, to our dismay, that we are no closer now to a linguistic litmus test for
agency than was Aristotle. After all, it would have been of some importance for
Aristotle's peers to decide between an agentive interpretation of
Alexander succeeded to the throne of Macedonia
and a non-agentive interpretation. For if Alexander was agentive in the matter
of his becoming king, then there was a prior choice of his which led directly to
* With the kind permission of Theoma, Belnap and Perloff 1988 is the basis of each section
of this chapter except 1D. That section is drawn from Belnap 1991, for the use of which we
thank the International Phenomenological Society.
3
4 Introduction to stit
that outcome, and he was likely guilty of regicide; while if he was not agentive
in the matter of becoming king, then there was no choice of his that guaranteed
his succession to the throne.
J. L. Austin said that "The beginning of sense, not to say wisdom, is to realize
that 'doing an action', as used in philosophy, is a highly abstract expressionit
is a stand-in used in the place of any (or almost any) verb with a personal subject
..." (Austin 1961, p. 126). In his essay he tried to throw light on the question of
"doing an action" by looking at the range of cases in which excuses are offered
both in everyday usage and in the law, and to arrive at a proper vocabulary for
action by "induction" on the proper uses of words. Many years have passed,
the lesson has been learned, and it is time for philosophy to go beyond the mere
beginnings of sense and progress toward a deeper understanding of an agent
doing an action. How then should we proceed?
The Austin legacy is of course a particularly rich one that is being carried on
in a variety of ways. One way attempts to explain certain facts about language
in terms of propositional content and illocutionary forces: "speech act theory" as
represented for example by Searle and Vanderveken. That is not our program,
as we make clear in many places hereafter, suggesting that a theory of "speech
acts" should begin with a prior theory of acts. Accordingly we follow in the
wake of Austin's suggestion that when faced with the question of the meaning
of a word or a term such as "doing an action," we should "reply by explaining its
syntactics and demonstrating its semantics" (Austin 1961, p. 28). A suggestion
of this book is that the next step in the progression toward greater sense and
wisdom is to have available the sort of clean and well-honed linguistic resource
that Austin, and other philosophers, have realized to be necessary. We think
that the most promising path to a deeper understanding of an agent making a
choice among alternatives that lead to action is to augment our philosophical
language with a class of sentences whose fundamental syntactic and semantic
structures are so well designed and easily understood that they illuminate not
only their own operations but the nature and structure of the linguistic settings
in which they function.
An example of a doing-an-action sentence that Austin might have had in mind
is
Ahab sailed in search of Moby Dick. (1)
It has a personal subject, "Ahab," and an action-like verb, "sailed," and seems
to be describing an action in Austin's deliberately wide sense. We take (1) not
only to be true, but to be agentive for Ahab, for Ahab's sailing in search of
Moby Dick was a direct result of a choice he made among alternatives available
to him. On the other hand, although the perfectly ordinary sentence,
The Pequod sailed in search of Moby Dick (2)
is surely true, and though we may be hard pressed to say exactly why, we are
not hard pressed to say that it is not agentive: It does not even have a personal
subject. Consider now
1. Stit: A canonical form for agentives 5
1A Agentives
Let us accordingly begin with the following convention: The agentive form that
we are about to introduce shall be set off with square brackets [ ... ]. It shall
have two open places as indicated, the first to take an agent term, the second to
take a declarative sentence (the declarative complement of the new form). The
point about the second open place is nontrivial: Having noted that declarative
sentences can either ascribe agency or not, we specifically include as possible
declarative complements for the second open place both those sentences that
do ascribe agency and those that do not ascribe agency. The resulting square-
bracket sentence is to say that the proposition expressed by the declarative
complement is guaranteed true by a prior choice of the agent. So
[the carpenter ... Ahab has a new snow white ivory leg]
is to be agentive for the carpenter, and is to say that he is the agent in the
matter of Ahab's having a new snow white ivory leg.
With what verb or verb phrase shall we replace the ellipsis in that senten-
tial form (it was after all only elliptical)?1 Among the candidate English verb
phrases that history suggests are the following:
1
Numerous other philosophers have considered this or similar questions, including at least
Anderson, Aqvist, Bennett, Chellas, Chisholm, Danto, Davidson, Fitch, Hamblin, Hilpinen,
6 Introduction to stit
1B.1 Imperatives
"Clear away the boats! Luff!" cried Ahab. Treatment of such imperatives is by
no means tangential to our concerns. We endorse the view that C. H. Hamblin
expresses in his masterful study:
Imperatives are not only among the most frequent of utterances;
they are also, surely, among the most important. If the human race
had to choose between being barred from uttering imperatives and
being barred from uttering anything else, there is no doubt which it
should prefer. (Hamblin 1987, p. 2)
Hamblin additionally reports that a full twenty percent of Shakespeare is in the
imperative mood. (We happily acknowledge that a study of Hamblin's book set
us under way and helped us avoid some threatening reefs.)
Imperatives, in our usage, constitute a grammatical category. Following an
established tradition, however, we think of each use of an imperative as having
a force and a content. With regard to force, Ahab's imperatives may have been
orders or commands, which many think the only possibilities; but Ahab might
instead have been inviting, requesting, suggesting, advising, ... the helmsman
to luff. Putting force to one side, however, we are after the content, about which
stit theory has a definite and (as it happens) helpful opinion that we sum up in
the "imperative content thesis."
IMPERATIVE CONTENT THESIS. (Stit thesis. Reference: Thesis 4) Regardless
of its force on an occasion of use, the content of every imperative is agentive.
For example, Luff! can have its content represented as The helmsman luffs,
which in turn, since it is agentive, can be paraphrased as The helmsman sees to
it that he luffs. Thus, Luff! can be paraphrased as Helmsman, see to it that you
luff! In this case the application of our thesis is easy because The helmsman
luffs, which looks to be the most plausible content for the imperative, is already
agentive. Still, there is more to be learned. To luff is to see to it that the bow
of the boat is heading directly into the wind. Accordingly, the content of the
imperative
Luff!
can be put into the canonical form
[Helmsman stit: the helmsman luffs] (15)
or equivalently,
[Helmsman stit: the boat is headed into the wind]. (16)
The two stit sentences are equivalent, but while the complement of the former is
agentive, the complement of the latter doesn't mention the agent at all.2 Unlike
Luff! the imperative
2
"Complement" versus "content" sometimes sounds confusing. The (grammatical) com-
plement of [a stit: Q] is Q. Its (semantic) content is an agentive proposition.
1. Stit: A canonical form for agentives 11
does not show its content so obviously; the stit apparatus, however, increases
in value as problems become more complex. Consider (17) as addressed to the
helmsman. The obviously non-agentive
The helmsman is on deck by dawn (18)
cannot, by our thesis, represent its content; we need an agentive form, and
fortunately there is already something at hand: the stit sentence,
[The helmsman stit: the helmsman is on deck at dawn].
This, in turn, can be transformed back into the imperative
1-1 REMARK. (On notation) The standard deontic notation is Op, Fp, and
Pp. We depart from the standard for good reason: We shall in the course of this
book deal with many other modalities, several of which would have an equal
right to single letters such as 0, F, and P. We therefore use abbreviations for
deontic and other modalities that we hope are short enough for the eye to take
12 Introduction to stit
in at a glance and long enough for the mind to remember. (It helps that we do
not often use these longer expressions in the course of complicated calculations
in which extreme brevity is a decided advantage.)
Among its many virtues, this approach allows such great latitude that it can
be connected to every possible action by any agent. On the other hand, agents
have in this grammar no distinguished place; if invoked at all, the agent's name
is only an accidental feature of the declarative complement. It is an easy mistake
to think that Oblg:Q says more than it actually does say. It does not say, for
example, who, if anyone, is obliged to see to it that Q. Consider
in either case the sentence form is not fit to tell us who is to see to it that
no cooks (or dogs) are on the bridge. While a deontic language of declarative-
sentence complements may be satisfactory for impersonal oughts, many within
the tradition have seen that agents need to be treated with more care. Per-
haps foremost among those who have argued that these standard alternatives
are inadequate is Castaneda, for example, in Castaneda 1974 (see additional
discussion in 3D).
Some deontic logicians have suggested the step of changing the grammar
so that Oblg:, Perm:, and Frbn: are taken to be adjectives modifying action-
nominals. The fundamental forms are then taken as
Oblg:a, i.e., a is obligatory,
Frbn:a, i.e., a is forbidden, or
Perm:a, i.e., a is permitted,
where a stands in place of a term designating an action. So Oblg:a might be
instanced by Sailing is obligatory, or Frbn:a by Bringing the boat into the wind
is forbidden. Since actions are always the actions of agents, this step is in the
right direction, but it still fails appropriately to recognize the agent. Further-
more, there is a considerable loss in expressive power, since clearly understood
declarative sentences, Q, are so much easier to come by than are clearly un-
derstood action-nominals, a. In order to regain the headway lost in the move
from sentences to action-nominals, some have tried adding negative doings such
as not-sailings, or disjunctive doings such as luffing-or-flensing; but even with
these additions, there remain two features missing from the overhauled model:
first, the flexibility that follows from permitting an arbitrary declarative within
the scope of a deontic statement, and second, the grammatical means to identify
and keep track of the agent.
We propose a combination having the strengths of both the declarative-
sentence complement plan, Oblg:Q, and the action-nominal complement plan,
1. Stit: A canonical form for agentives 13
Oblg:a. We propose to focus on deontic statements that have only agentives for
their complements and thus can always be paraphrased into one of the following
forms.
Oblg:[a stit: Q], i.e., a is obligated to see to it that Q
Frbn:[a stit: Q}, i.e., a is forbidden to see to it that Q
Perm:[a stit: Q}, i.e., a is permitted to see to it that Q
Though Perm:[a stit: Q], for example, is not agentive (in our technical sense),
we intend it as quasi-agentive in the loose sense that it involves both an agent
and an agentive, and in the stricter sense that like an agentive, it has the agent
itself as a recoverable part of its intension.
Our proposal calls, then, for a deontic language enriched by the restriction
that the complements of obligation, permission, and prohibition be limited to
stit sentences, a proposal that forms part of the "restricted complement thesis."
RESTRICTED COMPLEMENT THESIS. (Stit thesis. Reference: Thesis 5) A
variety of constructions concerned with agents and agencyincluding deontic
statements, imperatives, and statements of intention, among othersmust take
agentives as their complements.
Recall that (i) stit sentences always express an action, (ii) stit sentences never
lose or misplace the agent, and (TO) there are no grammatical or metaphysical
or semantic restrictions on the declaratives that may be put in place of Q. With
these features of stit in mind, applying Thesis 5 to deontic constructions re-
tains, we think, the most valuable features of both the declarative-sentence com-
plement account, Oblg:Q, and the action-nominal complement account, Oblg:a.
For example, in context the burden of the prohibition of cooks (or dogs) from
the bridge might be
Oblg:[the Third Mate stit: no cooks (or dogs) are on the bridge].
Or we might have
Frbn:[the cook stit: the boats are lowered].
Question: Does this imply
Oblg:[the cook stit: ~(the boats are lowered)]?
The application of stit theory to deontic concerns is a topic deserving treat-
ment at length. We return to it explicitly in 2B.9, chapter 11, chapter 12,
and chapter 14, as well as more indirectly in chapter 4 (via imperatives) and in
chapter 5 (via the idea of promising).
14 Introduction to stit
But more deeply, for the special case of negation, the result of embedding is
not always any kind of agentive; that is, by the paraphrase test for agentives
suggested by Thesis 3, to which we hope you have agreed, the declarative
it is false that a sees to it that Q
is not invariably paraphrasable (or indeed equivalent in truth value with)
a sees to it that it is false that a sees to it that Q.
Thus, by Thesis 3, (19) is not itself an agentive. In more colorful language that
speaks against taking a naive approach to inventing a "logic of imperatives,"
we may say that the negation of an imperative is not always an imperative. We
later discuss the interaction of negation with stit in a number of places; see for
example 2B.6 and 2B.8 on "refraining."
Finally, let us recall that when concerned with the modalities of agency, it is
similarly helpful to use stit sentences to keep track of the agentives. The English
is helpfully paraphrased as
while
18 Introduction to stit
In all these cases it is crucial that the content of the order and the obligation,
the ability and the action, be all the very same, all captured by a single stit
sentence.
theoretician nor private saint, the archbishop was deeply involved in controversy
with the tyrant William Rufus and later his brother Henry in regard to the
matters of lay investiture and clerical homage; he vigorously opposed the former.
These controversies were heavily freighted with the concepts of promising and
commitment and agentive powers. In order to make clear that his authority in
matters spiritual was not at the pleasure of the king, Anselm refused to accept
the papal pallium from the hands of William Rufus. Partly in consequence, the
archbishop was in effect exiled by the king. Anselm's brief notes on the modal
logic of agency were, we think, composed during this bitter exile.
In the document that Henry 1967 calls N, Anselm writes:
Quidquid autem 'facere' dicitur, aut facit ut sit aliquid, aut facit ut
non sit aliquid. Omne igitur 'facere' dici potest aut 'facere esse' aut
'facere non esse.' (p. 124; from N 29.8.10)
Paraphrase by Henry: For all x, if 'x does' is true, then x does so
that something either is so or is not so. Hence the analysis of 'doing'
will in fact be an analysis of x's doing so that p, and of x's doing
so that not-p [where 'p' is a clause describing a state of affairs, and
'not-p' is short for 'it is not the case that p']. (p. 124)
The next place we know a modal agentive construct to crop up, much more
explicitly but still embedded in the context of a normative expression, is in
Kanger 1957:
Ought(Y sees to it that F(X, Y)). (p. 42)
Although the locution "sees to it that" is displayed only in a normative context
and wholly without comment, it is clear from the general tenor of Kanger's
methodology that he intended to be isolating a norm-free concept of agency.
The explicit grammatical breakthrough for the logic of agentive modality
comes in Anderson 1962, who, reflecting on Hohfeld, introduces for the first
time a separate form of expression intended to disengage the concept of agency
from normative considerations.5 When on p. 40 Anderson takes
M(x, p, y)
to represent the case "when x executes what is regarded as an 'action' ... and
y is the recipient or patient of the action executed by x," he suddenly gives us
a clean target for some analytic questions that otherwise seem confusing.
Anderson sometimes reads M(x, p, y), with perhaps too little attention to
the connections between formal and English grammar, as "x does p to (for)
y." Evidently here agency is, for better or worse, not separated from patiency.
And certainly there is in Anderson no semantic theory of agency or patiency,
and only a trace of a deductive calculus (e.g., Anderson points out that the
implication between ~M(x, p, y) and M(x, ~p, y) goes only from right to left).
That is, Anderson pioneers in isolating agency and patiency, but he does so only
immediately to recombine them with deontic concepts.
In 1963 each of two logicians, Fitch and von Wright, advanced modal theories
of agency, each of them stressing syntactic developments. Fitch 1963 defines
"does A" in terms of two other modalities, "striving for" and "causes," and
offers a deductive calculus. The work has not been taken up by later logicians
and is seldom cited in the published literature. Indeed, although NB was Fitch's
admiring and fond student and colleague, he regrets to say that he had to be
reminded of this essay by Segerberg 1989, which contains a maximally useful
account.
Von Wright, beginning in 1963 and continuing at least through 1981, was, we
think, the other logician to be a first to treat agency (or action) as a specific
modal or quasi-modal topic, always with that specially honest von Wrightian
insistence on the lack of finality of the formulation in question, including at-
tending to nonmodal formulations in which complements are taken as terms
signifying specific or generic actions, rather than sentences. As in other cases,
the work keeps a close eye on deontic logic, to which he contributed so much.
We think von Wright did not succeed in disentangling agency from change,
5
Very likely the breakthrough for Anderson came about after correspondence or conver-
sations with his friend Kanger. Somewhat later Anderson visited Manchester, where Henry
was. Henry remarked in personal correspondence that during this year of 1965 there was
a colloquium involving a number of persons interested in agency, including, e.g., Hare and
Kenny.
1. Stit: A canonical form for agentives 21
d(p/p),
which is to be read as expressing some such idea as "the agent preserves the
state described by p" (pp. 43, 57). Like Anderson, von Wright tends to leave
to the reader the task of putting bits of logical grammar together with bits of
English grammar. In contrast with Anderson, however, agency here has been
separated out from patiency. (We further discuss von Wright in 3A.)
Kanger and Kanger 1966 introduce as a separate locution
X causes F,
where F is supposed to be a sentence, but in a fashion like Anderson's, they
logicize about it only by setting down that F may be replaced by its logical
equivalents, and that the proposition that X causes F implies that F. Three
influential lines of research began about the same time as that of the Kangers,
each of which highlighted the separate existence of agentive modalities, namely,
those initiated by Castaneda, by Kenny, and by Chisholm.6
Castaneda, whose views concerning deontic logic have informed both philoso-
phers and logicians for many years (since at least Castaneda 1954), has much to
say that is relevant to agency as a modality. Though his philosophical concerns
led him to pursue goals other than the formulation of a modal logic of agency,
he repeatedly urged the fundamental importance of the grammatical and log-
ical distinction between "propositions" and "practitions" (a distinction put as
clearly as anywhere in Castaneda 1981); but because there is no possibility of
constructing a Castaneda "practition" from an arbitrary sentence, in the way
for instance that Anderson's M(x, p, y] or von Wright's d(p/p) each permits
an arbitrary sentence in place of p, Castaneda practitions cannot themselves
serve as the foundation for such a modal logic of agency. (3D expands our
consideration of Castaneda.)
Kenny 1963, in the course of initiating a rich literature on the verbal struc-
ture of our causal and agentive discourse, says that any "performance" in his
technical sense is describable in the form
bringing it about that p.
And Chisholm 1964b takes the following as a basic locution on which to found
an extensive series of definitions and explanations in the vicinity of agency:
There is a state of affairs A and a state of affairs B, such that he makes
B happen with an end to making A happen,
6
Of course other work on the theory of action has also influenced the modal logic of agency,
but that literature is unsurveyably vast. We note as a passing example that there is hardly a
one of our past or present departmental colleagues who has not contributed.
22 Introduction to stit
where the letters stand in for "propositional clauses," and where the subject of
"makes happen" can be either a person or a state of affairs. The discussions of
Kenny and Chisholm, though relevant to logical questions, are themselves not
directed toward the formulation of either proof-theoretical or semantic prin-
ciples governing their respective basic locutions. They are sufficiently closely
connected to our project, however, that we return to Kenny and Chisholm re-
spectively in 3C and 3B.
This is as accurate a record as we can manage of the early history of the
modal logic of agency. If this story is right, then the following gives its gist.
Castaneda 1954ff,
Kenny 1963, relevant discussions
Chisholm 1964ff
As we note in several places previously, in chapter 3 we extend our discussion
of certain among these figures by more closely relating their thoughts to stit
theory: von Wright, Chisholm, Kenny, Castaneda, and Davidson. Also in 5B
there is a little more consideration of some Hohfeldian themes. But for now we
leave this early part of the history.
The first modal logic of agency with an explicit semantics is, we think, that
of Chellas 1969. The primitive locution is
to be read as "T sees to it that O," where T is an agent and O takes the place
of a sentence (pp. 62-63). Chellas only deploys this locution in one context,
namely, as the argument of an imperative operator.7
7
But Chellas does not restrict the complement of an imperative operator to sentences
having the form A-TO as is required by our Thesis 4.
1. Stit: A canonical form for agentives 23
The language that Chellas uses in this pioneering explanation, like the "relative
possibility" language of Kripke a few years earlier, is neither familiar in itself nor
further clarified by Chellas. Perhaps this is the reason that, like his predecessors,
Chellas in practice confines his agentive locution to the imperative context from
which his need for it sprang, and does not pause to investigate its separate
properties. 8
After Chellas there is a substantial group of logicians all of whom have de-
ployed a binary relation or a pair of binary relations in an effort to generate
a semantic understanding of an agentive modality that might be used as the
complement of an imperative or of a deontic operator; we know of Porn 1970,
1971, 1974, 1977; Needham 1971; Aqvist 1972; Kanger 1972; Hilpinen 1973;
Humberstone 1977; Lindahl 1977; and Talja 1980. For a critique of the line of
research being described, with special reference to Porn 1970, see Walton 1975;
also of note are Walton 1976b, 1976a, 1980, which develop some insights in an
independent and more nonsemantic fashion.
The earlier Porn articles and that of Aqvist use only a single binary relation;
the idea of using two binary relations seems to be independently due to Needham
1971, Kanger 1972, and Hilpinen 1973. (Unless we have overlooked it, there is no
cross-mention; we have not seen Needham's M.A. thesis, but make the inference
from Porn 1977.) The reason for the second binary relation is given as this:
Agency has not only a sufficient condition aspect but a necessary condition
aspect (Kanger 1972, p. 109; Hilpinen 1973, p. 119), and one needs a separate
relation for each. The later workers in this mini-tradition play variations on this
theme. In our judgment this line of investigation, although initially promising,
and although producing some useful insights, has not been much followed up
for the following reason: It has remained obscure what one is to make of the
binary relations that serve as the founding elements of the entire enterprise.
Kanger 1972 says, for example, that one of the relations holds between a person
and a couple of worlds or indices when everything the person does in the second
world is the case in the first; and the other relation holds when the opposite of
everything the person does in the second is the case in the first (p. 109). That
is far from clear, and no one in the tradition is, in our judgment, any clearer
than that. For a final example, we describe and quote at length from Porn 1977,
which among those we mention is the most developed grammatical and semantic
8
We discuss Chellas a bit more in 4A, and we rely on his agentive operator, ATO, in several
placeswriting it, however, as [a cstit: A] in order to conform to our standard symbolism for
the stit sentence.
24 Introduction to stit
treatment of agentive modality. (All page references are to Porn 1977 and all
words not inside quotation marks are ours.)
Dap is read "it is necessary for something which a does that p" (p.
4). It is said (pp. 7-8) that an equivalent concept is found as the
definition of "a sees to it that p" in Chellas 1969, chapter 3, section
4, and in Porn 1971.
"Consider all those hypothetical situations u' in which the agent
does at least as much as he does in u. If v is such a situation, it may
be said to be possible relative to what the agent does in u. ... if p
is necessary for something that a does in u, then there cannot be a
situation which is possible relative to what a does in u and which
lacks the state of affairs that p. ... A natural minimal assumption is
that the relation [of relative possibility] is reflexive and transitive."
(PP. 4-5)
D'ap is read "but for a's action it would not be the case that p"
(p. 5), and also "p is dependent on a's action." (p. 7)
"For the articulation of the truth of D' a p at u we require all
hypothetical situations u' such that the opposite of everything that
a does in u is the case in u' ... [the relation must be] irreflexive and
serial." (pp. 5-6)
Further, to connect the two modalities D and D', a condition is
imposed that "requires that worlds which are alternatives to a given
world under the relation [for Da] be treated as equals in contexts of
counteraction conditionality." (p. 6)
C' ap is read "p is not independent of a's action." (p. 7)
Eap is defined as the conjunction of Dap and C'ap, and read "a
brings it about (causes it to be the case that, effects that) p" (p.
7). Porn says (p. 8) that an equivalent concept is found in Needham
1971, p. 154, an essentially equivalent concept in Hilpinen 1973,
section 6, and explicitly in Porn 1974, p. 96.
Porn finds unacceptable (p. 7) an alternative E* a p, defined as
the conjunction of Dap and D'ap, which Porn says is equivalent to
a definition of Kanger 1972.
There are three points to be made about this extract. The first is that given the
available apparatus, Porn 1977 seems to us to offer the best explanations of and
the most detailed working out of the modal logic of agency as based on abstract
binary relational semantics. Second, even these best-possible explanations seem
difficult. The conclusion one might draw is that one should doubt the likelihood
that the abstract relational-semantic point of view itself can continue to serve in
the way that was hoped. But third, however, and counting against this conclu-
sion, is that Porn 1977 is evidently formulating, in the context of the relational
semantics, the very combination of "negative" and "positive" conditions that
much later were built into stit theory (see 2A.2 and 2A.3).
Aqvist 1974, 1978 provide a much less abstract and more intuitive semantic
setting; these articles are the first of which we know that make the fundamental
1. Stit: A canonical form for agentives 25
ways to fill out the sentential complement of "sees to it that": An agent can see
to it that the starter engages and passes through various stages, or that a certain
recursive program runs, or ... . But it seems best to make explicit our failure to
more than barely mention such a large literature just because so many persons
think that although it may be arguable, it certainly isn't plausible that it has
no special relevance to agency. On the other hand, NB once asked a well-known
computer scientist/mathematician after a lecture on parallel processing if he
had meant his use of "actor" and "agent" to be anything but an idle metaphor;
he was aghast that one should need to inquire.
The articles Brown 1988, 1990, and 1992 represent a sustained and important
investigation of the modal approach to ability and its connection with action.
Brown initially proposed a modal operator that has something of the force of
"can do." Horty 2001, which uses [a bstit: Q] for this Brown connective, explains
Brown's ideas and relates them to stit theory.
Penultimately there is von Kutschera 1986, which articulates in one form
or another nearly all of the essential underlying ideas concerning agency on
which we base the semantics offered in subseqent chapters. We can describe the
extent of von Kutschera's priority only by using some phrases not defined until
later. At the very least, one must credit von Kutschera 1986 with the no choice
between undivided histories condition, with generalization beyond the discrete,
with generalization to multiple independent agents (including the independence
of agents condition), with attention to strategies, and with semantics for the
"deliberative stit" that we study in Definition 2-5 and 8G.l. It also needs to
be remarked that von Kutschera 1986 cites the earlier von Kutschera 1980.
Finally there is Hamblin 1987, which in the context of a study of imperatives
provides a rich source of formal, informal, and semi-formal ideas on the topic at
hand, many of which have influenced the present work; in particular, collegial
reflection on Hamblin's "action-state semantics" was the immediate context of
the beginning of the research reported in this book. Our own recommendation
is that no one ought to try to move deeply into any part of the theory of agency
without reading Hamblin.
structure, promotes greater clarity in the way we talk and think about the
phenomena of our world, and thus justifies its added complexity.
In succeeding chapters we strive to deepen our understanding of agency by
providing stit sentences with careful and well-motivated semantic analyses, and
we apply and generalize on it in a variety of ways.
2
In chapter 1 we followed and extended the idea, going back at least to Anselm,
of treating agency as a modalitya modality that represents through an inten-
sional operator the agency, or action, of some individual in bringing about a
particular state of affairs.* We proposed that using the stit construction as a
normal form, when we are confused, is a happy way to clarify some aspects of
action and agency. In this chapter we sharpen our understanding of stitand
thereby, if we are right, of agency. The central idea is that the concept of action
must be understood in relation to an open future, and we formulate a rigorous
theory that tries to understand how action is compatible with and indeed re-
quires irideterminism. In this way we essay a contribution to what Kane 1998
calls ''the intelligibility question" (p. 105).
We often label the chosen approach stit theory, because it concentrates on the
linguistic form "a (an agent) sees to it that A," which we abbreviate simply as
[a stit: A]. Part of the theory, however, has nothing to do with language. This
nonlinguistic part instead purports to articulate in a general way how agency
fits into the overall causal structure of our world. For reasons that will emerge,
we often ponderously refer to this theory as "the theory of agents and choices
in branching time,'' or, more briefly but less memorably, as BT + AC theory
or, when endowed with instants or times, as BT + I + AC theory. Stit theory,
including its nonlinguistic part, provides a precise and intuitively compelling
semantic account of the stit operator within an overall logical framework of
indeterminism; the account is then used as a springboard for investigating a
number of topics from the general logic of agency, such as the proper treatment
of certain concepts naturally thought of as involving iterations of the agency
"This chapter draws on several sources. We thank John Horty for co-authoring 2A, which,
with the permission of Kluwer Academic Publishers, is drawn from Horty and Belnap 1995.
2B, except for 2B.10, finds its source in a portion of Belnap 1991, for the use of which we
thank the International Phenomenological Society, while 2B.10 is based on part of Belnap
1996a, with the permission of Kluwer Academic Publishers.
28
2. Stit: Introductory theory, semantics, and applications 29
tor Was: that represents the simple past tense: The definitions from standard
(linear) tense logic suffice. 3
Since these structures allow alternative possible futures, however, it is not
so easy to understand the operator Will:, representing future tense. Returning
again to Figure 2.1, suppose that, as depicted, the sentence A is true at m3 and
at m4, but nowhere else. In that case, what truth value should be assigned to
Will:A at the moment m1?
On the approach developed by Prior and Thomason, there is just no way to
answer this question. Evidently, Will:A is true at m1A really does lie in the
futureif one of the histories h2, h4, or h5 is realized; but it is false on the
histories h1 and h3. And since, at m1, each of these histories is still open as a
possibility, that is simply all we can say about the situation. In general, in the
context of branching time, a moment alone does not seem to provide enough
information for evaluating a statement about the future; and what Prior and
Thomason suggest instead is that a future-tensed statement must be evaluated
with respect to a more complicated point of evaluation consisting of a moment
together with a history through that moment. We let m/h represent such a
point: a pair consisting of a moment m and a history h from H( m ) .
Since future-tensed statements are to be evaluated at moments and histories
together, semantic uniformity suggests that other sentences must be evaluated
at these more complicated indices as well. We therefore define a BT model
(branching time model) as a pair m = <G, J>, in which G is a BT struc-
ture <Tree, <>, and 3 is an interpretation function mapping each propositional
constant from the background language into the set of m/h pairs at which,
intuitively, it is thought of as true. Where = represents, as usual, the relation
between a point of evaluation belonging to some model and the sentences true at
that point, the base case of the truth definition for branching temporal models
tells us simply that atomic sentences are true where 3 says they are (see 8F.l).
2-1 DEFINITION. (Truth for atomic sentences) m, m/h = A iff m/h E J(A)
for A an atomic sentence.4
And the definition extends to truth functions, past, and future as follows (see
8F.2 and 8F.5).
2-2 DEFINITION. (Truth for truth functions and tenses)
and
3
We remind the reader that, as explained in Remark 1-1, we use nonstandard notation for
various connectives. Taking into account the large number of connectives used in this book,
our purpose is to be as easy as possible on both the eye and the memory.
4
It is not usual for languages of this kind to admit the possibility that even atomic sentences
might be true at one point of evaluation m/h but false at another point m/h', for different
histories h and h' belonging to H(m). What we have in mind are situations such as the
following. If, in a restaurant, Karl is offered cake or pie for dessert, it seems that "Karl
chooses pie," which is at least not obviously non-atomic, might be true relative to one history
through m, but false relative to another. In any case, whether or not relativizing truth for
atomic sentences to both moments and histories is actually necessary for evaluating statements
of this kind, allowing for the possibility at least does no harm.
32 Introduction to stit
and a2 are simultaneous, however, one must impose the additional constraint
that Choicea1mand Choicea2mare "independent": Every combination of a choice
by a1 at m and a simultaneous choice by a2 is an open possibility. In other
words, no choice by a1 can make it impossible for a2 to make a simultaneous
choice. This is illustrated in Figure 2.3 by the fact that from each (inner) box
there proceeds at least one history. The independence-of-choices constraint is,
of course, to be taken to apply to the simultaneous choices of any number of
agents. Gathering the individual primitives into a tuple gives us {Tree, <, Agent,
Choice>, which we call an "agents and choices in branching time structure," or,
more briefly, a BT + AC structure. (The exact definition is given in 2.)
The final primitive is a set Instant of instants partitioning the moments of
Tree horizontally into equivalence classes. Intuitively, an instant represents a
set of contemporaneous moments from each of the various histories, with the
different moments belonging to a single instant thought of as occurring at the
"same time" in the different histories. The instant containing the moment m is
represented as i(m). It is supposed that each instant meets (intersects with) each
history at exactly one moment, and that instants respect the temporal order of
histories in the following sense: If the moment at which an instant i1 meets a
history h is later than the moment at which i2 meets h, then the same relation
holds between the moments at which the instants i1 and i2 meet any other
histories. These suppositions about instants amount to strong restrictions on
the structure of Tree, satisfiable only if all histories share an isomorphic temporal
ordering, which is then inherited by the instants themselves. The point of the
restrictions, of course, is to allow for temporal comparisons between moments
from different histories.
When the basic framework of agents and choices in branching time is sup-
plemented with the Instant primitive, the result is an "agents and choices in
branching time with instants" structure, or BT +1 + AC structure, of the form
(Tree, ^, Instant, Agent, Choice>, with BT + AC as before; and we can de-
fine BT + I + AC models as structures of the form m = <G, J>, in which 6 is a
BT + 1 + AC structure and 3 an interpretation mapping each propositional con-
stant, as before, into a set of m/h pairs. Also, when we combine branching time
with instants, but without agents and choices, into a tuple (Tree, <, Instant),
36 Introduction to stit
2-5 DEFINITION. (Truth for the deliberative stit) m, m/h = [a dstit: A] iff (i)
971, m/h' = A for each h' E Choice am(h), and (ii) there is some h" E H(m) for
which m, m/h" = A.
Evidently, clauses (i) and (ii) here are analogous to the positive and negative
requirements from the achievement stit. In the present case, the positive re-
quirement is simply that a should act at m in such a way that the truth of A
is guaranteed; a should constrain the histories through TO to lie among those
on which A is true. The negative requirement, again, is that A should not be
settled true, so that a's actions can be seen as having some real effect.
In addition to the primary, one-moment/two-moment contrast between the
achievement and deliberative stits, there are two other differences that should
be mentioned at once.
38 Introduction to stit
The first concerns the role of histories. Although the indices at which an
achievement stit is evaluated contain both moments and histories, the histories
are nearly idle in the evaluation rule, needed only as objects of quantification
in deciding whether A is settled true or not settled true at some moment. An
achievement stit, if true at some moment-history pair, must be true at every
history through that moment; this fact is reflected in the validity of
which tells us that any true achievement stit is settled true. For similar reasons,
the theory yields also the validity of
the complement of any true achievement stit must itself be settled true.
By contrast, since deliberative stit statements are evaluated at the very mo-
ment of an agent's choice or action, histories must play a more central role in
their evaluation; they provide our only access to the outcome of the agent's
action. The theory therefore yields validities such as the following:
The first of these tells us that a deliberative stit is never settled true; the second
that it can be true only if its complement is contingent, again, by contraposition,
reflecting Aristotle's idea that there is no deliberating about matters settled by
necessity.
The final point of contrast between the achievement and deliberative stits
concerns the role of instants. These play an essential role in the semantics for
the achievement stit, but no role at all in the deliberative stit. Because of this,
models for evaluating deliberative stits alone can be simpler than the stit models
described earlier: They need not contain Instant as a primitive, and so do not
require us to assume a notion of "same time" across different histories in order
to make sense of agency. They can be based on BT + AC structures instead of
on BT + I+AC structures.
One way of understanding the semantic differences between the achievement
and deliberative stits that result from the reliance of the former on instants is
by considering the following two sentences:
should be true at m1/h' for each h' in Choiceam1 ( h 1 ) is satisfied, the negative
requirement, that there should be some h" in H( m1 ) such that Will: A fail at
m1/h", is not. It is easy to see also that [a dstit: Will: B] holds at mi/hi, but
that there is no point in the future of m1 along h1 at which [a astit: B] holds.
In spite of these contrasts, it is well to keep in mind that each of the achieve-
ment stit and the deliberative stit represents, in its own way, a transition from
indeterminacy to a determinate outcome due to the choice or action of some
agent. In the case of the achievement stit, the settled outcome is at some
temporal remove, whereas in the case of the deliberative stit, the outcome is
immediate. The sign of this for dstit is the validity of the following:
Even though neither [a dstit: A] nor Will:[a dstit: A] is ever settled true, Was:
dstit: A] becomes a determinate, permanently settled fact immediately after the
choice is made. (We use the idea of (*) later, on p. 186.)
Of course it's possible, as Figure 2.6 makes obvious. In this and later figures,
we make clear which values are given, and which are calculated. The point of
Figure 2.6 is that [a stit: A & B ] holds at the moment m1, but even though B
is a logical consequent of A&B, nevertheless [a stit: B] fails there. The crucial
fact is that the negative condition for [a stit: A&.B] is satisfied in virtue of the
counter, m2, since A&B fails there because of the falsity of A. In contrast,
at the only potential witness mo, B is already settled true where each history
through mo intersects the relevant instant, i( m1 ). The negative condition for [a
stit: B] is therefore violated. B is true, all right; it is just that B is not something
to which a has seen. So what a sees to is not closed under logical consequence,
and obviously so. There is not the slightest paradox in saying, there is neither
"funny logic" nor grammatical subtlety required in calculating, that from the
fact that you see to it that there is at least one injured man who is bandaged,
it does not follow that you see to it that there is at least one injured man, even
though that there is at least one injured man who is bandaged logically implies
that there is at least one injured man. To the contrary, it is deeply built into
the real-choice-based idea of agency that such cases should be typical.
2B.2 Refraining
2-7 QUESTION. (Refraining) Is refraining both doing and not doing (in the
same respect, at the same time, etc.)?
The trouble with refraining, as most people appreciate, is that it is often hard
to do. The trouble with refraining, as most philosophers appreciate, is that it
is hard to pin down because it is both acting and not acting; discussions of this
topic in the recent literature show how difficult it is to avoid being confused.
Consider the two imperatives:
2. Stit: Introductory theory, semantics, and applications 41
from
by beginning to make noises about external versus internal negation, even when
we know perfectly well that we do not know what we mean by this distinc-
tion; and why it is tempting to invent the-act-of-not-turning-thy-back-to-the-
compass, thus enriching our ontology, also without having any sense that we
know what we are talking about.
In the framework of stit, with its clear recognition that embedding is not
only possible but encouraged, there is a natural solution to this puzzle about
refraining. It involves a simple series of steps:
"a is not turning his back to the compass" is equivalent to "~(a is turning
s back to the compass)."
Conclude that "a is not turning his back to the compass" is equivalent to
"~[a stit: a is turning a's back to the compass]."
is turning a's back to the compass]" is not agentive for
that is (by a negative use of the stit paraphrase thesis, Thesis 3), it is
not in general equivalent to "[a stit: ~[a stit: a is turning a's back to
the compass]]."
Recall that refraining is agentive, so that the content of a refraining sen-
tence must, by Thesis 4, have a canonical form [a stit: Q], and equally
keep in mind that the negative imperative "Turn not thy back to the
compass" must also have an agentive canonical form.
The following are then seen to be equivalent:
and
But let us continue more generally. We know that any agentive proposition
can be expressed in the form [a stit: Q], where Q might or might not itself be
agentive. We now add that any refraining proposition can be expressed in the
form
whether Q is agentive or not. (English will always demand that the subject of
Q is a, but it does not care whether or not Q is agentive for a.) The "Refrain
from turning thy back to the compass" example chose Q as agentive, but there
is also "Refrain from being (or don't be) on deck at dawn", or "Don't get caught
by the cook." Because in these cases Q is not agentive, they have simpler, non-
nested analyses, just "[a stit: ~(a is on deck at dawn)]," or "[a stit: ~(a is
caught by the cook)]."
Without postulating new "negative acts" as ontological items or strange un-
defined internal versus external negations, all is clear: The agency of "a refrains
from seeing to it that Q" flows from the agency of [a stit: ~Q], and the nega-
tion comes from what it is that a sees to; namely, that not Q. When Q is itself
agentive, this can of course be further filled out, as in the "turning thy back to
the compass" example.
Now we see the drive for the double use of "a isn't turning a's back to the
compass". If taken as non-agentive, it is canonically just
is turning a's back to the compass];
but if it is to be taken in an agentive sense, then you will have to read it as:
stit: a is turning a's back to the compass]].
English does not have a short and precise way to make the distinction; but the
stit locution does the job.
Who would have thought that refraining from acting involved an embedding of
a non-acting within an acting, a non-agentive within an agentive? Only through
attending to the grammar of the canonical form [a stit: Q], which promotes such
embedding by its very design, does the truth of the matter become accessible.
The concept of refraining is confusing and difficult to think about without the
aid of theory and pictures. Figure 2.7 vividly illustrates the differences required,
and permits the essential calculations. (This and succeeding pictures represent
choices only for a, since only those happen to be relevant to the particular
points to be made here.)
i. Easiest to see is that [a stit: ~Q] holds at mo, with witness Wo and counter
at m1.
ii. It will also be useful to note of Figure 2.7 that [a stit: Q] holds at m1,
with witness w1 and counter at my,. (Pause to observe that Wo cannot
serve as witness for [a stit: Q] at m1. Reason: Q fails at m3, which is
choice equivalent for a to m1 at Wo, so that there is a violation of the
positive condition for Wo to witness [a stit: Q] at m1.)
44 Introduction to stit
Figure 2.7: [a stit: ~Q] versus ~[a stit: Q] versus [a stit: ~[a stit: Q]]
iii. It is clear that at m2, one cannot attribute a guarantee of the fact that
Q holds there to any prior choice of a, for that fact was up to nature.
The same is true of m3 and m4: At all of m2, m3, and m4,
holds, which is the mere absence of seeing to it that. It is worth noticing
that we can make this statement about m4 without even knowing whether
or not Q itself holds there; all we need to observe is the failure of Q at
moment m3, which is choice equivalent to m4 at Wo for a. It follows that
the positive condition for Wo to witness [a stit: Q] at m4 fails in virtue of
the failure of Q at m3, and since Wo is the only potential witness for [a
stit: Q] at m4, it must be that [a stit: Q] fails at m4, which is to say, it
must be that ~[a stit: Q] holds there.
iv. Moments m2 and m3 on the one hand, and m4 on the other, are quite
different with respect to refraining. Figure 2.7 shows that the moment w1
does stand witness to a's responsibility for his or her own inaction with
respect to Q at m2 or m3: Not only does the right-hand choice for a at
w1 guarantee that a does not see to it that Q, but the left-hand choice
from w1, at which a does see to it that Q, testifies that at w1 a had a real
choice concerning his or her seeing to it that Q. The moment m1 stands,
that is, as the "counter" required for the truth of the claim that at m2, or
m3, a saw to it that he or she did not see to it that Q.
v. In contrast to moments m2 and m3, you can tell that in fact at moment
m4, a did not actively refrain from not seeing to it that Q. The only
potential witness is Wo; but since a did see to it that Q at m1, and since
m1 is choice equivalent to m4 at Wo for a, the positive condition fails,
and thereby the claim to agency. At 7774 not only does a fail to see to it
that Q, but he or she also fails to see to it that he or she fails to see to
it that Q. At 7774 you can therefore observe the difference between mere
2. Stit: Introductory theory, semantics, and applications 45
not seeing to something on the one hand, and positively refraining on the
other, for, as we have calculated, at m4 there is "not seeing to it" without
"refraining."
[a could-have-stit: Q]
which makes "could have stit" sentences look like the quasi-agentives that they
are.
The point is that when approaching questions about "could have," first be-
come clear about the complement of "could have," because that is essential to
being clear about "could have" itself; further, to focus on the "could have" of
agency, paraphrase into canonical form, that is, into a stit sentence.
You may think that such canonical forming is too much trouble, and some-
times it is; if you are concerned, however, to give a general account of agentive
"could have," then the availability of a canonical form is exactly what is re-
quired: (i) Since every stit sentence is an agentive, you will not have explained
"could have" unless you explain "could have seen to it that Q" for arbitrary
Q. (ii) Since every agentive can be paraphrased as a stit sentence, if you do
explain "could have seen to it that Q" for arbitrary Q, you will have done the
whole job.
Next, refine the question of "could have done it" by letting the "done it" be
specialized to "done otherwise."
2-9 QUESTION. (Doing otherwise) How should we understand "could have done
otherwise" ?
Our chief thought, expanded in chapter 9, is that there are hidden complexities
in the use of the phrase "do otherwise." Figure 2.8 helps with the following
discussion. Evidently a stit: Q] holds at TOO, with Wo as witness and m3 as
counter. The rest of the picture is just like Figure 2.7.
Observe in the first place that the other choice available to a at Wo obviously
does not guarantee that ~Q, so on that ground alone it was impossible for a
2. Stit: Introductory theory, semantics, and applications 47
to see to it that ~Q. That other possible choice only risks ~Q, but does not
guarantee it. So in this diagram the agent is in no position to "do otherwise"
in the sense of seeing to ~Q.
In the second place, the other choice available to a at Wo does not even
guarantee that [a stit: ~[a stit: Q]], that is, that a refrains from seeing to it
that Q. It is possible, but it is not guaranteed, since nature can take us straight
to m4, where we calculated that a does not refrain. If this is what "could have
done otherwise" means, then ''could have done otherwise" is by no means a
consequence of taking the future as open.
One can also see, however, that at Wo there exists a strategy for a such that
if (i) a knows about that strategy, and if (ii) a wishes to follow it, and if (iii)
a does not run into problems of weakness of the will, then a is in a position at
Wo, in this somewhat Pickwickian or conditional sense of (or absence of sense
of) "guarantee," to guarantee that he or she does not see to it that Q, that is,
that ~[Q stit: Q]. The strategy is simply to make the right-hand choice at each
of Wo and w1: Then, no matter what nature has in store, the issue is bound
to be ~[a stit: Q]. The exact statement and proof of this principle, which is
a much more complicated matter than one might have expected, is the main
theorem of chapter 13. Consult also the applications of the idea of a strategy to
promising in chapter 5. But proof aside, one should allow that this weakened
"strategic" sense is a long way from what your average expert on free will might
have meant by "could have done otherwise," though perhaps it is what the most
subtle dialecticians of the topic were implying. In any event, the pictures seem
to make the discussion easier to follow.
No.8 Oddly enough, as Gupta pointed out to us, it depends on the relative
order of the witnesses provided for [a stit: P] and for [a stit: P c Q } . If the
witness for P is earlier (or the same as) the witness for P c Q , the "detachment"
is guaranteed. The detachment might, however, fail when the witness for Pc
Q is earlier. This is clear from Figure 2.9 and Figure 2.10: In Figure 2.9 the
witness w1 for the seeing to it that P at m0 is earlier than the witness Wo for the
seeing to it of the conditional atTOO;in that circumstance one is bound to have
[a stit: Q} at TOO witnessed by W0, and with the same counter at m1 serving for
both and
In Figure 2.10, however, the witness w0 for the seeing to it that P atTOOis
properly later than the witness w1 for the seeing to it of the conditional at m0-
In this case it can be atTOOthat one sees to it that P and sees to it that PcQ
without seeing to it that Q. In particular, w1 cannot witness [a stit: Q] at TOO
because of the failure of Q at m2, which is choice equivalent toTOOat w1 for a.
8
But see (12) in chapter 11 for the opposite result for dstit.
48 Introduction to stit
And W0 cannot witness [a stit: Q] at mo, because by then the fact that Q at
i(m 0 ) is settledthere is no "counter."
The answer is in Figure 2.11. Suppose that a has been steadily running at a
ten-minute pace, and at frequent moments (of which there is no lastthis is the
critical condition) a has the option to drop out of the run. Consider [a stit: Q}
at mo as "a sees to it that a finishes the mile in just ten minutes." Evidently
[a stit: Q] should be true at mo, but it is equally evident that no single prior
moment such as W0 is adequate as a witness. The reason that W0 cannot serve
as a witness is not just intuitive. As the picture shows, the positive condition is
violated, for Q fails at a moment that comes out of a right-hand side of a box
that is later than W0, and hence Q fails at a moment that is choice equivalent
to m0 at W0 for a.
We therefore need to complicate our semantics (the underlying extra-linguistic
theories remaining unchanged) by permitting chains as well as single moments
to count as witnesses. The details are a little delicate, but you can catch the
idea. It is the whole chain of choices coming right up to the finish line that
stands as witness to the truth at mo of "a sees to it that a finishes the mile
in just ten minutes." One has only to generalize the positive and negative
conditions appropriately; see the discussion of "witness by chains" in 8G.4.
Figure 2.11 shows, incidentally, that the successful ten-minute miler is, in a
sense just about to be defined, a "busy chooser"!
refraining from seeing to it that you juggle sixteen balls in the air. Is there a
way in which you can do that without seeing to it that you juggle sixteen balls
in the air?
(refref)
We discuss this striking equivalence a number of times in this book, with the
mathematical facts being elaborated in chapters 15 and 18.
Although a full proof of the refref equivalence is not appropriate to this sketch,
Figure 2.12, which serves as a kind of partial "semantic tableau," may help you
to appreciate the flavor of verifying the implication from left to right. Suppose
that [a stit: ~[a stit: ~[a stit: Q]]] holds at m0. It needs a witness w0 and
a counter [a stit: ~[a stit: Q]], which we write at m0'. This in turn needs a
witness W0' and a counter [a stit: Q], which we write at m0"; a definite argument
is needed, however, that W0' is correctly drawn as later than (or identical to,
a possibility expressed by the double lines) W0. A similar argument as to the
need for (and placement of) a witness and a counter justifies the remainder of
the right side of the diagram. Then a reductio argument permits us to argue
that [a stit: Q] can be "moved over" to a moment m1 on the left that is choice
equivalent toTOOat W0 for a. The left side of the diagram is part of a subsidiary
reductio: The picture as drawn places the witness w1 for [a stit: Q] at m1 as
properly above W0, which can be shown to be impossible, provided a is not a
2. Stit: Introductory theory, semantics, and applications 51
Figure 2.12: With no busy choosers, [a stit: ~[a sht: ~[a stit: Q]]] implies [a
sht: Q]
busy chooser. In fact, if there are no busy choosers, the witness for [a stit: Q]
at m1 must be W0 itself. This easily implies that [a stit: Q] must be true at
mo, as desired. (See 15C for a full proof.)
(i) implies (ii) implies (iii), (i) also implies ( v ) , and (iii) is also implied by (iv).
Since (with no busy choosers) refraining from refraining from doing collapses
52 Introduction to stit
to doing, there are no more positive modalities. (The deliberative stit, 8G.l,
has exactly the same structure; see the diagram in Horty 2001.) The other five
modalitiesthe "negative" modalitiesare just the negations of these.
One can use the idea of modalities to obtain some further insight into "why"
the refref equivalence holds. Let Q <-> The Pequod was scrubbed, and let a
= Queequeg. Now think of yourself as building a "semantic tableau." Start
at the bottom with Q, and consider what you can consistently add. First, did
Queequeg see to it that the Pequod was scrubbed, or not? Evidently either is
consistent. Let us conjoin to Q the statement that Queequeg was not responsible
for that fact,
Now consider whether Queequeg was causally responsible for the fact that he
did not see to it that the Pequod was scrubbed. It could go either way. It might
be that Queequeg actually and agentively refrained from seeing to the scrubbing:
[a stit: ~{a stit: Q]]. Let us, however, examine the other possibility. Perhaps
Ahab confined Queequeg so severely that he had no more causal responsibility
for not seeing to the scrubbing than a table has for not seeing to it that the
moon rises: ~[a stit: ~[a stit: Q]]. Let us add this conjunct, yielding, so far,
the consistent story,
We are now ready for the last step. Queequeg either saw to the last conjunct or
not. In our particular story he did not, since his confinement by Ahab stripped
him of all agency in the matter of the Pequod being scrubbed. The question,
however, is whether on any consistent story you can add that Queequeg saw to
the third conjunct. Can you consistently add [a stit: ~[a stit: ~[a stit: Q]]],
giving the total story as
The refref equivalence, Ax. Conc. 1, says "no," since it makes that last conjunct
equivalent to [a stit: Q], which openly contradicts the second conjunct, ~[a stit:
Q}. And contrariwise, if you cannot think of a consistent story for (1), then you
are marshaling evidence for yourself in favor of the plausibility of the refref
equivalence.9
Figure 2.13: With busy choosers, [a stit: ~ a stit: ~[a stit: Q]]] does not imply
[a stit: Q]
Figure 2.13 reveals the failure of the implication from [a stit: ~[a stit: ~[a stit:
Q]]] to [a sizt- Q]. There is not enough room to show all relevant assignments;
the idea is that aside from the top right-most moment, those jumping out of the
right-hand side of a box have Q, and those that are limit points of an infinite
chain of busy choosing have ~Q. You can tell that the example of Figure 2.13
deeply involves a busy chooser. We are not going to discuss this awesome garden
of forking paths other than to refer the reader to the mirror game of the Example
with busy choosers 9-1 on p. 268 and to the elaborations of chapter 18. Also
see Xu 1995a for a consideration of deeper levels of complexity of busy choice
sequences and how they relate to the modal structure of the logic of agency.
That (2) fails is illustrated by the following example: Though the cook is
forbidden to see to it that the boats are lowered, he is not thereby obligated
to see to it that they are not lowered. The point about (Hi) is that since to
be obligated is always to be obligated to do something, and since not doing
something is not doing something, by our restriction that the complements of
deontic modalities be limited to stit sentences, it follows that Oblg:~[a stit: Q]
makes no sense. But the solution is obvious: To be forbidden to see to it that
Q is to be obligated to refrain from seeing to it that Q, and we know what
refraining means. Accordingly, the proper equivalences, forced on us by the
restricted complement thesis, are just
STIT DEONTIC EQUIVALENCES. (Axiomatics concept. Reference: Ax. Conc. 2)
and
The first is illustrated as follows: If the cook is forbidden to see to it that the
Pequod is headed up into the wind, then the cook is obligated to see to it that
he does not see to it that the Pequod is headed up into the wind, and conversely.
Symmetry suggests as well that to be obligated to see to it that Q is to be
forbidden to refrain from seeing to it that Q, so that we add the following to the
stit deontic equivalences Ax. Conc. 2: Oblg:[a stit: Q] <-> Frbn:[a stit: ~[a
stit: Q]]. If the helmsman is obligated to see to it that the Pequod is headed
up into the wind, then the helmsman is forbidden to see to it that he does not
see it that the Pequod is headed up into the wind, and conversely. And now the
surprise: If both the Oblg: and the Frbn: equivalences hold, then by substitution
(of ~[a stit: Q] for Q) and transitivity, then
must also be added to the stit deontic equivalences Ax. Conc. 2. The equiva-
lence (2) says that the only way to be obligated to refrain from refraining from
luffing is to be obligated to luff. Right or wrong? It would seem that the only
way the equivalence could possibly hold or fail to hold were if as a matter of
(nonnorrmative) stit fact itself, the followingnamely, the refref equivalence,
Def. 14held or did not hold:
2. Stit: Introductory theory, semantics, and applications 55
(refref)
We are therefore led to consider yet once again whether we can think of a case
in which the right side holds without the left. Does refraining from refraining
imply doing? The mathematics for the achievement stit is disposed of in 15C
and chapter 18, and Horty 2001 shows that the refref equivalence holds for the
deliberative stit; but we take it that our general point of view commits us only
to trying to get clear on the matter, not to already having accomplished that
desirable end. In any event, given the refref equivalence, (2) already follows
from the "normality" of the obligation modality.
The first condition tells us that $ is history-independent (in the natural sense,
which we define explicitly in Definition 6-4), so that since truth of c (A) at m/h
implies settled truth of c (A] at m, it is enough to say of o (A) that it is settled
true (or false) at a moment m. Because of their history-independence, we do
not need to mention a history for these sentences. The second condition says
that o is a "success" locution. It follows easily from these two provisions that
o (A) cannot, that is, be true, hence settled true, without the complement, A,
being settled true.
We are now ready to turn to Prior's argument. Let us substitute for his
"finding out" the philosophically more familiar verb "to know," and let us fur-
thermore fix the knower as Autumn Jane. Having made these substitutions,
consider the claim that
56 Introduction to stit
At 3:00 Autumn Jane knows that at 5:00 her father will have seen to it
that her dress is clean.10
premisses and the conclusion, (i) Like many standard propositional attitudes,
knowledge is history-independent. (ii) Knowledge (in the sense that gives rise
to Prior's puzzle) implies truth. (iii) No stit is settled to be true at an instant
until after its witness. Therefore, (iv) it is impossible to have knowledge of a
stit until after its witness.
Of course the essential content is that given (iii), the "epistemic" propositions
(i), ( i i ) , and not-(iv) constitute an inconsistent triad. For certain other senses
or uses of "knowledge" it may be preferable to resolve the inconsistency by
keeping not-(iv) while denying (i) or ( i i ) . The important thing is to follow
Prior in knowing about the inconsistency before settling matters.
3
59
60 Introduction to stit
3A Von Wright
In 1B.2 we argued the inadequacy of the two standard accounts of deontic
statements. The one allows any declarative sentence, no matter how divorced
from agency, to occur as a complement in a deontic account. The other lets
the complements in deontic statements be action nominals. Take for example a
permission statement symbolized by Pa. On the sentence-complement account,
one would replace "a" by a sentence such as "Persia is conquered," so that Pa
could be instanced by "It is permitted that Persia is conquered." According
to the action-nominal account, however, permission would be treated as an
adjective that modifies action nominals. So with "a" replaced by the action-
nominal, "conquering Persia," Pa might have as an instance "Conquering Persia
is permitted." Von Wright's program, along with both Castaneda's practitions
and the stit sentence, has, among its goals, providing an acceptable alternative
to these two unsatisfactory approaches. In this section we explain just enough
of an early proposal of von Wright's to compare it to the stit sentence. Page
references are all to von Wright 1966.
Both von Wright's sentence form and the stit sentence restrict the comple-
ments of deontic statements to a well-defined class of agentives. Both reflect
the fact that time and change must be reflected in the structure of agentive sen-
tences. Von Wright, however, incorporates time and change into his language
by restricting the complements of agentive forms to T-sentences. As we shall
see, that decision, in contravention of the stit complement thesis, significantly
weakens his account, particularly in its failure to illuminate the connections
between acting and refraining.
The T-sentence, which stands at the center of von Wright's language of ac-
tions and agents, takes, as complements, two propositionseach of which de-
scribes a state of affairsand yields a sentence that describes an event. An
event sentence such as p Tq "describes the transformation of or transition from
3. Small yet important differences from earlier proposals 61
a p-world to a g-world" (pp. 25-26), where the transition is either the alter-
ation of a state into its negation, or the maintenance of a state just as it is.
Von Wright tells us that transitions are to be thought of as ordered pairs of
states of affairs, whose ordering "is a relation between two occasions which are
successive in time" (p. 27). Since between two successive moments a state of
affairs may either remain unchanged or be transformed into its negation, there
are four elementary state transformations:
PTP
PT~P
~PTP
~PT~P.
as von Wright's notation for sentences of the form the agent sees to it that a
world in which not-p obtains is transformed into a world in which p obtains,
while
f(~PTP)
is the official notation for sentences of the form the agent refrains from seeing
to it that a world in which not-p obtains is transformed into a world in which p
obtains. (We ignore truth-functionally compounded T-sentences; that develop-
ment makes no difference to our story. See Belnap 1999 for additional discussion
of T-sentences in connection with the theory of concrete transitions and with
special reference to branching space-time.)
In order to begin the comparison, let us recall that according to the stit
paraphrase thesis the English sentence
The sentence is settled true just in case there was a prior choice of the agent
Alexander that guaranteed the truth of "Bucephalus was tamed."
On von Wright's account that same sentence, represented by d(~pTp), is true
just in case (i) "... the state described by ~p prevails on the first of the two
successive occasions" and (ii) "... the change described by (~pTp) does not
happen, as we say, 'of itself, i.e., independently of the action of the agent" (p.
43). That is, on the first of two occasions "Bucephalus was not tamed" is true,
on the second occasion "Bucephalus was tamed" is true, and the change did not
happen independently of Alexander.
Von Wright's agentive sentence forms, with T-sentences as their complements,
are an important step forward in our understanding, insofar as they explicitly
introduce at the heart of the discussion of agency considerations of time and
change. Such considerations focus our attention on the fact that a language
adequate to the description of agents and their achievements must deal with
the past, present, andabove allthe future.
Let us notice that the agent, whose importance is acknowledged in the original
English sentenceas well as in von Wright's statement of the truth conditions
is lost in the official idiom; when the agentive sentence occurs as the complement
of another sentence the agent is not easily retrievable. The stit sentence, on the
other hand, is specific in locating the agent as someone who makes choices
among alternatives and by those choices sees to it that a proposition is settled
true. The agent is essential in the things which he brings about; if "Alexander
tamed Bucephalus" is settled true, it is so because Alexander made a choice, a
choice that guaranteed the truth of "Bucephalus is tamed." For these reasons,
the stit sentence gives a distinguished place to the agent, by insuring that the
agent term always remains easily and obviously recoverable.
Further, the agent is essential in the things that he or she forbears from bring-
ing about. Forbearing, or refraining, presents an interesting challenge because
forbearing is not doing. As von Wright puts it, "forbearing is not the same as
not-doing simpliciter" (p. 45). The problem then is to talk about something
which seems to be both a doing and a not-doing and also seems to be neither a
doing nor a not-doing. As we saw earlier, von Wright's device for introducing
agentive refrainings or forbearings into the language is structurally analogous to
his introduction of agentive accomplishments; he simply prefixes each of the four
elementary change statements with a refraining operator. Recall that f(p T~p)
is the official version of a refraining sentence where the agent refrained from
seeing to it that a world in which p obtains was transformed into a world in
which p didn't obtain, so when
which exactly accords with 2B.2. This picture of refraining, in which an agent
sees to it that he or she doesn't see to it that some proposition holds, shows
in its structure the connection between seeing to it that and refraining. It is
equally at home with more modern constructions such as "Don't smoke in bed"
and "Don't sit under the apple tree with anyone else but me."
When the declarative complement of the refraining statement is not itself
agentive, yet has an agent as its grammatical subject, a simpler account is avail-
able. For instance, the imperative "Refrain from getting captured by Alexan-
der," as addressed by Darius to his wife, has a simpler non-nested analysis:
in particular
[Darius's wife stit: ~Darius's wife gets captured by Alexander].
As it happened, she failed to refrain from getting captured, but Alexander didn't
fail to refrain from freeing her.
Within deontic contexts, the conceptual gain provided by the stit sentence
in accord with 1B.2 and 2B.9 is just as pronounced. In harmony with the
restricted complement thesis, Thesis 5, deontic statements must have one of
the following forms:
Oblg:[ stit: Q]: is obligated to see to it that Q
Frbn:[ stit: Q]: is forbidden to see to it that Q
Perm:[ stit: Q]: is permitted to see to it that Q.
The equivalences
Frbn:[ stit: Q] Perm:[ stit: Q and
Perm:[ stit: Q] Frbn:[ stit: Q]
continue to hold, but as indicated also in 2B.9, what to take as the forbid-
den/obliged equivalence is not so obvious.
Frbn:[ stit: Q] Oblg[ stit: ~Q]
can be false when ~Q is not agentive for a. For example, a common soldier
may be forbidden to see to it that the prisoners are fed without being obligated
to see to it that the prisoners fail to be fed. Furthermore,
Frbn:[ stit: Q] Oblg[ stit: Q]
is grammatically unacceptable because, as we just saw, ~[a stit: Q], though the
negation of an agentive, need not itself be agentive.
The following equivalence, as suggested in 2B.9, seems likely to be the ap-
propriate one:
Frbn:[ stit: Q] Oblg:[ stit: ~[ stit: Q]].
3. Small yet important differences from earlier proposals 65
"The guard is forbidden to see to it that the prisoners are set free" is true just
in case "The guard is obligated to see to it that he does not see to it that
the prisoners are set free" is true. That is, the guard is forbidden to free the
prisoners just in case he is obligated to refrain from freeing the prisoners.
Symmetry suggests that whenever the guard is obligated to set the prisoners
free, he is forbidden to see to it that he does not set them free. Thus the
equivalence
Oblg:[ stit: Q] Frbn:[ stit: ~[ stit: Q
also seems correct. (In 3C we say a bit more about refraining and negation.)
In summary, von Wright's form marks a step forward in our understanding as
it properly restricts the complements of deontic structures to agentive sentences.
But because his agentive forms are constructed by affixing the d operator or
the / operator to T-sentences, there is no possibility of the nesting of agentives,
something the stit sentence promotes. With the nesting of agentives, we gain
the ability to express the connections between such agentives as refraining and
doing. With the ability to express these connections, we can demonstrate the
value of restricting the complements of deontic constructions to agentives when
the complements of agentives are all declarative sentences.
3B Chisholm
Von Wright, as we have seen, constructs his language of agency and agentive
achievements upon a foundation of changes and transitions. One difficulty von
Wright faced was providing an adequate account of the connection between do-
ing and refraining. A natural response is to introduce agentive intentions into
the syntactic structure of the language. Chisholm, in a series of essays, has
elaborated upon just such a sentence form, one whose foundation includes the
explicit introduction of agentive endeavors. Such an undertaking, involving as it
does the introduction of intentions within the syntactic structure, invites com-
parison to the stit sentence because stit incorporates all mental considerations
into the prior choice of the agent. Chisholm's work on this matter, beginning
in 1964, covers more than a decade and is to be found in more than a dozen
different places. Though there are terminological changes and subtle conceptual
alterations, the basic ideas and their modes of presentation remain sufficiently
unchanged to be presented in a unified fashion. For simplicity of exposition we
follow most closely the account in Chisholm 1969.
Chisholm's enterprise is built on the sentence form
into
66 Introduction to stit
Joan of Arc made it happen that she joined the Dauphin's forces in the
endeavor to make it happen that the English are defeated.
(M _,...).
There are three open places in the undefined locution: The first is the subject
term of the sentence; the second, the complement of "makes happen," is non-
intentional; the third, the complement of "in the endeavor to," is explicitly
intentional and teleological. We will discuss the grammar and metaphysics of
the first open place, then go on to speak about the grammar and metaphysics of
the other two open places in the undefined locution. We conclude this section
with a brief discussion of when an agent "could have done otherwise."
Because the agent term in the undefined locution has no distinguished place
either within the structure of that form or in the official abbreviation, it may
not be obvious that there are three open places in the undefined locution. There
appear to be only two, one represented by the dots and another by the dashes.
Chisholm's consistent use of the word "he" as the subject of the undefined
locution reinforces this misperception, as does the fact that the agent term is
dropped altogether in the official abbreviation (M , ...). The stit sentence, in
contrast, insures by its structure that the agent term is always recoverable no
matter where it occurs.
Though it is not clearly marked, there is no doubt that there is a place in the
undefined locution for an individual term. While Chisholm doesn't tell us much
about it, he does say that "[t]he subject term of 'makes happen' may designate
either a state of affairs or a person" (Chisholm 1964b, p. 615). By allowing
terms that refer to states of affairs to appear in the subject place, Chisholm
is trying to exploit the similarity of structure between saying that one state of
affairs makes another happen and saying that an agent makes-it-happen or sees
to it that a certain outcome eventuates. He says in identifying the two that "...
any instance of our locution will refer to the agent as a cause, and it will imply
that he makes something happen" (Chisholm 1969, p. 206).
It may not be a good policy to allow states of affairs to appear in the first open
place. For, while it is plausible to say that one state of affairs makes another
happen, it will never be plausible to say that one state of affairs makes another
happen in the endeavor to make something happen. So though it may be true
to say that
The French attacking the English at Patay made it happen that the En-
glish were routed at Patay,
it will never be true to say that
The French attacking the English made it happen that the English were
routed at Patay in the endeavor to drive the English out of France,
3. Small yet important differences from earlier proposals 67
3C Kenny
Kenny presents his important and influential exploration of agents and their
achievements through his discussions of emotion and the will in Kenny 1963.
Unlike von Wright, whose language is founded upon a metaphysics of elemen-
tary state transformations, and unlike Chisholm, whose starting point involves
endeavors and intentions, Kenny's account of agents and actions begins with
3. Small yet important differences from earlier proposals 69
language itself. In this section we shall see how the stit sentence, which is very
like Kenny's canonical form, "bringing it about that P" can be used to clarify
some important problems in the vicinity of agency, problems that require the
full power of the stit sentence for their solution.
Kenny proceeds by "considering the special logical properties of the finite
verbs we employ to report actions" in the attempt "to isolate a simple and
fundamental pattern of description of human activity, which reports of emotional
states and reports of voluntary action alike exemplify" (Kenny 1963, p. 151).
The finite verbs used to report actionsverbs of actionare those verbs "which
may occur in the answer to a question of the form "What did A do?" (p.
154). Because the answers to that question may vary widely, Kenny focuses
more specifically on "transitive verbs of action, such as occur in the sentences
'Brutus killed Caesar', 'Wren built St Paul's', 'Mary roasted the beef, and
'Shaw admired Caesar'" (p. 154). After locating a class of sentences in which
each member has an agentive subject, a transitive verb of action as its main
verb, and an object of the verb, Kenny argues that such sentences, that is,
agentive sentences, ought not to be accounted simply as relations of the form
Rab. There are at least five important differences, Kenny argues, between action
sentences and ordinary two-place relations. We will consider Kenny's five ways
later in this section.
Having established, to his satisfaction, the distinction between action sen-
tences and other relations, Kenny proceeds to a discussion of the verbs that
occur in action sentences. Throughout the discussion Kenny remains constantly
aware of the grammatics of his project, that is, of the important connections
between the grammar of English sentences to be studied and the grammar of
the logic to be used to represent them.
Kenny locates three species of verb in the genus verb of action: static verb,
performance verb, and activity verb. Static verbs, Kenny tells us, such as "love"
"understand," and "fear," do not have continuous tenses, and "where '(o ' is
a static verb 'A has od' implies 'A os'" (p. 173). Performance verbs such as
"discover," "learn," and "find" do have continuous tenses, and "where 'o' is a
performance-verb, 'A has od' implies 'A is not oing'" (p. 173). Activity verbs
such as "listen to" and "keep a secret" do have a continuous tense, but " 'A is
oing' implies rather 'A has o d ' " (p. 172).
Kenny's canonical form, "bringing it about that p," represents any sentence
whose main verb is a performance verb. He reasons that "... the primary form
of description of a piece of voluntary behavior takes the form of a performance
verb ... [and] any performance verb can be replaced by an expression of the
form 'bringing it about that p.' where '(p)' describes the state of affairs which
is the result of the process" (p. 236).
We can see from the quotation that Kenny has already decided to restrict the
complement in "bringing it about that p" to sentences that describe states of
affairs, more particularly states of affairs that are the results of processes. Once
again, the stit complement thesis specifically rejects that restriction. Kenny's
own example,
70 Introduction to stit
... when I learn French, I bring it about that I know French, i.e.,
that I have the capacity to speak French (p. 183)
can be used to reinforce the importance of the thesis. All of the following stit
sentences are, in this context, equivalent:
[Kenny stit: he learns French]
[Kenny stit: he knows French]
[Kenny stit: he has the capacity to speak French],
though the complement of the first is agentive, the complement of the second
describes a state of affairs, and the complement of the third is explicitly dispo-
sitional.
By allowing any declarative sentence to appear as complement, the stit sen-
tence changes the focus from actions to agents. It is the agent, in this case
Kenny, who learns French, is responsible for paying the tuition, or is required
to take the examination. The stit sentence reflects this fact by insuring in its
semantics that the agent term remains recoverable no matter how deeply the
stit sentence is embedded within wider contexts.
Exploiting the stit complement thesis yet again we can disentangle Kenny's
claim that "'John is taller than James' cannot be rewritten in the form 'John
is bringing it about that ...'" (p. 182). Doubtless he means that "John is taller
than James" is not agentive for John, because it is not equivalent to "John sees
to it that John is taller than James." But he might be interpreted to mean that
the sentence "John is bringing it about that taller than James" is incoherent
because ungrammatical. Or he might be interpreted to mean that [John stit:
John is taller than James], though grammatical and meaningful, will usually be
false. "Usually" because if John has tied his legs to a rack and is stretching
himself to gain the needed height, the sentence is bizarre but true. The stit
sentence provides a way to distinguish the cases.
Exploration of the modalities of agency requires breaking away from a picture
of actions as events. Kenny, in his perceptive discussion, is considering just that
alternative. He tells us that there are
two ways of reporting the same event ("A od B" and "B was od by
A"), and it is indeed possible to regard the notion of an event as an
abstraction from these two forms of expression, designed to enable us
to consider an occurrence without commitment to a special interest
in either A or B.1 (p. 180)
Restating the point with stit, we may say that there are two ways of reporting
the same action, both of which can be paraphrased as [a stit: Q] :
a sees to it that Q
or
1
The symbols "A" and "B," as Kenny uses them, take individual terms for either the
subject or the object of a "transitive verb of action."
3. Small yet important differences from earlier proposals 71
is equivalent to
3. Small yet important differences from earlier proposals 73
[a stit: ~Q].
We prove, that is, that the absence of busy choosers implies the validity of
the "refref equivalence," Ax. Conc. 1, expressing the idea that to refrain from
refraining from seeing to it that ~Q is the same as seeing to it that ~Q. Fur-
thermore, chapter 18 proves the converse, that the presence of busy choosers
implies the invalidity of the refref equivalence.
Kenny's claim in the fourth way is that any term in a relation can stand
equally in any place of the relation, but that the same is not true of agentive or
action sentences. But there is no real difference, in this regard, between agen-
tives and other relations. In "The painter of Mona Lisa is of Italian nationality,"
or "The element hydrogen has an atomic weight of 1," either of which might be
symbolized as Rab, confusion may result when the terms are interchanged. The
same kind of confusion might result if someone were to say "That the Mona Lisa
is painted sees to it that Leonardo." Though inequality of terms is not distinc-
tive of agentive sentences, Kenny has perceptively highlighted the contrast in
such sentences between the agentive place and the sentential complement place.
With regard to the fifth way, concerning existence and aboutness, the point
seems to be that very often individuals are identified and described by reference
to an antecedent. Though the antecedent is often agentive, it need not be. From
the relational sentence "The annual flow of mud down the Nile doesn't deposit
mud in the Nile Delta," we may infer that one of the terms of the relation, the
Nile Delta, does not exist.
Kenny's canonical form "bringing it about that P" is put to work in his theory
of volition. The account has three parts: (i) the agent brings it about that P,
(ii) the agent wills that P, and (iii) it is in the agent's power not to bring it
about that P.
With respect to ( i i ) , stit theory avoids explicit discussion of volition or will.
Instead stit deals with an approximation of sees to it that by having [a stit: P]
be settled true just in case a choice of the agent, a, guarantees the truth of P.
Questions of volition or will are thereby postponed by incorporating all mental
considerations into the agent's prior choice.
With respect to ( i i i ) , if it means that it was possible at the time the choice
is made for P to be false, the claim is correct. For if [a stit: P] is settled true,
then it was possible for P to be false at an alternative moment. On the other
hand, if (iii) means that it must be in the agent's power to see to it that he or
she doesn't see to it that P, then (iii) is wrong. The considerations are similar
those at the end of the preceding section. When
Leonardo sees to it that the Battle of Anghiari is completed
is settled true, it is not a necessary condition that there was a prior choice point
when it was possible that
Leonardo sees to it that the Battle of Anghiari is not completed
is also true. It might have been that Leonardo's students were going to complete
the painting at any point at which Leonardo stopped working. Further, if the
74 Introduction to stit
force of (in) is that it must be in the agent's power to see to it that ~P, where
P is specifically agentive, it is also wrong. When
Leonardo sees to it that he paints the Mona, Lisa
is settled true, then there was a prior choice point at which it was possible for
Leonardo paints the Mona Lisa
to be false, but it is not a necessary condition that there was a prior choice point
when it was possible that
Leonardo sees to it that he doesn't paint the Mona Lisa
is true. Further discussion can be found in 2B.2, including a pictorial demon-
stration of these results in Figure 2.6.
The point is that when [a stit: P] is settled true, then at the prior choice
point it was obviously possible for P to be false, therefore possible that [a stit:
P] would turn out to be false, but neither of these involve a'S power. Therefore,
neither the possible falsity of P nor the possible falsity of [a stit: P] need be
due to the agency of a.
As we have seen, because the complement of the canonical form "bringing it
about that P" is restricted to sentences describing states of affairs, that form
cannot shed sufficient light on the differences between agentives and other rela-
tional sentence forms. Nor can it shed sufficient light on the various negations
of agentive sentences. Further, because it does not have a precise semantics,
"bringing it about that P" cannot make the fine distinctions between agentive
powers and temporal possibilities. The stit sentence, although it can help clarify
all those issues, remains only subtly different from Kenny's canonical form.
3D Castaneda
Von Wright begins his study with considerations of change and alteration,
Chisholm with intentions and endeavors, and Kenny with the language of agency
and action itself. Castaneda, to whom we now turn, sets out to clarify mat-
ters by considering practical reasoning. On Castaneda's account, the language
of agents and their achievements takes actions to be the finished products of
causal sequences that originate in practical reasoning. Practitions, the contents
of practical reasoning, are markedly similar to stit sentences. This similarity is
not an accident. It arises from the fact that both practitions and stit sentences
are intended, among other things, to function as the descriptive core of deon-
tic constructions. The schematic account that follows merely skims the surface
of Castaneda's extensive theories. It is meant to provide just enough back-
ground to allow us to make some comparisons with stit. For further discussion
of Castaneda's language of action see Castaneda 1974 and Castaneda 1975, as
well as the commentaries of Bratman 1983 and Tomberlin 1983a.
Castaneda separates practical thinking from theoretical thinking, giving each
a different characteristic activity and a different characteristic content. The
3. Small yet important differences from earlier proposals 75
some resulting stit sentences may be distinctly odd and some obviously false,
such cases turn out to be revealing. They further our understanding of agency,
of agentive sentences, and of the wider environments in which agentives are
found. We have already seen that when Q is a tautology, [a stit: Q] will never
be true. The negative condition will never be satisfied, since the tautology is
settled true on every history, so cannot be true due to the prior choice of a.
Similarly, when Q is a contradiction, [a stit: Q] will never be true. In this latter
case, the positive condition will always be violated since there is no history on
which a contradiction is ever settled true. We also noticed in previous sections
that having these forms available is a significant aid to the understanding of
inferences involving agentives.
Castaneda's example in his objection to the stit complement thesis does not
happen to concern complements that are tautologies or contradictions, but
rather a class of English sentences involving two agentive sentences with dif-
ferent agents as subjects. Castaneda's conclusion is that neither one should be
allowed to appear as the complement of the other. Again, because there is much
to be said about such cases that cannot be said if such sentences are deemed to
be ungrammatical, it is not good policy to exclude them.
Consider the two agents John and Marilyn and the sentence "The door is
open." Let us assume that the English sentence
When John and Marilyn act jointly, perhaps by jointly heaving a heavy door
that neither could open alone, it is settled true that
and
Marilyn opens the door
3E Davidson
Because Davidson 1966 has given rise to a substantial literature, our commen-
tary will not be extensive. Several remarks are in order concerning the difference
between the stit sentence and Davidson's approach. Though our goals are sim-
ilar, Davidson intends to clarify the inferential structures in which action sen-
tences appear by looking at the parts from which they are constructed. We have
in mind the intention made explicit in the initial description of Davidson 1966
of its own task:
I would like to give an account of the logical or grammatical role
of the parts or words of [simple sentences about actions] that is
consistent with the entailment relations between such sentences
From the modal point of view, it is striking that the aim set out in this passage
includes only half of what is needed for a compositional account of meaning.
Davidson 1966 sets out to show how "the meanings of action sentences depend on
their structure," but does not begin with the aim of showing how the meanings
of sentences that contain action sentences depend on their structure. The stated
aim does not include, for instance, telling how the meaning of "Jones refrained
from buttering the toast" or "Mary demanded that Jones butter the toast" or
"Jones, butter the toast!" or "How speedily did Jones butter the toast?" or
"Jones brought it about (or saw to it) that Jones buttered the toast" depends
on the meaning of "Jones buttered the toast," or perhaps telling how it doesn't
if it doesn't. Half the compositional problem has been left out of the initial
statement of purpose.
The modal logic of agency tends to strike the other way. The modal logic
of agency should be interested in larger contexts containing agentive sentences.
The fact is that with regard to embedding agentive sentences in larger con-
texts, it makes a difference that they are agentive, and it makes a difference
who the agent is. Embedding contexts care about these things. The reason
3. Small yet important differences from earlier proposals 79
neither case would we say that the doctor removed the patient's appendix" (p.
86). But as we noted in 1A, these are not reasons to reject the stit sentence.
Rather they are reasons to embrace it, for we want to distinguish between
agentive and non-agentive interpretations of the same English sentence. By
now it is clear that
The doctor removes the patient's appendix
is agentive for the doctor if, and only if, it is equivalent to
[The doctor stit: The doctor removes the patient's appendix].
Objecting to Chisholm's undefined locution, Davidson suggests that we
... take as our example "Jones batted an eyelash." In this case I
think nothing will do but "Jones made it happen that Jones bat-
ted an eyelash" (or some trivial variant), and this cannot be called
progress in uncovering the logical form of "Jones batted an eyelash."
(p. 87)
On the contrary, we contend that progress has already been made, and that the
stit sentence will likely continue the progress. In English when "Jones batted
an eyelash" is not agentive, the appropriate verb is "blink." When "Jones
batted an eyelash" is agentive the appropriate verb is "wink." It may be that
Davidson is right that "when we understand the verb we recognize whether or
not it includes the idea of an agent" (p. 94); but since natural languages are fluid
and changing, someday that nice distinction might be lost. Nice distinctions,
comfortable enough this year, may be covered by linguistic barnacles, gobbled
up by more aggressive phrases, or simply lost in a sea of idle chatter. (Consider
that in our time English is losing a word to describe young men of a particularly
lighthearted demeanor.) Idealized constructions, on the other hand, keep firm
control of their environment as well the individuals and structures that reside
there. This permits us to maintain the distinction between
[Jones stit: Jones batted an eyelash],
which is agentive, and
Jones batted an eyelash, but he didn't see to it that he did,
which isn't. Apart from philosophical clarification, the difference is important
when we are trying to decide whether to respond with a wink of our own or
with an eyedropper. Fourth and finally, with respect to Davidson's objection to
a "bringing it about that" proposal of Reichenbach 1947, and by extension to
any stit-like analysis: If
[The astronaut stit: The astronaut flew to the Morning Star]
is settled true, then assuming that the Morning Star is necessarily identical to
the Evening Star,
3. Small yet important differences from earlier proposals 81
3F Conclusion
None of the differences between stit and its predecessors is extensive. Rather
they are small and subtle. But, as Reichenbach reminds us, little deviations
often lead to deep insights. "It is," he says, "as if nature discloses its fundamen-
tal relationships in the minute errors of current theories" (Reichenbach 1980, p.
32).
Von Wright's sentence form, Chisholm's undefined locution, Kenny's canon-
ical form, Castaneda's practition, and the stit sentence all attempt to provide
the foundations for a modal language of action. The stit sentence advances that
project. The stit sentence has no restrictions on declarative sentences to oc-
cur as complements. The stit sentence is sufficiently rich to serve as a canonical
form for all agentive sentences. The stit sentence equips itself with the resources
to maintain focus on the agent. The stit sentence is rich enough to display the
value of restricting the complements of deontic statements to agentives. The stit
sentence has a semantics that pictures agents as making choices against a back-
ground of temporal logic and branching time. As we have repeatedly seen, these
minute differences, taken together, yield significant theoretical advantages.
4
Suppose that we humans were to be deprived of all forms of speech, except just
one that we might choose to keep.* Which would it be? On p. 10 of chapter 1
we cited and endorsed, as a much-neglected truth, Hamblin's confident assertion
that we should choose imperatives over all other forms. The central concern of
the following is to explore ways that stit can help in the understanding of this
important category. Among the factors underlying the relationship between stit
and imperatives is the fact that on the one side stit is a theory of agents seeing-
to-it-that, while on the other issuing and receiving imperatives are distinctive
forms of agentive behavior: Agents use imperatives to tell other agents what
to do, or what not to do, to advise them about how to act and what to wear,
to invite other agents to dinner and to the theater, to request favors and to
demand attention.
The concerns of this chapter are narrowly defined, focusing primarily on three
previous studies of imperatives: Hofstadter and McKinsey's theory of fiats,
Chellas's exploration of the logical form of imperatives, and Hamblin's in-depth
study of the logic and grammar of imperatives. Our aim is to show both how
stit has been informed by these works, and how stit, in turn, attempts to carry
forward the projects there begun.
82
4. Stit and the imperative 83
focus on the fiat, which makes "the satisfaction of an imperative ... analogous
to the truth of [an indicative] sentence" (p. 447), yields a simple and power-
ful theory of imperatives. The syntax is uncomplicated, using the exclamation
point as the only special symbol. The theory is expressively complete in the
sense that every indicative can be associated with a corresponding imperative
by having an exclamation point appended to it. The semantics incorporates an
already developed truth-tabular account developed for indicatives. Imperatives
are assimilated to indicatives via the notion of satisfaction. "We understand an
imperative to be satisfied if what is commanded is the case. Thus the fiat 'Let
the door be closed!' is satisfied if the door is closed" (p. 447). So, the ordinary
language imperative chosen by Hofstadter and McKinsey as their paradigm,
Ross comments that this result shows us that the theory of fiats "is surely not a
logic of such content which we have in mind in the case of practical inferences"
(p. 61).
The linguistic setting in question, the setting Ross describes as a case of "prac-
tical inference," is just the sort of setting where agency is essential. Since the stit
sentence is designed to provide a normal form for agency, let us reconsider Ross's
paradox with that machinery at our disposal. (For an alternative approach to
the constellation of issues surrounding imperatives, see Segerberg 1988b and
Segerberg 1990.) Our discussion will depend both on the stit theses listed in
1 and the semantics for stit offered for the deliberative stit, 8G.l, and for the
achievement stit, 8G.3 (here it will make little difference which stit is at issue).
is true, then
The letter is either burned or posted
is also true.
The imperative content thesis, Thesis 4, however, recommends that we treat
the content of any imperative sentence as agentive. (We return to the topic
of agentive constructions in 4D.) By the stit paraphrase thesis, Thesis 3, we
know that an agentive construction is always appropriately paraphrased as a stit
sentence. Further, since stit extends us the opportunity to clarify by normal
forming, we are in position to verify that
follows from
4. Stit and the imperative 85
4C Chellas's theory
The theory of fiats sets the study of imperatives on the right path and with a
good footing. Fiat theory shares a helpful point of view with stit, and also with
Chellas 1969, who puts the matter as follows:
... imperatives are a species of sentence the logical form of which can
be investigated by examining the logical properties of their counter-
parts in a suitably articulated, well-defined language (p. 4).
It is to Chellas's forceful and influential study that we now turn. (Horty 2001
provides additional discussion of Chellas's contributions.) We saw earlier that
while Hofstadter and McKinsey consider both fiats and directives, they provide
86 Introduction to stit
a formal account only for fiats, and we saw that their semantics is classically
based. Chellas supplies imperatives with a modal semantics and is emphatic
in the need for adding a temporal structure. The "recognition of temporal
elements," he says, "compels us to acknowledge the dependence of imperative
obligation upon time; an (imperative) obligation to the effect that such-and-
such be the case may hold at some times and not at others" (p. 77). In stit
theory, the temporal structure is portrayed by branching histories.
Chellas's logical form for the directive,
and by that utterance creates an obligation, in this case the obligation that Ben
clean his room, she has succeeded in ordering him to clean his room. When Ben
utters the same sentence, he does not succeed in ordering his mother to clean
her room. Even if he intends it as an order, he succeeds, at best, in offering
a piece of advice, or perhaps making a joke. Without the ability to impose
obligations of one sort or another, imperatives must fail to be orders. This
approach evidently differs essentially from that of Searle 1965 and Searle and
Vanderveken 1985 in its concern with what is accomplished with an utterance
rather than with what an utterance is intended to accomplish.
We learn some important lessons from Chellas: (i) the need for an agentive
form as the complement for imperatives; (ii) the need for a modal semantics
for imperatives; (iii) the need for temporal elements in the semantics and (iv)
the close relation between obligations and imperatives. All of these lessons are
taken account of in the syntax and semantics of stit. Still further progress,
however, is possible if we widen our understanding of imperatives. We proceed
toward that destination with Hamblin as our guide.
4D Agentive constructions
In the last several sections, the notions of an agentive and an agentive construc-
tion have been invoked repeatedly. The agentiveness of stit thesis guarantees
that a stit: Q] is always agentive for a. The stit paraphrase thesis tells us that
Q is agentive for a iff Q is appropriately paraphrased as [a stit: Q}. The re-
stricted complement thesis requires that imperative and deontic constructions,
among others, take agentives as their complements.
We now propose to show how the stit understanding of this important idea
continues along the trail blazed by Hamblin 1987. The most striking difference
between stit and Hamblin is that Hamblin's tests for agentives look to fea-
tures of ordinary usage and ordinary language, while stit utilizes the benefits of
its theoretical apparatus to separate agentive from non-agentive constructions.
(Hamblin credits the distinction between agentive and stative to Vendler 1957
and the application to imperatives to Kenny 1963.) That is, Hamblin is trying
to uncover the relevant differences between agentives and non-agentives in the
ways we ordinarily speak about such matters. Stit, on the other hand, specifies
the distinction with the aid of a theoretical structure. The point in both is the
same: to discriminate between those constructions that are, and those that are
not, agentive.
Let us first look at Hamblin's five tests (pp. 54-56) for distinguishing agentive
from stative predicates and then proceed to the stit versions.
i. "Agentive, but not stative, predicates can take a 'continuous' tense"; so
Ben is running
is acceptable, but not ordinarily
Ben is knowing the answer.
88 Introduction to stit
ii. "Agentive, but not stative, predicates fall under the genus do something."
In answer to a question about what Jessica did, you may answer
Jessica cleaned the house
but not
Jessica was a student.
Hi. "Agentive, but not stative, predicates can be augmented with the adverb
deliberately, or with adverbs of manner such as carefully, enthusiastically";
We may properly say
Mark worked deliberately,
but not
Mark rested deliberately,
or
Mark slept carefully.
iv. "Someone may be persuaded or reminded in respect of an agentive pred-
icate such as to go to bed, but not in respect of a stative one such as to
understand the instructions or to be asleep."
v. "The present tense of an agentive, but not stative, verb can be used with
future reference"; so we might have
Kathy leaves for school soon
but not
Kathy sleeps soon.
Because his approach relies on the vagaries of ordinary speech, Hamblin recog-
nizes that these tests are, at best, indications of the distinction between agentive
and non-agentive constructions rather than precise markers. Notice, for exam-
ple, that "Kathy sleeps soon" passes the first test, fails the third, fourth, and
fifth, and remains in doubt with respect to the second. It is perhaps inevitable
that attempts to find absolute distinctions in the ways we ordinarily speak are
doomed to imprecision. Stit, by contrast, specifies the difference between agen-
tive and non-agentive constructions in terms of the theory itself and validates
that account in application.
Consider the sentence
Jessica trips. (11)
Is it agentive or not? The stit view is that (11) is agentive whenever Jessica's
tripping is guaranteed true because of a prior choice of the agent, in this case
Jessica. If (11) is true because of Jessica's prior choice, then it will be appro-
priately paraphrasable as
4. Stit and the imperative 89
4E Negations of imperatives
Having in hand the basics of stit theory, and a fuller account of the concept of
an agentive, we are now prepared to continue our exploration of imperatives,
concerning ourselves with negation. Before presenting the stit versions, let us
survey Hamblin's classification of imperative negations (pp. 64-71).
Hamblin's type 1 negation is the first of five distinct kinds of negations for
imperatives. This negation, he tells us, applies to statives as opposed to agen-
tives. Type 1 negation is "the one closest in spirit to propositional negation."
In this sense the negation of "Let the door be closed" is "Let the door not be
closed."
Type 2 negation applies to agentives. Negation of this sort changes an order
"from an instruction to carry out that action to an instruction to refrain from
doing so." In English, one has the passage from
Close the door (13)
to
Don't close the door. (14)
For negations of type 3, Hamblin supposes that there "is a natural order of
events that will prevail provided nobody, or nobody relevant, interferes." This
sort of negation, Hamblin tells us, can be expressed with "the use of emphasizer
words" as in
Don't actually close the door.
90 Introduction to stit
The idea is that this delicate use of emphasis is supposed to convey an order to
let the event occur.
Negations of type 4 comprise those "locutions whose role it is to negate ...
pressures, enticements or encouragements to undertake a course of action."
Hamblin characterizes such negation as what some deontic logicians have called
'external' negation. Type 4 negations transform the order to close the door into
the permissive
You may refrain from closing the door.
is a negation of the fifth type and "different from explicitly permitting noncom-
pliance."
Type 2 negations present stit theory with a nice problem. In stit theory
there is only one primitive form of negation, sentential negation. It can be used
to negate any sentence, whether agentive or not, whether imperative or not.
Let us notice that because a stit sentence takes any arbitrary declarative as
complement, each stit sentence naturally admits a sentential negation in either
of two places. For example,
[ sitt: the door is closed]
can be negated either by
or
In (15) the scope of the negation is the declarative complement and the result-
ing construction remains agentive. That's good, but nevertheless (15) cannot
represent the content of the type 2 negation, (14), since (15) evidently answers
instead to "See to it that the door is not closed," or perhaps even, in context,
"Open the door."
In (16) the negation is external, but (16) is no longer agentive. Although (16)
expresses not doing, because it is no longer agentive, it cannot transform an
order "from an instruction to carry out that action to an instruction to refrain
from doing so." (16) cannot be the type 2 negation of (13). Something more
is needed, and it is a decided advantage of stit that it gives us that something
more. The stit sentence
4. Stit and the imperative 91
is both agentive and expresses not-doing. (See also the discussion of refraining
in 2B.) Now we can see why the appropriate normal form for (14), "Don't
close the door," is (17) rather than either (15) or (16).
Type 3 negations invoke a distinction between letting something happen and
making it happen. Negations of type 3 assume "a natural order of events."
Because in stit the future is open, replete with real choices and real possibilities,
it is a matter of policy and not oversight that this type of negation is not
accommodated in the foundations of stit. Note that it is not inconsistent with
stit theory to characterize some outcomes as "seen to" and others as "allowed
to happen" by adding additional concepts to the basic structure.
Type 4 negation, characterized by Hamblin as "external negation," trans-
forms an order into a permissive, or may transform an imperative such as (14)
into "You may close the door." But these reflect the standard deontic equiv-
alences, equivalences likely to be represented in stit theory, with the obvious
abbreviations, as
STIT DEONTIC EQUIVALENCES. (Axiomatics concept. Reference: Ax. Conc. 2)
that is, Kathy sees to it that she does not order Ben to close the door. In short,
Kathy refrains from ordering Ben to close the door.
In stit theory there is only one negation, sentential negation. As we have
just seen, however, that negation together with stit can help to represent all
the negative forms described by Hamblin that don't depend on a distinction
between making and letting.
We may go a little further if we permit "ordering" to be paraphrased, roughly,
as "seeing to it that the recipient is obligated," so that (20) comes to
92 Introduction to stit
This normal form of an order makes it evident that in each of the numbered
blanks there is room for a negation. A worthwhile stit exercise is to see how
instructive it is to fill these blanks with negations in various combinations, thus
pulling together into a single schema some otherwise confusing observations.
For example, filling 3 and 5 with negations describes the positive granting
of permission or authorization, while filling 2 and 5 with negations describes
the positive act of refraining from laying on an obligation, which is sometimes
thought of as a kind of permission, Finally, filling 1 and 5 with negations
describes the non-act of not laying on an obligation, which some also might
think of as falling within the precincts of permission. This suggestion is to
be compared with the more refined idea of Wansing 1998, according to which
agency is made intrinsic to the concept of obligation; see 11H.
[Kathy stit: Kathy advises that [Ben stit: Ben comes to her party]]
then as Hamblin tells us, they "are accountable" (p. 13). nonwillful imperatives,
like advice, are accountable as they involve the issuer and addressee in the asking
for and giving of reasons. Kathy may, upon giving such advice, be asked to show
how going to the party serves Ben's interests. Her advice may then be judged
as good or bad, appropriate or inappropriate, sound or unsound. Of course, she
might be asked to give reasons for an order, but then, as Hamblin points out,
because an order is a willful imperative, she may properly reply that she has
given an order or issued an invitation and that's all there is to it. (If Kathy's
speech act constitutes an invitation, we need a more complex analysis. We
consider that account at the end of 4G.)
Such analyses are not the only ones available using the machinery of the stit
sentence; they are only some examples of the possibilities the stit sentence opens
up to theorizing.
4G Embedding imperatives
We have seen that the stit sentence not only accommodates Ross's counterex-
ample, but illuminates the conceptual territory surrounding that inference. We
have just seen, in 4F, how the richness and diversity of imperatives is better
accommodated in stit theory than by a logical grammar of fiats. There is yet
another feature of stit that recommends it: Imperatives embed, and the stit
sentence not only allows embedding, it encourages it. 2
Consider the following from Kathy to Ben
I advise you to order Jessica to close the door. (27)
Though we can extract from (27) the fiats
Ben is advised to order Jessica to close the door!
Jessica is ordered to close the door!
The door is closed!
there is, in the structure of fiats, neither grammatical nor semantic clarification
of the relations among the embedded imperatives. Let us see how stit's positive
encouragement of embedding contributes to our understanding of (27). As we
already know, the stit sentence facilitates regimenting both the act of issuing
an imperative and the content of the imperative. So (27) might be represented
as
[Kathy stit: Kathy advises Ben to order Jessica to close the door]. (28)
Next, by the restricted complement thesis, the complement of Kathy's advice
in (28) must be agentive, represented as
2
See 1C for additional consideration of this fact. That section emphasizes that imperatives
embed in exactly the same sense as declaratives and interrogatives.
4. Stit and the imperative 95
4H Conclusion
In the preceding, we have carried forward the exploration into the nature and
structure of imperatives by looking at some of the ways in which stit, especially
via the stit theses listed in 1, incorporates the insights of previous philosophers
concerned with the same issues and builds upon them.
With the help of the imperative content thesis, we have seen that the stit
sentence is a notational variant of the directive, a construction that (by the
agentiveness of stit thesis and the stit paraphrase thesis) is suitable to represent
agentive constructions, a task the fiat is unsuited for.
Because stit provides a modal logic of agency, it is able to make headway that
was not possible so long as the study of imperatives remained more narrowly
confined. Imperatives are thus treated as a special case, albeit an important
special case, of the language and logic of agency.
We have tried to show the appropriateness of the stit normal form thesis,
Thesis 6, with respect to imperatives; that is, since imperatives are among the
constructions whose contents must be agentive, nothing but confusion is lost if
we take those contents to be expressed by all and only stit sentences.
5
97
98 Introduction to stit
and hence no act of smiling? And so on. Conclusion: It might not be a good
idea to assume that the only way to articulate the complexity of a promise is
in terms of structures appropriate to events, for example, part-whole or causal
relations. All that reification of acts as events may begin by sounding scientific
and hardheaded, but after one or two steps it may lead directly to perplex.
An alternative idea represents some aspects of the complexity of a promise
by structures provided by stit. This gives us a speech-independent approach to
the idea of promising that is even more hardheaded, but that does not have to
stop to worry about the perils of reification. Without the intent to minimize the
potential value of an "acts as events" analysis of promising, we urge that the
light of stit shine in corners notor not yetilluminated by reifying actions.
The purpose of this chapter is to explore that idea. We begin with a kind of
triple pincer attack on promising. In the first part of the chapter, 5A, we
use stit theory to suggest approaches to the structure of promising. In the
second part, 5B, we begin with Thomson's approach in The Realm of Rights,
asking how the modal logic of agency might contribute to what is there said
about promising, and, more critically, how the point of view of that book can
enrich a stit-based account of promising. Then finally, in 5C, we call on the
theory of strategies of chapter 13 to make good on what seem to be the common
deficiencies of both of the first two sections.
5-1 SLOGAN. (Feature the agent) Always give pride of place to agents.
5. Promising: Stits, claims, and strategies 99
If you did something, then you could have done otherwise. (3)
We discuss the complexities of (3) in chapter 9. Here let us consider (2). Stit
theory complains about the following ambiguity: Is the "something" in (2)
agentive or not? Why is this important? Well, it is nearly a logical truth that if
you are morally responsible for the fact that you saw to it that Q, then in fact
you saw to it that Q. There needs to be no discussion (or not much) about (2)
when "something" is taken agentively, as made explicit in the stit paraphrase.
It borders on incoherence to hold someone morally responsible for seeing to
something that he didn't see to. But consider the case when "something" in (2)
is not agentive at all, but a non-agentive state of affairs. The form is this: If
you are morally responsible for the fact that Q, then in fact you saw to it that
Q. This, as we also argue in chapter 9, is far from a nearly-logical truth. Our
use of italics should make the matter obvious.
Consider an example in the vicinity of promising. Your friend entrusted you
with some money. You chose to go to the race track. You are morally responsible
for the non-agentive fact that you were at the race track, and indeed you did,
in this story, see to it that you were at the race track. Further, you are morally
responsible for the non-agentive fact that you are in the way of temptation to
gamble with your friend's money, and in fact you did, in this story, see to it
that you were in the way of temptation to gamble with your friend's money.
But do not generalize from this example. Continue the story. You bet your
friend's money on a very fine horse with every chance to win, and at wonderful
odds. You made a good bet with your friend's money. However, against all odds,
the horse lost, and so was your friend's money. Here, because you deliberately
put your friend's money at risk, there is reason to say that you are morally
responsible for the fact that your friend's money is gone. But that remains true
even though you did not see to it that your friend's money is gone. You put the
money at risk, with full agency, which is a ground of your moral fault; but its
loss was due in part to chance, not entirely to your agency.
Some moralists prefer to confuse this issue. They prefer to say "Well, in a
way you did see to it that your friend's money is gone." They should resist this
temptation. The reason is this. To speak sloppily in this way is to interfere with
the enterprise of clearly explaining the connection between the facts that (i) your
friend entrusted you with some money, that (ii) the money is gone, that (in)
you were fully agentive in putting it at-risk-from-the-chances-of-nature while
not fully being the agent of its loss, and that (iv) nevertheless you are morally
responsible for the fact that the money is gone. Moralists should stay aware of
what an enormous difference there is between moral responsibility for what we
do and moral responsibility for complex non-agentive or partly agentive states
of affairs. The stit approach helps keep us honest.
5-3 SLOGAN. (Try stit) If a form of speech takes a complement that concerns
action, try thinking about its complement in terms of the stit paraphrase thesis.
That may help in understanding how the complement contributes to the larger
sentence in which it is embedded.
We take up promising as an extended illustration of Slogan 5-3. What gets
us started is that to promise is to do something; "a promised ..." is agentive.
Promising is a paradigm "speech act," therefore an act. That it is an act implies
that it should or might be enlightening to consider promising as a stit:
[a sitt:_]
What goes in the blanks? It seems always best to begin in the safest way, by
putting in the selfsame thing we started with:
[aa siit:a promised ...]
That may seem like "no advance," but in fact the stit normal form does en-
courage us to zero in on promising as an accomplishment. What is it that was
actually done when the promise was made? How did the world change due to
the agency of the promisor? It is no mean beginning to notice that promisings
occur at concrete moments with a "before" and an "after," and with alternative
possibilities for the future. As a stit, a promise involves a choice that rules
out alternative possibilities. These remarks give us a way of asking what oc-
curs when a promise is made: How is the world different, after an occasion on
which a promise is made, from what it would have been after some other choice
available on that occasion? We come back to this.
In shedding light on this question, stit can provide further assistance. When
one promises, as we earlier recorded, one does something. In addition, what one
promises itself pertains to agency. A helpful but too-strong statement of this is
that promising is promising to do, which is to say that the complement of the
promising is itself an agentive. This is the essence of the restricted complement
thesis, Thesis 5, as it applies to promising. Here is an easy example:
Paul promised to pick up Marie at the station at 9:00. (5)
To see how stit helps (if it does) even in this easy case, let us repeat the ap-
plicable version of the restricted complement thesis, which we stated earlier in
ordinary philosophical prose:
Now let us refine that version by applying the stit paraphrase thesis, Thesis 3,
which suggests that we word the "promising to do" thesis (6) as follows:
Our verdict on this thesis is that it is helpful for the present, but falls short in
the end. Here we are exploiting the extent to which it can help.
In (5), the paraphrase suggested by Thesis 5-4 is so easy as to not yield much
information:
Paul promised that [Paul stit: Paul picks up Marie at 9:00]. (7)
The rephrasing forces on us that we need Paul as a double agent, once of the
promising and once of the up-picking. With stit he never gets lost in some
philosopher's impersonal reductionist neverland.
Here is an example where stit helps a little more. Let us start with
Paul promised to be at the station at 9:00. (8)
describes an agentless "state of affairs." That makes it look as if one can promise
such a state of affairs in an agent-free sense, and perhaps so. Thesis 5-4, however,
bids that we not start at that place. Until we are forced by the data, as we will
be, we refrain from wondering what it is to promise a state of affairs. It is
more natural, for this example (we'll later come to more difficult examples), to
begin with the thought that to promise to be at the station at 9:00 is as much
a promise to do something as is the more agentive-sounding promise (5) to pick
up Marie at 9:00. Stit makes it clear:
Paul promised to see to it that he would be at the station at 9:00, (9)
or, with the stit made even more explicit,
Paul promised [Paul stit: Paul is at the station at 9:00]. (10)
That's clearly a useful paraphrase of (8), is it not? Even though being at the
station at 9:00 is not at all the same as seeing to it that one is there at 9:00,
nevertheless promising "to be at the station at 9:00" does pretty much come to
the same thing as promising "to see to it that one is at the station at 9:00." The
stit complement thesis for promising, Thesis 5-4, nicely explains why we should
have one paraphrase and not the other. Furthermore, paraphrasing reduces (but
does not remove) the temptation to make up a philosophy that tries to construe
promising as a relation between a person and an agentless state of affairs. By
concentrating on the stit paraphrase, we at least postpone the necessity of such
a construal.
Let us put the conclusion a little more sharply. Thesis 5-4, which we in-
troduced as a good thing to mean by the thesis (6) that promising is always
promising to do, was threatened by promises having the form "promise to be ...,"
as in (8). We met that threat with the claim that a "promise to be" is at least
5. Promising: Stits, claims, and strategies 103
uses; of course one might also use that form in speaking to a child as a kind of
pretense.)
The reader who continues on to 5B will recognize that we have in (12) a good
case of word-giving in the sense of Thomson 1990. Our aim here, however, is to
do as much as we can with the stit apparatus, even though it may temporarily
lead us astray in ways that will need to be corrected. And certainly we are not
yet at the end of our resources for defending the helpfulness of the thesis that
a promising is always a promising to do, that is, a promising to see to it that
something. Let us first try insisting that yes, promising that the sun will shine
is promising to see to it that something. What could that something be, given
that it obviously isn't that the sun will shine? We do not think that stit gives an
automatic answer to this question. Nevertheless, consider a concrete situation
featuring the promise reported in (12). Imagine some circumstances. What
was the point of the promise? Even without supposing that there is anything
like a uniform answer to this question, the generalized shape of the story is
not so difficult to make out. Say that Marie had been wondering whether or
not to take her umbrella. Paul in promising that the sun will shine in effect
says "don't bother." That, however, is not precise. There are many ways to say
"don't bother" that are not promises. What is special about the promise? Well,
as we might say, and as Thomson explicitly makes clear, the promising entitles
Marie to rely on Paul's declaration that the sun will shine. We are nearly in
a position to break out of the circle. The stit approach suggests looking for
something that Paul promises to do when Paul entitles Marie to rely on his
word that the sun will shine. Put this way, it is natural to look at the situation
when the sun fails to shine, as Paul promised. So suppose that 9:00 comes, and
there is Marie huddled in the pouring rain without her umbrella. What then?
The first thing to observe is that there is no hint of Paul's "breaking" a
promise in this situation as so far described. Presumably "breaking" involves
agency, whereas in our story the sun failed to shine without any help from Paul.
And yet we cannot let Paul entirely off the hook. To make the example easy
to follow, suppose Paul himself had a umbrella, but refused to share it with
Marie in the downpour. Suppose he said that all along that was his plan; in
promising Marie that the sun would shine he had no idea of helping her out
if his declaration turned out false. Then, even though promise-breaking is not
to be attributed to Paul, we ought to say that he made a lying promise. We
ought to say that the act carried out in uttering a sentence (12) is constituted
in part by a conditional promise to do, namely, a promise that Paul see to it
that Marie is helped or that her troubles are alleviated or that she receives
restitution or recompense for damages (or something) in the case in which the
sun does not shine. At the very least, Paul is promising to see to it that he is
graceful in taking blame if things go wrong. It is not only that he, as patient,
receives demerits if the sun doesn't shine, but that he, as agent, sees to it that
he accepts those demerits, even though, as seems important, he might have done
otherwise.
Clarification: In labeling Paul's act a "conditional promise to do," it is obvious
that we intend the condition to apply to the "doing," not to the "promising."
5. Promising: Stits, claims, and strategies 105
Warning: A conditional stit is not always a stit (see chapter 11). Further
warning: The conditional stit is just about bound to be in the scope of a future
tense, and so subject to Remark 5-5 on p. 103. For both of these reasons, and
inspired by Thomson 1990, this account of (12), while imbued with some of the
spirit of Thesis 5-4, tries not to confuse the spirit with the letter. More of this
later.
One last example, with a complement of a sort that lies between the agentive
complement, "that Paul picks up Marie at 9:00," and the agentless complement,
"Paul is at the station at 9:00." Suppose
Paul promised to see to it that Marie buys a train ticket. (13)
This of course makes eminently good sense. According to stit theory, however,
the complement should not be rendered by "[Paul stit: [Marie stit: Marie buys
a train ticket]]," since one person cannot guarantee the choice of another. In
chapter 10 we discuss various ways of understanding such examples, pointing
out the many differences among them. According to one of these ways, (13) is
usefully paraphrased using some idea of "probability":
Paul promised to see to it that it is highly probable that Marie (14)
buys a train ticket,
for the probability is something that Paul can, according to stit theory, reason-
ably see to. If, however, we think about many choices by Paul, as we did in
considering (12), it is natural to point out that there are and will be available to
Paul many choice points at which he can render it more or less likely that Marie
buys a train ticket. A minimal such notion is this: Paul never makes a choice
that excludes the possibility of Marie buying a ticket, if there is available to
him another choice keeping that possibility open. If we think about the content
in this way, we make at least some objective sense out of saying that Paul's
agency is relevant to Marie's buying a ticket. We can do this even though it
is (conceptually) impossible for Paul to guarantee that result. So this case lies
properly between the cases (5), in which Paul can, at least in approximation,
see to what he has promised, and (12), in which Paul's agency is irrelevant.
Let us summarize what is easy to see about promising when viewed from the
point of view of stit.
one cannot choose today what can only be chosen tomorrow (no choice between
undivided histories). For this reason, it is certain that the promise does not
guarantee that on the morrow Paul chooses to pick up Marie. So what can Paul
guarantee that wasn't guaranteed before his promise? One answer is this: He
guarantees that he is obligated to pick up Marie. He was not so obligated before
the promise; he is obligated after the promise; and it is the promising that is
responsible for this change in the world. It therefore seems helpful to articulate
(7) as follows:
[Paul stit: 06/0: [Paul stit: Paul picks up Marie at 9:00]]. (15)
5-6 SLOGAN. (Exploit nesting) Exploit the recursive power of modalities; that
is, look for ways in which to clarify by embedding one stit in the scope of
another.
5B From RR to promising
Thomson 1990 (The Realm of Rights, henceforth "RR") shares with stit theory
a reliance on a modal agentive construct that finds its source in Hohfeld 1919.
This makes for enough commonalities to have allowed us already, in 5A, to
have relied on some of the key ideas of RR, and we shall suggest that conversely
the use of some central ideas of stit theory can help to refine the RR account
of promising. (When we quote from RR, we make a few small changes in
symbolism.)
108 Introduction to stit
Cx,YPittDY,xP.
In this representation of claims there are two places, marked by X and Y, for
singular terms and one, marked by P, for a sentence. Thus, RR carries the
ideas of claims, duties, and privileges with the help of modal operators.
The example RR gives for the usefulness of the modal form is that this form
makes it easy to express the generalization that claims are closed under entail-
ment. The stit approach helps by suggesting that this generalization may not be
straightforward when one mixes agentive and non-agentive complements. The
stit approach, even without the introduction of stit notation, makes it natural to
question entailment-closure. In analogy with a favorite stit example (see 2B.l),
one wonders whether that X has a claim against Y that Y bandage at least
one injured person does in fact imply that X has a claim against Y that there
is at least one injured person.
RR explicitly states the following principle.
5-7 THESIS. (RR's unrestricted complement thesis for claims, duties, and priv-
ileges) Any sentence whatsoever may be put in for "P" (see p. 41).
This thesis is to be compared with the unrestricted stit complement thesis, The-
sis 2, and with Thesis 5-4 for promising. We especially note that even though
the two announce different targets, Thesis 5-4 and RR's Thesis 5-7 appear to
conflict in spirit, and we will say something about that shortly. We shall see
that for promising RR seems to adopt, or nearly to adopt, not indeed our Thesis
5-4, but some restricted complement thesis. We shall wind up endorsing this
line, and thereby suggesting, for promising, an account somewhere between our
Thesis 5-4 and RR's Thesis 5-7 (see 5C). Our immediate thrust, however, will
be to take the foundations of RR as they are, its unrestricted complement thesis,
Thesis 5-7, and all, and to follow up what RR makes of promising on that basis.
It nevertheless seems best to indicate at this point how the stit approach can
help make us wonder about Thesis 5-7, and especially about the quite different
ways that agency figures in the three English readings (16), (17), and (18).
5. Promising: Stits, claims, and strategies 109
Sure, that's confusing. That's the point: There is work to be done, and it might
not be easy. The considered choice by RR to use the locutions "discharge"
and "let" in connection with duties and privileges imply that a simpleminded
insertion of stit is not going to suffice. One may nevertheless hope to be guided
by the suggestion that arises in applying stit theory to deontic constructions,
namely, that being permitted to see to it that P is equivalent to not being
obligated to refrain from seeing to it that P (see 2B.9):
Note the delicate placement of the internal negation: neither on the outer stit
nor on P. One can imagine that something like the equivalence (19), which
unquestionably relies on the restricted complement thesis for deontic construc-
tions, might connect privileges and duties. Certainly this seems to work in the
simplest case, the case where the complement of both claim and privilege is
restricted to stit. That is, the following seems illuminating.
X has as regards Y a privilege of seeing to it that P <-> Not-( Y
has a claim against X that X see to it that Not-(A' see to it that (20)
P).
Since stit theory says that "refraining from doing" is "seeing to not seeing
to," stit theory allows the equivalence (20) to be dramatically shortened for an
agentive P, as in the following Hohfeldesque example.
110 Introduction to stit
We want to make it plain, however, that although one may reasonably hope that
the simple case will provide guidance, and that stit theory will clearly explain to
you how you are cheating yourself if you interchange "refraining from crossing"
with "not crossing," it would be optimistic to hope for a quick solution to more
difficult cases without extensive work.
5B.2 Word-giving in RR
Leaving aside foundational questions as to the relation between agentive or
quasi-agentive concepts and those of Hohfeld, we go on to promising. That
RR explicitly prefers unrestricted complements helps to explain the conceptual
order in RR according to which promising is taken as a species of word-giving,
which itself is taken to be a special case of assertion. Here is a sequence of
examples that accords with RR.
Paul asserts to Marie that he will be at the station at 9:00.
Paul gives his word to Marie that he will be at the station at 9:00.
Paul promises Marie that he will be at the station at 9:00.
This sequence minimizesand we think is intended to minimizethe need for a
restricted complement thesis for promising. In this respect the entire movement
is opposite to the spirit of stit theory, in the context of which one would be happy
to construe word-giving as a kind of promising (as in our treatment of promising
that the sun will shine at 9:00), and asserting as a kind of word-giving.4 But
that is not here to the point. Among these three, RR takes assertionand
especially assertion to someoneas a primitive among the array of concepts it
develops. It is consonant with the thrust of RR that assertion should be taken
as seeing to the existence of some kind of normative state, as we also explicitly
suggest in 6E. It is, however, a feature of the RR analysis that definiteness
about this seems not required. In any event, the other two members of the triple
are given sharp characterizations. Postponing for a while any consideration of
promising per se, we begin with two theses of RR, the most important of those
that govern word-giving (our italics).
We are led to see that on the RR account, at the end of the process what X
does when X accepts the invitation to rely is simply this: X sees to it that he
or she has a claim against Y that P. Let's write this down:
[X stit: X has a claim against Y that P].
That expresses what X "does" in accepting the invitation: X binds Y. So what
does Y do? On the RR account, Y's invitation amounts to giving X the power
to create the claim. But we do not need to invoke the metaphors "give" and
"create." Both are at bottom stits. Working from the inside out, the power is
nothing but a can-do:
Can:[X stit: X has a claim against Y that P}.
That expresses exactly the power that X has when X is in a position to accept
the invitation.5 Furthermore, "giving" X the power is also nothing but a stit:
[Y stit: Can:(X stit: X has a claim against Y that P]]. (22)
That is what Y does when Y invites X to rely on his or her word: Y sees to it
that X can see to it that X has a claim against Y that P.
Having gone so far with RR, one can see the likelihood that the time-sequence,
represented in (22) by the nesting, has been overspecified. It seems not to matter
that Y goes first and X second, that is to say, that Y empowers X. It seems
equally all right if X empowers Y:
[X stit: Can:[Y stit: X has a claim against Y that P]}. (23)
Why not? If for instance X asks a solemn question (p. 295 of RR gives an
example involving keys on a bureau), there seems no call to ask for an additional
act on X's part, after Y's answer, in order to construe Y as having been a word-
giver in the full sense, including uptake.
Nor does it seem wrong to think of a situation in which neither act precedes
the other, so that we should have a stit that is founded in the simultaneous
choices of two agents, a "strict stit" in the sense of chapter 10:
[{X, Y} sstit: X has a claim against Y that P}. (24)
What these three, (22), (23), and (24), have in common is that the complement
"X has a claim against Y that P" has come true due to the prior choices of
X and Y, both being essential. This important agentive idea, although not
attached to any of the stit locutions so far treated, seems to be a common
generalization of [a stit: Q] to [F stit: Q], where F is a set of agents, and of [a
stit: Q] as witnessed by a moment to [a stit: Q] as witnessed by a chain. Let us
describe it by means of an alternative semantic clause for strict stit that builds
in both witness by chains and the essentiality of each of a group of agents. (By
5
"A power is an ability to cause, by an act of one's own, an alteration in a person's rights"
(RR, p. 57). We have separated the "power" into its two natural components: A can-stit,
which represents "an ability to cause," and an expression of a claim, which is one sort of right.
The ability to do so is an advantage of stit.
5. Promising: Stits, claims, and strategics 113
5B.3 Promising in RR
We move from word-giving to promising. RR almost seems to make a point of
not taking the complement of "promise" as an agentive. We think, however, that
the details of RR can be improved by taking seriously Thesis 5-4 for promising,
even while rejecting it.
What, on the RR account, is the difference between a mere word-giving and
a promise? We quote four additional requirements from scattered locations on
pp. 299-301.
5-9 THESIS. (When a word-giving is a promising)
First, a word-giving is a promising only if the proposition asserted is in
the future tense.
Second, a word-giving is a promising only if the proposition asserted has
the asserter as its subject.
114 Introduction to stit
[F]ourth ..., unless the word-receiver cares whether the proposition as-
serted is true, and indeed wants it to be true, then even if there is uptake,
so that a word-giving does take place, what takes place is not an instance
of promising.
There you have your future tense, your promisor as subject, and your ascription
of agency (the stit). The only difference is this: In (26) the three pieces are
taken together, not separately.
Our replacement is obvious, and worth trying. This replacement is a signif-
icant modification of the spirit of RR's unrestricted complement thesis, Thesis
5-7, for it insists that the complement of "promise" must be a future-tensed
agentive with the same agent as the promisor. That may sound like the "same
subject" condition of RR, but it is not. From the beginning stit theorists have
supposed that the agent is a distinguishable or recoverable "part" of the "con-
tent" of an agentive. This is no mere verbal matter. See especially 1C and
3A. The replacement, however, is also a significant modification of a restricted
complement thesis, Thesis 5, in the sense of Thesis 5-4.
In order to appreciate the exact form of (26), one probably needs to read the
stit as a "deliberative stit" rather than an "achievement stit," though neither
of these constructions is entirely apt except for simple cases. The present point
is this: The choices that are available to the agent, and not just the outcome of
those choices, are to be envisaged as in the future of the promising. It is easier
to get straight on this with the deliberative stit. It is a separate point that Q
itself is likely to be well represented by a sentence in the future tense, so that
the outcome represented by Q follows after the choice that witnesses the stit.
The stit part of the sentence (26) expresses agency. But Thesis 5-9 (its third
part) gives three disjuncts. There is in the reach of that third condition not only
a promise (i) to do, but also a promise (ii) to refrain; and there is a promise
(in) to be. Of these only (i) seems to be covered by (26). What shall we do
about the other two disjuncts of the third condition given by Thesis 5-9 as to
when a word-giving is a promise? Luckily, however, stit already covers both
of these extra casesand in a unified way. With regard to ( i i ) , a promise to
refrain, it is an explicit thesis of stit theory that refraining is expressible by a
stit, and is therefore very different from mere not-doing. For example, that X
not trespass can be true of nearly everything, whereas only sometimes is X in
a position to refrain from trespassing, that is, in a position to see to it that X
does not trespass. For the latter, but not for the former, X must certainly be an
116 Introduction to stit
agent in the matter. With regard to ( i i i ) a promise to be, for example a promise
by B to be at A's office at 4:00, we think that here above all the distinction that
RR intends between (mere) word-giving and promising comes to the fore. If it
is merely a matter of where B is. then we have word-giving. But it becomes a
promise if it is a matter of B exercising some agency in getting to the office,
that is, if it is a matter of B's seeing to it that B is at A's office at 4:00.
The central point is this: To the extent that there is a difference between
(mere) word-giving and promising, the complement needs to be or at least in-
volve an agentive. A mere agentless state-description will not do.
In short, we suggest that a simple promise, when represented in the same
style as before, would involve a claim of X against Y that Y will see to it that
P. In full dress:
Y promises X to see to it that P [Y stit: Can:[X stit: X has a claim
against Y that will:[Y stit: p]]]
a verbal convenience, we say that the strategy "requires" that the agent choose
in accord with the strategy, even though the austere idea of a strategy does not
depend on deontic ideas. That, austerely, is it.
It is basic to the theory of stit and of strategies that you cannot stit the content
of a strategy. This is an elementary consequence of the principle of no choice
between undivided histories. It therefore seems that we have a sound theoretical
basis for modifying the stit complement thesis, Thesis 2, according to which
the complement of stit must be a declarative sentence. We have not, however,
laid sufficient groundwork for such a modification. The stit complement thesis
is at least in part linguistic, and although in chapter 13 we suggest an abstract
theory of strategies, we have not suggested any linguistic devices with which to
carry the content of a strategy. Perhaps none are necessary; perhaps it suffices
just to consider, at least for a first approximation, sentences, Q, for which
there is a strategy such that Q is true at the moment of promising paired with
each member of the smallest set of histories that the strategy really guarantees
(Definition 13-5). Some such suggestion would appear to cover, for example,
future-tensed stit sentences, and also combinations of conditional stits such as
"if the sun doesn't shine, I will apologize, and furthermore if it rains, I will lend
you my umbrella."
It is not, however, sufficient to consider just this first approximation. The
difficulty is that the approximation will only work for the simplest cases such as
Paul's promise to pick up Marie, (7). It will not work to cover our understanding
of Paul's promising something that Paul can influence but cannot guarantee, for
example, (13), nor of something over which he has no control, for example, (12).
We offer a better approximation, in 5C.2, that is intended to cover these cases
uniformly.
wonders, however, if one can use the theory of strategies to draw a distinction
in an interesting and helpful way. Such a theory might permit one to refine
the "complex responsibility" of (21) in two different ways, ways that distinguish
(27) from (28). Speech act theorists sometimes distinguish the two by saying
that assertion or word-giving embodies a fit of words to the world, whereas a
promise embodies a fit of the world to the words. (The idea comes perhaps from
Anscombe 1957, p. 4.) We shall show that the theory of strategies can help give
definite meaning to the distinction drawn by these metaphors.
The theory of strategies would allow the following, which we present in rough
form.
For (27), the word-giving, the content might be represented by
a strategy that requires no choices whatsoever that are relevant
to Paul's seeing to it that he is at the station at 9:00, but that (29)
prescribes that Paul compensate Marie if he does not see to it
that he is at the station at 9:00 (see Definition 5-15).
This is a "words-to-world-fitting" strategy. Although it does call for some
choices involving compensation when the words do not fit the world, the cen-
tral point is that the strategy does not call on Paul to make choices favoring
(objective) possibilities fitting his words. In other words, Paul can follow this
strategy without doing or choosing anything that favors his seeing to it that he
is at the station at 9:00. We propose this as an explication of the metaphorical
phrase, "fit the words to the world."
Note also that the word-giving strategy is "primary" rather than "secondary"
in the sense of Definition 13-7. That is, it says nothing whatever if Paul makes
an excluded choice, for example, if Paul chooses not to compensate Marie.
For (28), the promising, the content might be represented by a
strategy that (i) prescribes that, if possible, Paul choose in a way
that guarantees that he see to it that he is at the station at 9:00,
(ii) prescribes that, if possible, Paul keep open the possibility
of Paul's later guaranteeing that he sees to it that he is at the (30)
station at 9:00, and (iii) prescribes that Paul compensate Marie
if in fact he "infringes his promise" by arriving at a moment at
which it is no longer possible for him to see to it that he is at
the station at 9:00 (see Definition 5-13).
This is a "world-to-words-fitting" strategy. It calls for choices relevant to getting
Paul to the station at 9:00. We propose this as an explication of the metaphorical
phrase, "fit the world to the words."
Observe that this strategy is, crucially, "secondary" rather than "primary,"
since it imposes requirements on Paul (that he compensate Marie) when he
violates the strategy by infringing his promise. As secondary, the strategy con-
tains a "contrary to duty" (so to speak) component, namely, the compensatory
sub-strategy.
There is evidently a large difference between the strategies described in (29)
and (30), which (inconclusively) suggests that the theory of strategies can help
5. Promising: Stits, claims, and strategies 119
clarify the view of those of us who feel that there is an important difference
between (27) and (28). It offers hope that there is a useful distinction to be
made between word-giving that an action will be performed and promising to
carry out that action. It provides a way to try to make rigorous and objective
sense out of the metaphorical difference between "fit of words to world" and "fit
of world to words."
can see exactly how "double time references" work without presuming anything
whatever about the various syntactic ingredients of P.
The fundament of those semantics is this: Because of indeterminism, the
truth value of a sentence may depend both on the moment of evaluation and
also on the history of evaluation.8 That is, the metalinguistic predicate "true
on" must be relativized both to a moment and to a history, so that one says
"A is true on m/h," where h is one of the histories to which the moment, m,
belongs.
It follows that the sentential complement, P, of Target 5-10 has a rich se-
mantic content in terms of its truth value at each moment-history pair, m/h.
In principle, the entire semantic content of P is available to us in explaining
the contribution of P to the content of Target 5-10. An essential feature of
our account, however, is that promising uses only a small part of this richness;
namely, promising uses only the pattern of truth values of P at pairs that feature
a single moment,TOO,the moment of promising. The sentential complement, P,
of Target 5-10 contributes nothing but its pattern of truth values at the various
pairs m o / h , where mo is fixed once and for all as the moment of promising, and
where h ranges over all the histories to which this moment belongs.
It is striking that this be so. One would have thought that one would have
to consider the truth value of P at moments after the moment of promising in
order to get the right "world to words" fit. As we saw in considering Target
5-11, however, in looking at these later moments, we are not considering the
truth value of P itself. In this special example, we look instead at the truth
value, at later moments, of an ingredient of P, namely, (32). We made a point
out of the fact that we do not consider the truth value of P = (31) itself.
By attending to the pattern of values of P only at the moment of promising,
we obtain a uniform account that works for arbitrary P, regardless of syntactic
structure. (Naturally, in the context of this book, we presuppose that P exhibits
a structure for which it is illuminating to speak of a rigorous branching-histories
semantics. We offer no contribution to other aspects of the theory of promises.)
We claim that it is possible to forget the grammatical complexities of P, and
consider only its pattern of truth values at m0/h, whereTOOis the moment of
promising and h is any history through that moment. Our claim is that this
pattern supplies exactly the right semantic content for the promising to be a
promising that P. We claim that in analyzing the content of Target 5-10, we
need use only "P is true on m0/h," where mo is the moment of promising, and
h is one of the histories to which that moment belongs.
8
This idea was introduced in 2A, and will be spelled out at length in chapters 6 and 8. If
context-dependencies are present, and they must be considered for an accurate understanding
of the language of indeterminism, then truth also depends on the moment of utterance, and
perhaps also on the place, speaker, etc., in a manner indicated in those chapters. (There
is, given indeterminism, no such thing as "the" history of utterance; see 6D and Question
8-4.) The moment and history "of evaluation" must be seen to be separate from these. In
these terms, observe that the moment of promising represents a moment of evaluation, not the
moment of utterance of a promise-report such as (9) or Target 5-10. With regard to Target
5-10, mo need not be "now," X need not be "I," and Y need not be "you." Compare the
treatment of direct-discourse assertion in 6E.
122 Introduction to stit
A is true at m0/h1
A is true at m0/h2
A is not true at m0/h3
A is not true at m 0 /h4
In these circumstances, Sett-true(mo, m1, A), since A is true at m0/h for every
history through m1; and Sett-false (mo, m2, A), since A is false at m0/h for
every history through m2. The critical point is that the role of a later moment,
for example m1 or m2, is not as a place to evaluate A. Instead, the later moment
supplies a group of histories over which "h" is to range while evaluating A at
m0/h. In double time reference, the earlier moment supplies the moment of
evaluation, while the later moment supplies a group of histories of evaluation.
should not be likely to say in this case that X "broke" the promise, since Y was
paid after all, and indeed it seems likely that we might say that X "kept" his
promise. Nevertheless, putting at risk the formerly secure ability to keep the
promise seems to deserve some sort of evaluative expression.
5-14 TARGET. (Analytic target) At m0, X gives Y his or her word that P.
ii. Suppose X arrives at a moment m1 such that Sett-false (m0, m1, P), so
that no matter the historical continuation of m1, P was false at the mo-
ment of word-giving with respect to that historical continuation. Then
at that moment m1, we say that the word-giving-that-P has been "im-
pugned." At any such moment, the strategy, S0, prescribes that X "com-
pensate" Y in a way that we do not spell out, but that would certainly
involve a sub-strategy that would work toward compensation until com-
pleted, or perhaps until some "deadline" was reached.
iii. Suppose X arrives at moment, m1, at which neither Sett-true(mo, m1,
P) nor Sett-false (m0, m1, P); that is, whether or not P was true at the
moment of word-giving is historically open at m1. Here is where there is
room for X 's agency directed toward P. Since, however, it is only a matter
of word-giving-that-P, and not of promising-that-P, nothing is prescribed
for X to choose. No effort of fitting world to words is called for. It just
doesn't matter.
The last clause, the one dealing with moments at which it is open whether P is
true at mo, is where the two types of strategies differ. Here is where the present
analysis puts the essential difference between promising that P and word-giving
that P. Promising typically requires that the promisor do or choose something
at some such moments, moments after the moment of promising and before it is
settled whether the promise is satisfied or infringed. Word-giving never imposes
any such requirement. The analysis of these two concepts via their strategic
content emphasizes this.
Q is true at m1/h1.
Q is not true at m1/h2.
Q is true at m2/h3,.
Q is not true at m2/h4.
In this simple case, one can use stit theory (the dstit variety) to calculate (or
just see) that at each of these four pairs, "X sees to it that Q" has exactly the
same truth value as Q:
Also, since m1 and m2 are the only moments at which X has a nonvacuous
choice, "X sees to it that Q" is false everywhere else. Therefore, by ordinary
tense logic, "X will see to it that Q" must follow along, history for history,
when evaluated at the moment of promising or word-giving:
The strategy for the promise represented by (34) therefore dictates that X must
not "turn right" upon reaching either m1 or m2- If he does so, if he chooses h2
upon reaching m1, or if he chooses h4 upon reaching 7712, the strategy for (34)
will be infringed, and will call for compensation of Y. We might well say that X
has broken his promise to Y since by "turning left," X could have guaranteed
satisfaction of the promise. The point is that if X throws away a "last chance,"
X has violated the strategy that represents the content of a promise, even if
later X compensates Y.
The strategy for the word-giving represented by (35) is different precisely with
respect to what it dictates at m1 and m2. At these moments, it says nothing
at all. X is free to choose as he likes, so far as the word-giving strategy goes.
If X chooses at m1 or m2 to "turn right," X has not violated the strategy that
represents the content of the word-giving that he will see to it that Q. If at
mo X has merely given his word to Y that he will see to it that Q, then on
the account we have given, X undertakes no responsibility for guaranteeing the
truth of (33) at m0. For avoiding violation of the word-giving strategy, it suffices
if X compensates Y even after choosing to guarantee that (33) is false at mo!13
13
We have taken the word-giving target from RR. It is therefore essential to observe that the
analysis of word-giving to be derived from RR via quote (21) does not give the present result
for word-giving. The quote from RR fits our theoretical account of "promising" better than
our account of "word-giving." Our aim in choosing to speak of "word-giving" is not, however,
to saddle either RR or indeed ourselves with a particular analysis. Its sole purpose is to exhibit
how the theory of strategies provides a chance to help sharpen any such analysiswithout
invoking mentalistic concepts.
5. Promising: Stits, claims, and strategies 129
By giving us insight into the difference between two ideas naturally (but not
inevitably) associated respectively with the English expressions "word-giving
that one does" and "promising that one does," the proposed application of the
theory of strategies passes a critical test.
This page intentionally left blank
Part II
Foundations of
indeterminism
This page intentionally left blank
6
133
134 Foundations of indeterminism
6A Preliminary considerations
Before turning to the exact analysis that we think is required in addressing the
"assertion problem," we briefly lay out some preliminary ideas: a generalization
of McTaggart's A- and B-series, a sketchy account of the Thin Red Line, a
clarification of the idea of indeterminism with which we work, and a rehearsal
of the basic postulates of branching time theory. Throughout it is essential to
keep in mind that we are not considering "second order" indeterminism, that
is, indeterminism as a feature of our mind or of our language or of our theories.
We are always taking our questions to concern "first order" indeterminism, that
is, indeterminism as a feature (if it is one) of the causal order of our world.
from the point of view of an A-theorist, consult McCall 1976, McCall 1984, and
McCall 1994.)
more, the coin will land heads, but it needn't do so." (See OhrstrOm 1981 for
an explicit call for a Thin Red Line; Brauner, 0hrstr0m, and Hasle 1998 for
discussion; and Barcellan and Zanardo 1999 for formal development, discussion,
and a suggestion of applicability in computer science.)
We shall argue that in spite of its stalwartness the Thin Red Line should be
forsworn. We shall argue, against an established opposition, that one can make
much better sense of an indeterministic, branching structure for our world by
abandoning the idea of an actual future as distinguished among the possibilities.
We shall furthermore argue that the Thin Red Line doctrine turns out, on closer
scrutiny, to have unpalatable consequences. We shall call the view that in spite
of indeterminism one neither needs nor can use a Thin Red Line the doctrine of
the open future. We intend this terminology to make contact with the intuitions
(but not the style) of many decades of thinking about indeterminism, and to
make contact with an analogy that we develop in detail between the "openness"
of expressions like "the coin will come up heads" and the well-known "openness"
of "x1 is brindle."
We first lay down constraints upon the notion of indeterminism relevant to
the present discussion (6A.3). Explaining the open future under those con-
straints involves two related tasks. The first and conceptually prior task is
extra-linguistic, requiring the elucidation of a notion of indeterminism that ap-
plies to our world (6A.4). This account will embody an understanding of a
B-order that is not a series, but instead is free from any assumption of linear-
ity, an understanding that will explain how this very world, if indeterminism
is ever true, is replete with possibility. The second task is a linguistic one, in-
volving the careful development of a semantic theory for temporal discourse in
an indeterministic world (6B). At this point we will have finished our positive
development of the open future doctrine.
A natural basis for doubt about that doctrine is that it appears unable to
make sense of one who asserts that, bets that, or wonders whether there will
be a sea battle even when it is clear that there might not be. A special case of
this problem is formulated in 6C as the "assertion problem." One superficially
reasonable response is to postulate a Thin Red Line. In 6D, however, we argue
that the Thin Red Line doctrine, in any of the versions of that doctrine that we
consider, has unacceptable consequences, ranging from a mistreatment of actu-
ality to an inability to talk helpfully about what would have happened had what
is going to happen not taken place. In 6E we "solve" the assertion problem
by arguing that our framework does make sense of such acts and attitudes once
we come to see that to assert that A is to do something that has a normative
significance no matter how history carries on.
EXAMPLE. (Throw of a die) This morning we threw a die. It showed six; but
there were five other possible outcomes. Then we loaded the die, and threw it
again. Again it showed six; but this time, that was the only possible outcome.
The first throw-to-six transition was indeterministic. the second was not. We
need a concept that can be used in this local and particular way to describe
individual transitions from one event (the initial of the transition) to another
(its outcome). In this study we simplify by worrying only about the temporal
dimension of locality. We will suppress the relativistic considerations arising
from the spatial aspects of indeterministic transitions. (See Belnap 1992 for
an exploration of indeterminism in a relativistic setting.) The idea of locality
just described should rightly make you think of "single case probabilities." But
we need a pre-probabilistic concept of indeterminism. Perhaps when the first
throw was made there was a 1/6 chance of six. Before the numbers come into it,
however, there is the idea that given the throw that in fact came up six, there
were five other possibilities. To say that does not require any probability mea-
sure; only possibilities are needed. These possibilities are themselves "local."
The actual transition was throw-to-six, the other possibilities were the transi-
tions throw-to-one, throw-to-two, and so on. Nothing as global as a "world" or
"theory" or even "law" comes in this early, and no numbers representing proba-
bilities are part of this concept. Quite to the contrary, any concept of probability
must rest on a concept of possibility (sometimes called "the probability space").
We need an objective concept of indeterminism. We mean that the question of
how many possible outcomes there were for a certain throw shall be classed with
the question of how many ears there are on a certain Scottie, and contrasted with
questions that are explicitly about who thinks what about what, and whether it
is reasonable to do so. Our aim is to theorize about a concept of indeterminism
that does not require simultaneous explicit theorizing about people and their
thoughts or norms or culture. Thus, we are after a concept of indeterminism
that does not put the number of possible outcomes of a certain throw in anyone's
head, or make it relative to laws or theories, or have it depend on the status of
a conversation, or depend on what people care about. All to exactly the same
extent as the number of ears on a certain Scottie. The most explicit contrast is
with "epistemic indeterminism," also called epistemic possibility, as codifying a
form of ignorance.
There is a difference between calling a transition indeterministic relative to
some feature, and calling it feature-independently indeterministic. By defini-
tion a transition is feature-independently indeterministic if there is more than
138 Foundations of indeterminism
one possible outcome for its initial, and otherwise it is feature-independently de-
terministic. In this study we shall be content with these "feature-independent"
ideas because of their foundational role. The feature-relative ideas are, however,
of great importance. When every possible outcome of an initial has a certain
feature, we may well say that the transition was deterministic with respect to
that feature. Given the throw, for example, not only did the die come up six,
but it landed on the floor. Since it landed on the floor on every one of the
other five outcomes, the transition was deterministic with respect to landing on
the floor. This sort of feature-relative determinism (for a particular feature),
which is doubtless of enormous importance both for science and common sense,
is evidently consistent with feature-independent indeterminism.
We want a concept that can apply to a transition de re, without requiring
some description under which the initial falls. No matter how you describe the
throwing of the die, even if you are confused enough to refer to it as the holding
of a martini by that man in the corner, still, it has just six possible outcomes.
Take the ancient example:
EXAMPLE. (Rest or motion) She remained at rest. But she wasn't tied up. So
she could have moved or remained at rest. On the other hand, given the exact
history of her beliefs and desires, there was evidently but one possible outcome,
not two.
Here the initial has been described in two different ways. The first patch of
rhetoric suggests free will, the second suggests instead iron-clad determinism.
To insist on a concept of indeterminism that is fundamentally de re is to disbar
pretending to plausibility of contradictory phenomena via colorful redescrip-
tions. You can't have it both ways. Either her concrete situation, no matter
how described and no matter what was "similar" to it, admitted two possible
outcomes or it admitted only one. Thinking de re prevents you from evading
the problems of indeterminism (or determinism) by switching descriptionsas
if you could change the number of ears on a particular Scottie by describing it
as very like a whale.
For our continuation to have interest we need to consider only the weakest
possible indeterminist claim, namely, the existential claim that some transi-
tions are indeterministic. The existential claim is obviously consistent with the
commonsense view that numerous transitions are more or less feature-relative
deterministic in interesting ways: The stars in their courses bravely run, fire
burns here and in Persia, and although the ways of men are various, the same
motives are followed by the same actions.
A concept is hard if you have a rigorous theory for it, or at least if you wish
you had such a theory and are miserable to the extent that you don't. If in
contrast you are happy with some interesting stories, some "paradigm cases,"
or with a sketch of an outline of a skeleton of a never-to-be-supplied theory,
the concept is soft. Wittgenstein's concept of game is soft; von Neumann's is
hard. We are interested in indeterminism as a hard concept. We want a rigorous
theory.
6. Indetcrminism and the Thin Red Line 139
right to think perspectivally of m1 as in the past of m2. On the other hand, one
should say that mo is in the "future of possibilities" of m1not simply in its
"future." The reason is that it is intuitive that a future of possibilities, unlike
a future history, can contain incompatible moments.
The final idea is Our World. Start with this very moment (yours or ours; at
this level of idealization it doesn't matter). Now form the set of all moments
that are connected to this very moment by means of any zigzag combination
of the causal ordering or its converse. That is, include all moments that you
can reach by means of a "causal path," no matter how complicated. That is
what we mean by "Our World" construed as a set of moments.3 The theory of
branching time is, as we give it, a theory only about Our World and not about
any other. Thus by "moment" in what follows we always mean "moment of Our
World." Moments are thereby not to be confused with creatures of the mind.
Moments are mundane.
The theory technically has only two primitives, Our World and the causal
order. The theory uses these primitives in four postulates: nontriviality, par-
tial order, historical connection, and no backward branching. These, which we
introduced in 2A, we summarize in proper form in 3 and study in chapter
7. Here, for easy reference, we briefly repeat their meaning. Nontriviality is a
technical postulate that says that Our World is nonempty. Partial order de-
scribes the causal relation, /<, as reflexive, transitive, and antisymmetric. This
last signals that moments of Our World must be understood as concrete, unre-
peatable events rather than abstract "states" or "times." Historical connection
says that every two moments in Our World have a common historical ancestor.
It is the postulate that describes Our World as one world. Finally, no backward
branching says that all branching is forward, never backward. No backward
branching reflects the uniqueness of the past. Starting from any moment there
is exactly one chain of moments in the backward direction, its past history.
The conviction is that although Our World contains alternative incompatible
possibilities in the future of possibilities of a given moment, there are no incom-
patible moments in the past of any moment. A moment may have more than
one possible outcome, but not more than one possible "income."
A history (we use "h") is defined as a maximal chain of moments, a complete
possible course of events stretching all the way back and all the way forward.
We may consider a new "same-time" primitive, Instant, that renders all his-
tories isomorphic, so that it makes sense to have a doctrine of linearly ordered
instants of time to complement the theory of branching moments. An "instant"
of linear time can be defined as a set of same-time-mated moments; see Post.
5. Given Instant, it makes sense to ask what might have happened "at this
very instant." The primitive seems convenient both for scientific approxima-
tions and for fashioning persuasive illustrations, and earlier chapters have relied
upon it in conceptualizing the achievement stit. One may question whether it
is finally respectable either scientifically or from the point of view of common
3
In the theory of branching space-time, Belnap 1992 uses "Our World" for a similarly
motivated set of point events, instead of moments. In both cases the defined term names a
set that exhauststo the extent admissible for an idealization of the given kindour world.
6. Indeterminism and the Thin Red Line 141
6B Parameters of truth
As we indicated in 6A, there is a problem about speech acts using future-
tensed sentences in the language of branching time. We take up assertion as
a specialbut surely centralcase. Crudely put, "the assertion problem," as
we call it, arises because given indeterminism, it would seem as if future-tensed
statements "have no truth value." (This dangerous phrase receives a technical
definition and an accompanying warning in 6B.7.) The reason that this is a
problem is that it seems to make sense to assert a predictive statement even
in the face of indeterminism. Since, however, such a statement "has no truth
value," how can it make sense to do so?
In order clearly to see the nature of the assertion problem, we adumbrate
a perspective on a variety of linguistic devices that have been considered by
philosophical logicians. All of these devices involve relativization of truth to
one or more parameters such as domain, interpretation, and assignment of
values to variables as in first-order logic, and to further parameters such as
moment, history, world, time, place, speaker, addressee, demonstrated object,
presupposition-set, and so on. By speaking in this general way about the rela-
tivization of truth to parameters, we hope (i) to elucidate the assertion problem,
(ii) to sort out proposals to solve that problem by introducing a Thin Red Line
representing the "actual future," (iii) to explain our own discontent with those
proposals, and finally (iv) to suggest our own solution. We do not, alas, know
how to do all this without introducing some auxiliary semantic ideas, chiefly
ideas that are derived from Prior 1967, Thomason 1970, and Kaplan 1989and
of course Tarski. In chapter 8 we go over many of these ideas in detail; our
treatment here, which is casual, is intended as a modest introduction.
is, for each point, ( z 1 , ..., zn), whether or not z1, ..., zn
depends on the following: for each point ( z 1 , ..., z'n), whether or
not z1, ..., z'n \= A. In other words, we suppose that our semantic
understanding of, for example, O(A) in terms of certain parameters
is a well-determined function of our semantic understanding of A in
terms of those parameters.
In short, the semantics of L is given by a recursive explanation or definition of
"z1, ..., zn \= A"
Easily lost amid the abstractness is the central idea, derivative from Tarski,
that if you wish to give your semantic explanations or definitions by means
of a simple compositional recursion that follows grammar, then denotation and
truth must be relativized to parameters. It will be important to our treatment of
the assertion problem that we keep track of the individual nature and purpose
of these various parameters. We do this as an approach to a key question:
Given that some philosophers have introduced the "Thin Red Line" (the "actual
future"; see 6A.2) to help with understanding the future tense, exactly how
does the Thin Red Line work as a parameter of truth? Treating parameters in
general will provide a firm foundation when we turn to answering this question.
From the point of view of the assertion problem, the most profound division
of parameters is into the "mobile" versus the "immobile," a rigorously definable
division that we now informally and roughly explain.
6-1 DEFINITION. (Mobile versus immobile parameters)
If L has any connective whose recursion clause requires shifting some
parameter, Zi, then Zi is mobile with respect to L.
More ponderously, if for some connective, O, there is a point ( z 1 , ..., zi,
n ) such that the recursion clause determining whether z1, ..., zi, ...,
(A) requires considering whether or not z1, ..., z'i, ..., z'n \= A for
some point such that zt is not identical to z't, then the parameter, Zi, is
mobile with respect to L.6
If, however, the value zi of Zi can be held fixed throughout the entire
recursion, then Zi is immobile with respect to L.
More ponderously, suppose that for every connective, O, of L, "z1, ...,
can be defined without considering whethe
except perhaps when zt = zi. Then Zi is immobile with respect to L.
This division into mobile versus immobile parameters, even though difficult to
keep in mind, is essential to understanding the assertion problem as we conceive
it. We return to this point. Now we lay nearly all of our cards on the table,
all at once, by listing the various parameters of truth whose understanding we
think is critical to understanding indeterminism. We divide parameters for L
primarily into mobile versus immobile, and then we subdivide them into types
that depend on the underlying purpose of each parameter.
6
This (informal) definition takes into consideration that some connective might need to
vary some Zi along with Zi.
144 Foundations of indeterminism
and let us consider (5) as stand-alone. We know that the past tense moves
evaluation of "Meg is hungry" backward to an earlier moment; but we need an
initial moment at which to start this backward motion. This initial moment is
naturally identified as the moment determined by the context of use. We will say
that the context initializes the moment of evaluation for sentences considered
as stand-alone.
This is the second role of the context parameter, adumbrated by Kaplan, p.
595, in (of course) his own vocabulary and in a slightly different connection. The
first role of the context parameter is to help with context-dependent expressions,
and its second role is to start the process of semantic evaluation of sentences
considered as stand-alone at the very context of the speech act that uses that
sentence as a vehicle.
At this point we have merely stated how the three mobile parameters are to
be classified. Since this classification is crucial to the assertion problem as we
eventually describe it, mere statement is not enough. Here is our reasoning.
instead of "O, c, a, m/h \= A" as in (4) on p. 147. The notation (6) will
serve for every sentence, including those considered as embedded. Second, we
cater to the special needs of sentences considered as stand-alone. We do this by
evaluating them only at points such that the moment of evaluation is identified
with the moment of use:
Observe that the moment of evaluation is identical to the moment of use; that
is what we mean by a context-initialized point, the kind of point at which we
must evaluate A when it is considered stand-alone.
however, do not in fact provide an initial value for variables such as x1. Not
only is there no fact of the matter, but it is probable that no language on Earth
subscribes to a conventional context-of-use-determined assignment of values to
its bindable variables.8 There just is no such convention; and if there were, it
would have no discernible purpose. We mean that there would be no purpose for
the users. It is not germane that pretending that such a convention exists might
make things easier for some descriptive logician. We intend to be agreeing with
Kaplan 1989, pp. 592-593, which contains the only pertinent discussion known
to us. Our chief point, however, is that the context is supposed to be fact-of-
the-matter, and there is no fact of the matter regarding an initial assignment
of values to the variables.
crucial respect histories are not like possible "worlds": Unlike worlds, histories
overlap, so that a single speech act will typically belong to many possible histo-
ries; and that is why the phrase "the history of the speech act" is impermissible.
These points will become sufficiently central to our treatment of the assertion
problem as to warrant highlighting.
In one sense, these statements are true by convention, since we built their truth
into the design of our mini-language, L. We nevertheless label them "theses"
because at a deeper level they are debatable, and we have given arguments in
their favor.
Quantifiers. m, mc, a, m/h \= VxiA iff m, mc, a1, m/h \= A for every
assignment a1 that does not differ from a except perhaps at the variable xi.
Settled truth and historical possibility. m, mc, a, m/h \= Sett:A iff for
every history h1 to which m belongs (that is, for every possible future for m),
m, mc, a, m/hi \= A. For Poss:A, change "every" to "some."
6. Indeterminism and the Thin Red Line 153
x1 is brindle. (9)
So the assignment and the history parameter are deeply alike: Each of these
two parameters is mobile but uninitialized by context, and each offers examples
of sentences that, when considered as stand-alone, are open.
In the service of setting the assertion problem, we take our semantic claims
just one step further. We claim:
In the absence of a specific convention, if a stand-alone sentence, A, is
open in a certain (mobile and non-initialized) parameter, Z, then there
is at least one context-initialized point such that it makes no sense to say
that A has a truth value at that point independently of Z.
In carefully controlled but more colorful language, we express this claim by
saying that if a stand-alone sentence is closed neither by independence nor by
initialization with respect to some parameter, Z, then "it has no truth value"
at some context-initialized point save relative to Z. More generally, if there is
any parameter relative to which a stand-alone sentence is open, we say that
by definition the sentence "has no truth value" at any context-initialized point
that bears witness to the openness. Use of the expression "has no truth value"
is dangerous; we try to minimize the danger of misleading by the upcoming
"Observation," and by emphasizing here that the phrase is defined. 9 Because
of the definition, we have the following.
6-6 SEMANTIC THESIS. (Stand-alones open in either the assignment or the his-
tory parameter sometimes have no truth value) Fix the model, m, and the
moment of use, mc. This suffices to give (8), considered as standing alone, a
truth value: "m, mc \= (8)" makes sense. So (8) "has a truth value." But m and
mc do not suffice to give a truth value to sentences, considered as stand-alone,
that are either assignment-open like (9) or history-open like (10), even if their
evaluation is restricted, as is appropriate for stand-alones, to context-initialized
points. Sentences like (9) or (10) may well "have no truth value." In symbols,
i. "m, mc \= Meg is hungry" makes sense since "Meg is hungry" is closed
in each of the mobile parameters. We could indeed define "truth" given
model and context as follows: For every (or some, or most, or your favorite)
a and h, m, mc, a, mc/h \= Meg is hungry.
ii. "m, mc \= x1 is brindle" does not make sense since ux1 is brindle" is open in
the assignment parameter. The definition proposed in (i) would of course
be formally correct; our complaint is that it would not be "materially
adequate." Note in particular that in contrast to (i), since "x1 is brindle"
is open in the assignment parameter, here it would make a difference
whether one framed the definition with "every" or "some" or "most" or
"your favorite." In short, it can happen that (9) "has no truth value."
iii. "m, mc \= WilL:the coin lands heads" does not make sense. (Ditto.)
9
We do not prejudge whether or not there is a useful connection between our defined idea
of "no truth value" and other notions such as those used in cases of presupposition failure,
vagueness, ungroundedness, etc.
156 Foundations of indeterminism
OBSERVATION. (No gaps) To say that open sentences such as (9) or (10) "have
no truth value" is not to say that open sentences have some third truth value,
or a third special status.10 It is even a mistake to think of open sentences as
introducing "truth value gaps" that might sometimes be filled by introducing
"supervaluations."11 No one makes these mistakes about assignment-open sen-
tences. As is evident from our strictly parallel treatment of the semantics of
assignment-open and history-open sentences, we think that they should never
be made. As a preventative against the temptation to start speaking of "truth
value gaps," it is healthy to intone that "given a model and a context, an open
sentence always has a truth valueonce a suitable value is supplied for each
parameter in which it is open."12
mentions assertion. They are purely semantic. Now we state some claims
about how various kinds of open sentences relate to assertion. In the following,
when we speak of "asserting A" we intend it as short for "using A as a vehicle
of assertion, so that what is asserted is expressed by A."
6-7 ASSERTABILITY THESIS. (Which stand-alone sentences are assertable?)
When we ask in the following whether or not a sentence is "assertable," we do
not mean anything like "warranted" or "correctly assertable." These meanings
are not what is wanted for our assertability thesis.
ii. Assignment-open sentences such as (9) "x1 is brindle" are not assertable.
It makes no sense to use one as the vehicle of an assertion.
hi. History-open sentences such as (10) "Will:the coin lands heads" are gen-
uinely assertable. It makes perfect sense to use them as vehicles of asser-
tions.
These claims do not go without saying; we argue for them.
Ad ( i ) . The assertability of triply closed sentences is paradigmatic. Seman-
tically, m, mc \= A makes sense for these sentences, so that they have a clear
model- and context-dependent truth value, a fact that seems to warrant their
eminent assertability.
Ad ( i i ) . It makes no sense to assert an assignment-open sentence such as
"x1 is brindle." It is literally senseless to use that sentence as a vehicle for
assertion. To put the matter a little more circumspectly, such anassignme
open sentence is not a proper vehicle for assertion. A purported assertion using
such a sentence would be defective. The defect is radical: It is not just that
no one could know what was asserted; more than that, nothing is asserted by
making such an utterance.
Observe that since (ii) is about assertion rather than just semantics, it does
not follow from Semantic thesis 6-3, Semantic thesis 6-5, and Semantic thesis 6-
6. We lay (ii) down as a new fact, a fact that describes our practice of assertion
(as opposed to mere utterance). No one would understand you if you said, using
direct quotation, "Jack confidently asserted the sentence, 'x1 is brindle'."13
13
Red herrings abound. There could be a special convention permitting, e.g., dropping
universal quantifiers on stand-alone sentences used as vehicles of assertion, and indeed there
is such a convention among some groups of mathematicians and other technical workers. The
convention is, however, "merely conventional," requiring addition to our underlying agree-
ments about semantics and the practice of assertion. Furthermore, it remains the case that
the convention covers only a few cases.
Indirect discourse would offer a separate topic. "Jack asserted that x\ is brindle" makes no
sense as a stand-alone sentence (since that sentence is itself assignment-open), but it might
make some sense if it is embedded in a quantificational connective: "For some x1, x1 is owned
by Mary, and Jack asserted that x1 is brindle."
Switching from asserting to uttering is yet another way to change the topic. Of course,
using direct discourse, we can truthfully say that Jack uttered the sentence, "x1 is brindle,"
158 Foundations of indeterminism
i. Our semantic account of triply closed sentences jibes with our assertability
claims: They do have a definite truth value (given model and context),
and they are paradigmatically assertable.
ii. Our semantic account of assignment-open sentences fits with our asserta-
bility claims: They "have no truth value" (given only model and context)
and they are not assertable.
iii. Our semantic account of history-open sentences seems severely in tension
with our assertability claims almost to the point of apparent contradiction:
Like assignment-open sentences, history-open sentences "have no truth
value" (given only model and context). But if they have no truth value,
it would seem that they would be no more assertable than assignment-
open sentences. After allif we may be permitted language known to be
untrustworthyit is certain that to assert A is to assert that A is true.
We nevertheless claim that, in spite of "having no truth value," it is indeed
proper to assert a history-open sentence, and even to assert that it is true.
In the end we confront the assertion problem head on, but first we discuss
three tempting ways to avoid the problem. These all have in common the view
that, contrary to Assertability thesis 6-7(iii), it is not in fact possible to assert
a history-open sentence, precisely on the grounds that even given model and
context, a history-open sentence has no truth value.
Way number one admits that our language contains (something like) history-
open sentences, but argues that they are not assertable. The reasoning might
be that the practice of assertion makes no sense unless assertors are reliable, so
that it is not part of the practice to assert in the face of indeterminism. Or it
might be that assertors should warrant what they assert, and they cannot do
that when making a risky prediction. Or it might be that it sounds funny to
the ear to say "The coin might or might not land heads, and I assert that it
will."
All of these suggestions, and their cousins, seem at first glance reasonable, and
one could doubtless speak at length on one side and the other, citing details of
our practice of assertion and (for the funny-to-the-ear argument) conversation.
We do not, however, think the discussion would advance our investigation, since
the concepts involved are too soft to help us with understanding a hard concept
of indeterminism.
The second and third ways of avoiding the assertion problem have something
in common: Both claim that our language (or any sensible language?) is free
of history-open sentences. Because of the conjunctive nature of our account of
"open" in Definition 6-4, two such ways are possible.
Way of avoidance number two works as follows: It claims that our language
is entirely free of history-dependent sentences, whether stand-alone or not, even
given indeterminism. The most common form that this claim takes is the
Peircean view that future-tensed statements are intrinsically closed by indepen-
dence. Put in our terms, it is the claim that Will:A really means Sett: Will:A.
If that were true, one would have the advantage of being able to drop the his-
tory parameter altogether, since the Peircean future tense can be understood
without it:
6-9 DEFINITION. (Peircean future tense) m, mc, a, m \= Will:A iff for every
history, h, through m, there is a moment, m1 that lies on h and is later than
TO such that m, mc, a, m1 \= Will:A.
(There is quantification over histories, but there is no parameter for histo-
ries.) The trouble with the Peircean account of " Will:" is that it makes no
sense of someone who purports to assert that the coin will land heads even
though it might not, that is, who sincerely asserts both that Will:A and that
Poss:~Will:A. See the discussion of "Antactualism" in Burgess 1978 for an
elaboration of this view. A variation of the Peircean position is offered in
McArthur 1974, according to which there are in general two things that could
be meant by a use of Will:A: either the Peircean Sett:Will:A or Poss: Will:A.
According to this view, which of these two things is meant depends upon the
speaker's intentions. We believe this does not help.
160 Foundations of indeterminism
One of the most compelling arguments against the claim that no sentence in
our language "really" is history-dependent comes from the analogy with betting.
A strict Peircean would have to say that
If one has objective leanings at all, it is easy to feel a need to treat both conjuncts
with equal objectivity. It seems good to try to combine objective indeterminism
with an objective actual future. One is thereby tempted to continue to represent
objective indeterminism by postulating that our world (up to an idealization)
is treelike, but to hold in addition that there is a distinguished history, the
Thin Red Line, which we abbreviate as TRL. One may posit a TRL without
shifting from an objective to a subjective construal of indeterminism (a contrast
discussed in 6A.3), and we must understand the TRL proposal in this objective
way. The proposal succeeds in avoiding the assertion problem by postulating an
"actual history" in addition to the "moment provided by the context of use."
Parametrically speaking, one has a point something like (m, TRL, mc, a, m),
where we specifically note (i) that the TRL parameter is immobile, and (M)
that the mobile history parameter is omitted.
In the semantic theory of branching with a TRL, the future tense, Will:A,
moves you forward along TRL, the past tense, Was:A, moves you backward
along it, and the temporal connective, At-instt:A, moves you to the moment
lying at the intersection of the TRL and the instant denoted by t. For instance,
for the future tense we have
The semantics of "actually" is likewise bound to the TRL itself. Talk of pos-
sibility or necessity or inevitability is explained without reference to a history
parameter by means of quantifications that involve moments that lie off the
TRL. For this purpose we introduce two new connectives, Settled-will:A and
Possibly-wilLA that take over the jobs formerly performed by Sett: Will:A and
Poss: Will:A.
6-11 DEFINITION. (Possibly-wilLA and Settled-will:A)
m, TRL, mc, a, m \= Possibly-wilL: A iff for some m1 such that m < m1
(whether or not m1 lies on TRL), m, TRL, mc, a, m1 \= A.
m, TRL, mc, a, m \= Settled-will:A iff in every history, h, through m,
there is a moment, m1, later than m such that m, TRL, mc, a, m1 \= A.
The definiens of Settled-will: mimics the definiens of the Peircean account of
Will: given in Definition 6-9. (We might also have given that same definition to
the connective, Inevitably:, of 8F.6.) Unvarnished Will: sticks with the TRL.
The assertion problem is thereby avoided.
Very likely there are endless possible versions of TRL theory; we shall consider
only a few simple ones. In what follows we shall reserve plain "TRL" for generic
use, which will permit us to distinguish the particular versions by subscripting
the TRL acronym. In each case we try to make clear both our misgivings about
the "logical" or linguistic explanations offered by the version, and our doubts
about the extra-linguistic commitments of the version.
162 Foundations of indeterminism
ence.) If we cannot find anything in our world to ground the idea of a uniquely
"actual" history, perhaps we should go outside our world, say to a Leibnizian
God who might have some basis for such a choice; and indeed it might seem as
if theological concerns might well direct one's feelings toward the TRL. That
would, however, be a mistake: TRL theory presupposes mdeterminism, and
should never be confused with determinism. According to Prior, for example, it
is determinism itself that Augustine, Luther, Calvin, Pascal, Barth, and Brun-
ner all defended "in the name of religion." See Brauner, Hasle, and 0hrstr0m
2000, who also point out that a theologically inspired (and later abandoned)
determinism is what first led to Prior's interest in tense logic. In any event, our
own view is that if a theologian wrestles with what is required by indeterminism,
he or she need not and should not invoke a TRL.
The TRL abs theory also has troubles with actuality. (Thomason 1984, p.
145 and p. 160, makes remarks that the argument of this paragraph may be
seen as elucidating.) As Lewis has argued (Lewis 1970), this world's being the
actual world does not favor it over any others, but is just a reflection of the fact
that this is the world at which we are conversing. To suppose that there is one
from among the histories in Our World that is the absolutely actual history is
rather like purporting to stand outside Lewis's realm of concrete possibilia and
pointing to the one that is actual. But this is wrong in both cases.
Additional dismay accrues from asking how we could know whether we are
on TRL abs. How could we find out? Perhaps by seeing whether or not we
eventually find out "what will happen next Tuesday"? That seems to make a
certain sense, since it seems that, according to the theory, we might not find
out what will happen next Tuesday. What will happen will of course happen,
but does that recipe include a guarantee that TRL abs is our future?
or not. Suppose, first, that we postulate (15). Then let m1 admit two incom-
parable future possibilities, m2 and m3. First use (15) to calculate that
This postulate is weak enough to avoid the trouble caused by (15), but strong
enough to preserve (i). (17) is furthermore conceptually natural for a branching
+ TRLfcn theory, so natural that it seems best to think of it as an essential part
of that theory.15 Figure 6.2 gives a sense of what the theory calls for. In
this picture, TRLfcn(m) is indicated by a "red" line beginning at the various
moments, m. Note that each moment belongs to exactly one "red" line, and that
it is only after a "break" that a new TRLfcn(m) can be started. This means
that Will: Will: is bound to behave properly (given "standard" properties of
each history). Our "logical" complaint about branching with TRLfcn therefore
reduces to the loss of ( i i ) , which still fails even in the presence of (17).
To see why one should regret the loss of this implication, consider Figure
6.3.The picture indicates that the coin is flipped at mo, where heads will be the
outcome (at m1) but tails might be the outcome (at m2) Now picture Jack at
the moment of use, m2, where the coin landed tails at 2:00 P.M. It would seem
that in order to speak truly at m2, Jack would be obliged to say
The coin has landed tails, but this is not what was going to happen at
1:00 P.M. At 1:00 P.M. the coin was going to land heads. It's just that it
didn't.
15
The point is clearly made and defended in Brauner, 0hrstr0m, and Hasle 1998. We
are also reminded by these workers that (17) was explicitly introduced by Thomason and
Gupta 1980, and used in another form in McKim and Davis 1976.
6. Indeterminism and the Thin Red Line 167
In symbols:
This is an odd thing to be forced to say. As Diodorus would have it, if the coin
lands heads, then the coin's landing heads is what was to be.16
McKim and Davis 1976 sketches another variation on the TRLfcn idea ac-
cording to which each moment provides a future, rather than a history, where
(putting the idea in our own terms) / is a "future history" in the sense of Def.
6 rather than a entire history, including a past. An analog of the problem that
we posed for branching with TRLfcn arises for the McKim-Davis view.
To make clear where we are so far, we have seen that shifting from branching
with absolute TRL to branching with context-dependent TRL solves the prob-
lem of evaluating (stand-alone) predictions off the TRL; and we have seen that
the additional shift to branching with TRLfcn solves the problem of evaluating
embedded future-tensed sentences. We have also seen that the shift does so in
a fashion that, under reasonable assumptions, saves much of linear tense logic,
while failing to save all of it. That A implies Was: Will: A is left in the dust, as
is that At-insti.oop M -'At-mst2 oo p.M.:A is equivalent to At-inst2 oopM :A. One
might also point out that although on this theory one does not have that Will:A
implies Settled-will:A (that something will happen does not imply that it is in-
evitable that it will happen), nevertheless one has that Will:A implies Settled-
will: Was: Will: A. That something will happen does indeed imply that it is
inevitable that it will be true that it was going to happen. (Please forgive our
contorted English.)
We do not wish to rest our case, however, entirely or even chiefly on logical or
linguistic oddity. For one thing, the history of contemporary semantics testifies
that someone can always cook up a definition of "point of evaluation" and some
semantic clauses that will legitimate any desired technical result. We want our
semantic investigation to be not like that. We are trying to stick to structure
16
It is simple tense logic, not "fatalism," that A implies Was: Will: A, given ordinary linear
properties of histories. Equally a matter of simple tense logic is the fact that At-inst2.oop.M :A
implies At-inst1.ooP M : At-inst 2-ooP M: A (added time references are vacuous; see 8F.5).
168 Foundations of indeterminism
parameters and context parameters that, in our opinion, are firmly rooted in
the (idealized) nature of our world. We are trying to follow our own policy of
not making things up as we go along.
The objective reason that we do not agree that there is a "real" history
determined by a moment is this: Our moments represent what is settled, what
is a definite matter of fact, what is determinate; and the future course of history
is still unsettled, is not yet a matter of fact, is indeterminateif indeterminism
be true. One ought not be taken in by a definite description such as "the future"
or "what will happen" (as in (13) on 160). Given indeterminism, even when all
parameters are initialized to context, the referent of these definite descriptions
depends on what is going to happen. Que sera, sera is like "Whoever is x1 is
x1that's certain!"
Our objective reservation to TRLfcn theory is that TRLfcn (and indeed each
form of TRL theory) involves commitments to physical facts (for example, the
coin will land heads) that do not supervene upon any physical, chemical, bio-
logical, or psychological states of affairs. The fact, if it is one, that at a given
indeterministic moment m there is some history such that it is the one that will
occur is not a state of affairs that supervenes upon what is true of particles,
tissues, or organisms that exist at m. Those of us who do not postulate a Thin
Red Line have no need of such a mysterious realm of physical fact. (We hope
you join us in regarding as spurious a reassurance having the form, "but it's
only a logical fact." That's bad logic. See also the argument of 8D.l.)
The TRLfcn approach also has troubles with actuality. Thinking now about
Our World, suppose we imaginatively locate ourselves at a certain moment. For
a history to be actual would be for it to be the history to which the moment
we inhabit belongs. It is not, however, in general the case that the expression
"the history to which the moment we inhabit belongs" secures a referent, since
uniqueness fails in the face of indeterminism. One does on the other hand
always succeed in referring with the expression "the set of histories to which
the moment we inhabit belongs," for which an alternative description might be,
"the actual situation."
Put another way, suppose you agree, as twice before, that "the actual world"
does not, in Lewis's framework for modality, privilege any one world over any
other. Then we hope you will agree that from among the Lewisian worlds that
are exactly alike up to a given time, no one of them has a firmer claim to
actuality than any of the others. It would then be most natural as well to say
that at a given indeterministic moment m of Our World, there is no privileged
actual history or future from among those on which m lies.
The worry about knowing whether or not one is on the TRL might take the
following form: A person might not know whether what is happening is what
was going to happen (while still being sure that what is happening will have
happened). That's a bit like Jane thinking "I don't know whether or not I am
east of some point west of here (whereas I am sure that I'm west of some point
east of here)."
6. Indeterminism and the Thin Red Line 169
6D.4 Epicycles
Permit us another epicycle or two before we engage in our positive argument.
Suppose someone adds the absolute TRLabs not as an immobile part of structure,
but as a mobile parameter. Either this parameter is used in exactly the same
way that we use the history-of-evaluation parameter, or it is not. If so, we
think the addition represents a rhetorically misleading change of philosophical
terminology since, on the open future view, there is nothing "real" or "true"
about the current value of the mobile history-of-evaluation parameter. And if
there is a difference in use, in spite of our pessimism, we must necessarily reserve
comment until the difference is made clear.
Suppose instead one adds the TRLfcn function not as an immobile part of
the structure authorized by the nature of our world, but as a mobile parameter
shifted by some non-truth-functional connective. Will it be initialized by context
or not? If not, there would be no TRLfcn determined by the context of use.
In this case, it would seem that since this new parameter would be mobile
and uninitialized, the assertion problem would return in exactly the same form
as before. Contrariwise, if the mobile TRLfcn is initialized by context, then
the TRLf c n must, we suppose, also exist as a context parameter. This causes
no really new problems; it just deepens old ones. Recall that with TRLC we
deemed it a mystery just how a little speech act could determine the exact
course of world history from now on, forever. With TRLfcn the situation is even
heavier: TRLc-fcn, "the TRL function of the context of use," even determines
many counterfactual such courses, each course beginning with a moment that
is perhaps far off in some causally distant region of our world, hidden among
those possible events that might have happened had things gone otherwise in
the past. The basis in reality of these counterfactually actual historical paths
seems an additional mystery worth pondering.
A special case is the proposal of Brauner, Hasle, and 0hrstr0m 1998 to add
a TRLf c n parameter, not in order to help with the plain future tense, which
they treat in the Prior-Thomason way, but instead as a basis for defining for
each moment a set of TRLfcn functions that determines what is "immediately
possible" at that moment (compare Def. 4 for our concept of immediate possi-
bility). The proposal is too new for us to jump in with an opinion, but naturally
questions concerning the objective meaning of the TRLfcn arise. We mean our
questions, however, to be real questions, not rhetorical. We are entitled to ask
with regard to all the TRL theories what their added parameters would mean,
and how they would help us to understand (or perhaps hinder us from under-
standing) the objective features of, for example, agency, strategies, promising,
assertion, betting, causality, and so on. Perhaps these additions could help on
all fronts; if so, such helpfulness might support a claim for the usefulness of this
new version. That is certainly our claim for the history-of-evaluation parameter.
In pointing out these features of the TRL, we do not take ourselves to have
proven that belief in the Thin Red Line is inconsistent or incoherent or even
false. The chief thrust of our thinking is not so much that adding the TRL
causes trouble, but that, as we argue in the next section, its addition does
170 Foundations of indeterminism
no serious work (beyond causing confusion). To help make our position clear,
consider again a version of a TRL theory on which a point (m, TRL, mc, a,
m/h) contains both a history-of-evaluation parameter and a TRL abs or TRLC or
T R L f c n . Surely such a combination is mathematically consistent. Consistency,
however, is not enough. You may take the arguments to follow in support of
the prediction that when history-of-evaluation and TRL are combined, you will
find that the TRL is mere idle filigree. Our conclusion in the following section
is that the mobile history-of-evaluation parameter is useful for making sense of
assertion (and much else), but that the TRL is not. Insisting on adding a TRL
to branching time is exactly like insisting on adding a privileged assignment of
values to the semantics of quantifiers. To do so would be technically consistent,
but pointless.
The trouble with branching exactly is that it conflicts with our or-
dinary presupposition that we have a single future. If two futures
are equally mine, one with a sea fight tomorrow and one without, it
is nonsense to wonder which way it will beit will be both ways
and yet I do wonder.17 The theory of branching suits those who
think this wondering is nonsense. Or those who think the wonder-
ing makes sense only if reconstrued: You have leave to wonder about
the sea fight, provided that really you wonder not about what tomor-
row will bring but about what today predetermines. (Lewis 1986,
pp. 207-208; we discuss this passage again in 7B.2, where we also
quote its continuation.)
17
"It will be both ways" does not apply to branching time. That doctrine always insists
that there are alternative incompatible ways the future might be, which fit together not by
"both ways happening," but precisely by having a branch point at which they both were
possibilities. At the branch point, the future can be either way, but not both ways.
6. Indeterminism and the Thin Red Line 171
Consider Lewis's suggestion that on the open future view the only way to make
sense of a person purporting to wonder what the future will bring is to construe
her as wondering what the present predetermines. If this suggestion is right,
then it is a strike against the open future view, since it would appear to be a
mistake to identify wondering whether there will be a sea battle tomorrow with
wondering whether it is settled true that there is to be a sea battle tomorrow.
(We are also indebted to C. Hitchcock for formulating an objection along these
lines.)
We shall argue that it makes sense to wonder about what history has not yet
decided so long as history will decide the matter. We shall also argue that it
makes sense to assert A when A 's truth value is not settled at the moment of
use; the idea is that assertion is an act that has consequences for the speaker
no matter how things turn out.
Assertion can be treated in either indirect discourse, "a asserts that the coin
will land heads," or in direct discourse, "a asserts 'the coin will land heads'."
Both are valuable approaches, and each has its own awkwardnesses and limita-
tions. We won't try to stick to one or the other when conveying rough ideas,
but when we are fine-tuning, we shall take L to have the direct discourse version
of assertion.
To this end, we endow the grammar of L with a quote-function such that
when A is a sentence of L, 'A' is a term of L that denotes A.18
In order to avoid having to worry about the consequent threat of circularity
or paradox, however, we severely limit the resources that L can bring to bear in
this regard: A must must involve neither assertion concepts, nor any remotely
semantic concepts. Furthermore, with the same end in view, when we introduce
"a asserts " into L, we insist that the blank be filled only with a quote-name:
a asserts 'A,'
'A'" as embeddable, not merely stand-alone, so that we can make sense out
of for example the past-tensed "Was:(a asserts '.A')." It is therefore necessary
to observe that in (18), the "moment of assertion" is to be identified as the
moment of evaluation, m, rather than the moment of use, mc. That is, as in
any case of embedding, when the assertion-statement is considered as embedded,
one must distinguish the moment of use of the (presumably) larger sentence
in which the assertion-sentence is embedded from the moment at which the
embedded assertion-statement is evaluated. There are additional complications
when we turn recursively to evaluating A itself, (i) Direct discourse requires
that we consider A as stand-alone, letting its moment-of-evaluation parameter
be initialized by context to be the very moment of its use as a vehicle of assertion.
(ii) Direct discourse requires that we throw away the moment of use, mc, of the
larger sentence in which "a asserts 'A'" is embedded. Instead, direct discourse
requires that we identify the moment of use (and also, by ( i ) , the moment of
evaluation) of the asserted sentence, A, with the moment, TO, of evaluation of
the (considered-as-embedded) assertion sentence, "a asserts 'A'." (m) Direct
discourse requires that we do not consider assertion of assignment-dependent
sentences; accordingly, except in certain later dialectic passages, we require
that A in (18) contain no free variables. We shall therefore be interested in
without knowing what value has been assigned to x1. An analogous line of
thought would apply to attempts to evaluate other attitude and performative
verbs, such as "believe," "wonder" and "predict." The difficulty is that assertion
(etc.), unlike mere utterance, relies on the meaning or semantic content of the
6. Indeterminism and the Thin Red Line 173
the sense that assertion is an act that has implications for the speaker and others
no matter how things eventuate. That is why, in coming to an understanding
of assertion, every history through the moment of assertion is treated equally,
and why it is a step backward to bring in a Thin Red Line.
Semantic account 6-12 is amenable to revision in two different directions. The
first concerns the causal structure of assertion. A far better version would avoid
claiming that vindication/impugnment occurs at the very moment of assertion.
The revision would spell out that vindication/impugnment occurs only later, at
moments m1 when it becomes settled that A was true (false) when evaluated
(not at m1 but) at the moment, m, of assertion. (And if on h1 it never becomes
settled one way or the other, then a is neither vindicated nor impugned any-
where on h1.) On the revised version, it would be clear that if one is making
an assertion concerning a matter not yet determined, then one has to wait for
vindication or impugnment. This better formulation would involve double time
references, as in 5C.2.1 and 8F.5.
Either formulation puts us in a position to explain the difference indicated in
Assertability thesis 6-7 between assignment-open sentences, which have no truth
value and are not assertable, and history-open sentences, which also have no
truth value, but are assertable. In relation to assertion, the content of " Will:(the
coin lands heads)" differs in principle, and not just technically, from that of "x1
is brindle." For the content of "Will: (the coin lands heads)," unlike that of ll x1
is brindle," is the sort of thing that can be borne out or not, depending upon
what comes to pass. This is particularly evident for the double-time-reference
refinement: Time will tell whether we arrive at a moment at which the truth
value (at the moment of assertion) of "Will:(i\\e coin lands heads)" becomes
settledand whether we do or not determines whether the person who asserted
"Will:(the coin lands heads)" is vindicated or impugned. On the other hand,
finding an object that is brindle gives us no guidance whether or when one who
purports to assert "x1 is brindle" is vindicated or impugned.
In another direction, Semantic account 6-12 can be fleshed out by adding
further concepts and conditions while still eschewing a Thin Red Line. In this
way the vindication/impugnment skeletons help not only with assertion, but
also with other speech acts that face the open future. One notices, for instance,
that the right side of Semantic account 6-12 applies equally to assertions and
to conjectures, even though these are evidently different: An assertion is liable
to challenge by means of such words as "Why do you think so?" A question
such as this is, however, inappropriate for a mere conjecture. We may envi-
sion an improvement that marks this difference by building in further norms
for assertions, conjectures, and so on, that go beyond liability to impugnment
and vindication, and that attribute normative requirements and entitlements at
future moments other than those at which vindication and impugnment come
into play. For instance, one could note that one who asserts A, in addition to
being vindicated or impugned at certain moments, is also obliged to respond at
any future moment at which there is an appropriate challenge having the form
"How do you justify that claim?" There is, in contrast, no such requirement on
one who conjectures A. (See Green 1999 for further discussion.)
176 Foundations of indeterminism
21
We neglect the possibility that a denial of the presupposition of the question counts as
an answer. For simplicity we also neglect consideration of questions whose answerability itself
depends on future happenings; for example, "Will the sea battle be followed by a full-scale
war?" may turn out to be badly posed (if there turns out to be no sea battle), and is therefore
risky.
7
177
178 Foundations of indeterminism
of the beach by the hurricane, which happened far away from (and in causal
independence of) Alfredo's finishing his pasta primavera, which was followed
by a coin flip that by chance came up heads. It is in this spirit that we pro-
pose branching time with agents and choices as a high-level, broadly empirical
quasi-geometrical theory of our world that counts equally as proto-physical and
proto-humanist. Like geometry, it does not pretend to be that famous "theory
of everything." It concerns above all the structural aspects of how the doings
of agents fit into the indeterministic causal structure of our world. We pos-
tulate an underlying temporal-modal-agent-choice structure (Tree, \<, Instant,
Agent, Choice, Domain) subject to certain constraints. You will find a list of all
u
BT + I + AC postulates," with BT + I + AC standing for "agents and choices
in branching time with instants," in 3. We go over these postulates one at a
time, concentrating on foundational questions.
Greatest lower bounds and least upper bounds of sets of moments are as
usual in the theory of partial orders.
M is the causal past of a future history, /, iff M is the set of all proper
lower bounds of /.
We often say just past because given no backward branching, a causal past
is the same as a historical past.
1
We use "p" for past histories and "f" for future histories, but also for other purposes; we
believe, however, that no ambiguities arise in this book.
7. Agents and choices in branching time with instants 183
The postulate of no backward branching tells us that each future history has
exactly one past history: If (p1, f) and (p2, f) are historical cuts, then p1 =
P2. Where / is a future history, this uniqueness entitles us to introduce the
phrase "the past history of f." Since no backward branching implies that the
past history of / is the set of all moments preceding /, by strict analogy with
special relativity, we may also call the past history of / the causal past of /, or
just its (plain) past.
As a special case of no backward branching, our causal past is a history
in which everything has a place in terms of earlier/later; our past is not an
assemblage of incompatible possibilities. If two possible events (i.e., members
of Tree) each lie in our past, then one of them was causally earlier, and lies in
the past of the other, giving rise to an actual linear causal sequence, earlier-
later-now. This is not merely a matter of temporal dating; it concerns causal
linkage. This discussion does, however, falsely but conveniently presuppose that
causal linkage is between entire spatial "slices" instead of between small local
events. As previously noted, this defect is remedied in Belnap 1992, but not in
this book.
Although a future history has but one past history, the converse is by no means
true. If determinism is not permanently true, there will be past histories p to
which more than one future history can be appended: One can have historical
cuts (p , f i ) and (p, f2) with f1 /= fa. There is therefore no "rigid" sense to the
expression "the future history of p." We may, however, reasonably speak of a
unique "future of possibilities." Each future of possibilities will look something
like a tree, and will be a subtree of Tree.
When at a given moment Lee-Hamilton says "the past is stone, and stands
forever fast," his use of the phrase "the past" safely refers to a past that is
uniquely determined by the (idealized) context of utterance. 3 Furthermore,
this past is a portion of each history of which his utterance is a part: No matter
2
In earlier publications we sometimes used "future" as a short form of "future of possi-
bilities." That was an expository error; in delicate discussions of indeterminism, it is better
to avoid this contraction because of its confusing conflict with ordinary English usage. One
should be resolute in speaking of either "future histories," or "future of possibilities," never
of just "futures."
3
It is a confusion to infer from this that "it was true that Q" implies "it is settled that it
was true that Q"; instead, as we note at the very end of this section, one has only that "it
was settled true that Q" implies "it is settled that it was settled true that Q."
184 Foundations of indeterminism
the future history, that past stands forever fast. Given no backward branching,
talk of so acting as to "influence" the past is therefore just talk.
When, however, at a certain moment Tennyson says "I dipt into the future,
far as human eye could see, / Saw the Vision of the world, and all the wonder
that would be," we must be careful, especially when we learn that the Vision
involves a sea battle. There is no philosophical problem if we interpret the poet
as denoting his future of possibilities, the tree that fans out from the moment
of his utterance in intricate arborescent profusiona proper source of wonder
indeed. Nor is there a problem if the poet is using "the future" as "the future
history," but with its "nonrigid" denotation relative to a poetically visionary
historyas long as we do not forget this relativization and take proper care to
disambiguate Tennyson's expression accordingly, just as we would in the case
of an occurrence of "the river" in The Lady of Shalott. (This simply adapts
to singular terms the sentential semantic insight of Prior-Thomason that we
adopted and first explained in 2A.l, and that we make more explicit in 8E.)
If, however, by the phrase "the future" the poet intends to denote a unique future
history of which his moment is a part, then his intention cannot be carried out,
for (unless determinism be true from that moment as far as human eye can see)
there is no such unique future history, as from time to time we repeat.
We believe (if that is the right word) in forward branching, and in the impossi-
bility of backward branching. Sometimes one hears a philosopher or a physicist
maunder on about distinct pasts that coalesce in a present moment, and doubt-
less it is good that our conceptual limits be tested. We confess, however, that we
ourselves cannot follow these fancies. That we face alternative future histories
seems to us right; that we are faced away from alternative pasts seems to us
wrong. That starting with the concrete event that occurred yesterday morning
there were incompatible possible events each of which might have transpired
seems to us right; that more than one of these incompatible possible streams of
events might have finished up in this very concrete situation seems to us wrong.
In common with antisymmetry, no backward branching makes sense only for
objective, concrete events. First, we advance no theory at all about what is
possible (not objectively but) "for all one knows." A given concrete situation
could obviously have been preceded by any of various inconsistent predecessors,
"for all one knows." It is precisely to preclude this epistemic or doxastic use of
"possible" that we so tiresomely repeat that our present concern is with "objec-
tive" possibilities. Second, no backward branching fails to apply to "states" or
other repeatable carriers of partial information. There is no doubt whatsoever
that a present "state" may be accessible from either of two earlier incompatible
states. There is no doubt about this because there are so very many senses to
the word "state." Surely there are physical "systems" with a favored family of
"states" that branch only forward and not backward, others that branch only
backward and not forward, others that are doubly deterministic in terms of their
favored "states," and still others with more exotic structural properties or with
no interesting properties at all. After all, everything happens. None of this,
however, is relevant to our postulate of no backward branching. To discuss any
of it is to change the topic.
7. Agents and choices in branching time with instants 185
that makes sense; but we ourselves have absolutely no idea (in the absence of
doublethink) of how to develop a theory of promises in such a setting.
One pressure for backward branching comes from certain interpretations of
certain physical theories. According to these interpretations of these theories,
the world looks just the same upside down. We of course have no special exper-
tise in physics, and so turn to the quantum-field theorist Haag 1990.
I want to suggest here thatonce one accepts indeterminismthere
is no reason against including irreversibility as part of the fundamen-
tal laws of nature. ... It should be stressed that this picture does
not touch CTP-invariance or detailed balancing. ... the term "time
reversal" should be replaced by "motion reversal." (p. 247)
Evidently even physicists disagree about these matters; it is, however, surely
respectable to take Haag as our governing authority. If we do so, backward
branching receives no aid and comfort from any physical theory.
The tense-logical principle that seems to us most closely associated with no
backward branching is this: "If something was settled true, then it is (now)
settled that it was true." This principlewhich is not to be confused with its
invalid sound-alike, "If something was true then it is (now) settled that it was
true"is the one whose loss would leave us with a feeling of having lost our
grip. Settled features of past moments stay settled. In particular, if we look
back at earlier choice points, it is and will always remain a settled matter which
choice was made at that earlier point.
postulate, above all, by means of which the theory insists that we not admit
histories that are idle creatures of the imagination, or histories that could be
"defined" by piecing together some arbitrary array of logical or conceptual or
scientific possibilities. Instead, every possible history, h, has a definite causal
relation to the very moment in which we converse, since h must share with it
a common past. Historical connection is the postulate that endows the theory
with a sense of robust reality. In other words, historical connection makes Tree
a single world, Our World, instead of a mere collection of "worlds." In this
sense it is historical connection that ensures that agents are actual and that our
choices are real.
Historical connection puts all moments of Tree "in suitable external relations,"
as Lewis 1986 says (p. 208). It gives content to exactly the sort of real possibility
that is pertinent to an understanding of stit. Just to make things clear by an
example, we are disallowing that it is or was really possible that there should be
blue swans unless there is some definite moment in our past that has a moment
in its future of possibilities at which there are blue swans. Naturally, as armchair
philosophers of indeterminism, we do not claim special insight into what is really
possible; that is a matter for common sense or science or metaphysics. The point
of the example is only to express our doubt that it is easy to be sure that it
is or was really possible that there should be blue swans. Of course something
terminological is going on here: We are using "really possible" as what is or
was determined as possible in the world of our context of utterance, and thus
in a sense much narrower than that sought by, for example, Lewis 1986 through
the idea of recombination. But there is also something nonterminological: We
think that the Humean picture of enormous recombinational possibilities for
the immediate future (e.g., blue swans on our desk one nanosecond from now
Lewis 1986, p. 91, says that "anything can follow anything") is not relevant to
what can be seen to, and that instead what counts is only the currentmuch
narrowerset of real possibilities.
When the intersection h1 n h2 of two (distinct) histories h1 and h2 is not only
nonempty, as promised by historical connection, but has a least upper bound,
m0, we say that h1 and h2 split at mo (Def. 7). We write h1 Lmo h2 (Def.
4). And if every two (distinct) histories split at some moment, we say that
the semi-lattice condition is satisfied (Def. 7), since in context this holds iff
every two moments have a greatest lower bound. Is it true? Do each two two
histories split at a moment? In the branching space-time theory of Belnap 1992
we postulate the principle of "prior choice," which says that if a point-event,
e\, belongs to one history, h1, and not to another history, h2, then there is a
particular point-event, e0, in the past of e1 such that h1 and h2 split at that
point. In the context of branching space-time prior choice is, as far as we can
see, a deep causal principle, and in fact in that context, we do not know how to
carry on without it. The same words in the context of branching time express
the semi-lattice condition. We are therefore certainly tempted to enter that
postulate as a strengthening of historical connection, especially because there
is no reason of which we know to suppose that the semi-lattice condition is
false. On the other hand, by careful formulations that do not rely on that
7. Agents and choices in branching time with instants 189
condition, one can come to see that it does not make any difference to the
theory of agents making choices in branching time. The fundamental reason for
this seems to be that in our theory of choice, we postulate that regardless of
the semi-lattice condition, when an agent makes a choice, there is always a last
moment of indetermination, which is therefore the greatest lower bound of two
moments belonging to different choices. In other words, if the fact that we are
in one history rather than another is to be explained by choice, then those two
histories, anyhow, split at a moment.
All this abstruse talk is to explain why we waver when it comes to postu-
lating the semi-lattice condition. In the end we decide somewhat arbitrarily to
do without it. Wavering to one side, however, even in the rudimentary con-
text of branching time, one ought to see historical connection, especially when
strengthened to the semi-lattice condition, as a powerful causal principle that,
like prior choice, says that a real cause for something being one way rather than
another always lies in the past. If histories were disconnected, one could not
say that without bringing in soft ideas such as "similarity."
We also refrain from postulating that all the moments in Tree have a common
lower bound, which would imply a kind of nonrelativistic Big Bang.
The idea of a "transition" as an ordered pair of events now falls into place.
H(m) is the set of histories in which m lies (or the set of histories "passing
through" m): h E H(m) iff m E h.
H [ M] = {h: M C h}, so that H[M] is the set of histories entirely containing
M. For suitable M, this is the set of histories in which M passes away.
H<M> = {h: (Mnh) = O}, so that H<M> is the set of histories that pass
through at least one member of the set of moments, M. For suitable M,
this is the set of histories in which M comes to be.
7. Agents and choices in branching time with instants 193
Consistency. Without belaboring the point, we note that the interplay be-
tween propositions and events in branching time generates a firmly based family
of consistency/inconsistency concepts. We may take from possible-worlds theory
the idea that when propositions are represented as sets of histories, consistency
between them is definable as having some history in common, a history in which
both propositions are true. We add that it is helpful to define the consistency
of two initial events I1 and I2 by the consistency/inconsistency of the proposi-
tions H[I1] and -H[I 2 ], so that the question is whether or not there is a history
in which both preparations are (not just started but) completed. Dually, two
outcome events O1 and O2 are consistent/inconsistent iff the propositions H<O1>
and H<O2> are consistent/inconsistent, so that the question is whether or not
there is a history in which both outcomes begin to be. Even more delicately,
194 Foundations of indeterminism
one may ask whether a certain initial is consistent with a certain outcome by
way of well-chosen propositions. Finally, one has the right definition of what
makes a transition "contingent."
CONTINGENT TRANSITION. (Definition. Reference: Def. 8) <I, O> is a con-
tingent transition iff <I, O> is a transition, and if some history is dropped in
passing from the completion of / to the beginning of O: H[I] -H<O> = O.
These remarks are intended to indicate with almost excessive brevity that
branching time permits and indeed suggests a rigorous and modestly enlighten-
ing theory of the interrelations of (possible) concrete events and propositions.
We mention that some additional ideas on initials and outcomes are offered in
Szabo and Belnap 1996 and in Belnap 1996c, and in the unpublished set of notes
Belnap 1995.
iii. Order preservation. Instants never distort historical order: Given two
instants i1 and i2 and two histories h and h', if the moment at which i1
intersects h precedes, or is the same as, or comes after the moment at
which i2 intersects h, then the same relation holds between the moment
at which i1 intersects h' and the moment at which i2 intersects h'.
We next offer some convenient definitions and simple facts, after which we
comment on the postulates.
i|>m = {m0: m < mo & mo E i}. We say that i|> m is the horizon from
moment m at instant i.
Where i1 and i2 are instants, we may induce a linear time order (not a
causal order!) by defining i1 < i2 iff m1 < m2 for some moment m1 in i1
and some moment m2 in i2. Instants can also be temporally (not causally)
compared with moments, m: i1 < m iff m1 < m for some moment m1 in
i1 and m < i2 iff m < m2 for some moment m2 in i2 .
Postulates Post. 5(ii) and Post. 5(iii) on Instant are very likely too strong
(too oversimplifying); our justification is that agency is already hard to under-
stand, so that it won't hurt to try to see what it comes to in circumstances that
are not altogether realisticas long as we keep track of what we are doing so
that later we can try to move closer to reality. Thus, which most reality-oriented
persons think not so plausible, but which greatly simplifies our picture of time,
all histories are said to have exactly the same temporal structure. It follows that
all histories are isomorphic with each other, and with Instant, which justifies
the ordering on instants defined in Def. 9. On the other hand, no assumption
whatsoever is made about the order type that all histories share with each other
and with Instant. For this reason the present theory of agency is immediately
applicable regardless of whether we picture succession as discrete, dense, con-
tinuous, well-ordered, some mixture of these, or whatever; and regardless of
whether histories are finite or infinite in one direction or the other.
The theory of Instant is not, as we have said, as fundamental as that of <,
and perhaps it is too strong, even pre-relativistically. Certainly the present
assumption that all histories have isomorphic temporal orderings is stronger
than comparable assumptions of Thomason 1970 or Thomason and Gupta 1980,
and probably it should ultimately be weakened. In the meantime, while it is
good to be concerned about oversimplification, the justification of our procedure
is that the assumption can be clarifying when it comes to thinking about certain
aspects of agency. Instant gives us a theory of linear "time" (based on <
between instants) to play off against the theory of branching "time" (based on
< between moments).
very like one of the histories on which we turn right. Perhaps this happens at
some point after the heat death of the sun; why not? But "very like" is not
the same as sharing particular moments (and is not even very like). The gap
between identity/distinctness on the one hand, and similarity/difference on the
other, is unbridgeable. A helpful analogy is to points of space, some of which
are well known as both very like and very distinct. The analogy is particularly
apt since the theory of branching time is, in its general spirit, geometrical.
Every theory has merits and demerits. All Tx W theories share the following
two demerits: They need for their foundation a prior story about (i) times and
(ii) worlds. Both of these stories are likely to be tall.
(i) The notion of a time as an entity independent of events gives pause. It
would seem that in order to make sense out of indicating a particular time,
one would have to explain quite a lot about clocks or cesium atoms or seasonal
social practices, or something. This is something that Tx W theorists need to
do. The problem is perhaps best expressed in worrying over the requirement that
one can make sense of "same time" across different worlds, which is normally
entered as an unexamined and unquestionable presupposition of Tx W theories.
Of course in the theory of instants we rely on a postulate that is similar to
this presupposition. That we introduce this postulate explicitly and separately
after giving the fundamental ideas of branching time is conceptually significant,
since we can develop the main ideas of tense and modality without it. The
Tx W approach hides this independence of tense and modality from the "same
time" problem. Further, our "same time" postulation is explicit rather than
presuppositional, so that it is easy to give it up when it does not help, or when
it does not ring true.
In short, a Tx W theory obviously provides no way to avoid presupposing
instants or times shared across histories. Therefore the extent to which the
concept of agency does not presuppose same-time comparisons across histories
is the extent to which we have a good reason for not using Tx W as a foundation
for the theory of agency.
(ii) The idea of a possible world is the idea of something Very Big and (we
suppose) hard to understand. As possibilities go, there are (we suppose) none
bigger than worlds. It seems somehow a pity to start with something so big when
what we want to understand is Caesar's situation when deciding whether or not
to commence crossing the Rubicon. It would seem better to begin with a theory
about more local incompatible possibilities, such as those available within ten
or fifteen minutes, or available within ten or fifteen seconds, or (best) available
immediately. The argument is that it is easier to credit a foundation built upon
small possibilities, such as "the alternative moves in an actual chess game that
I did not finally make" (Marcus 1985/86), rather than a foundation built upon
something as large as worlds.5
Different from Tx W is the framework employed in Chellas 1969, which one
could call an "h:T> S" theory: Each history, h, is taken as a mapping from
5
Marcus herself allows the point as persuasive, but argues against taking it seriously. It
would be out of place to respond to her argument here.
198 Foundations of indeterminism
same work. There is, however, something about bundled trees that is indeed
of considerable significance: Apart from perspective, bundled trees result from
our trees <Tree, <> in exactly the same way that "Henkin general models" of
second order logic result from standard models of second order logic (Zanardo
1996). Like second order logic, general models take the domain of individuals
as primitive. Quite unlike second order logic, however, general models also take
the range of predicate variables as primitive, which permits them to license the
omission of some of the subsets of the domain of individuals, provided certain
constraints are satisfied. Analogously, bundled trees take the set of branches to
be primitive. This permits bundled trees to license the omission of branches,
provided constraints are satisfied. When one adjusts for ontological perspective,
this is tantamount to yet a fifth representation: One takes both the set Tree of
all moments and the set History of all histories as primitive, in what might be
called a moment-history structure <Tree, < , History). One is then in a position
to license the omission from History of some of the histories (maximal chains)
on (Tree. <>, subject to the requirement that each moment belongs to at least
one history. In this way one is able to replace second-order conditions with first-
order conditions. Since first-order theories are always more technically tractable
than second order theories, professional logicians tend to prefer to study them.
This is by no means a misplaced "preference," since the study of these more
general moment-history structures provides a great deal of illumination in the
way that, for example, much illumination is to be had by studying nonstandard
first-order models of arithmetic, or of second order logic.
As a reading of Thomason 1984, pp. 151-152, suggests, however, there remains
a question about the descriptive adequacy of the more general moment-history
structures when taken as theories intended to be true to the facts of our world
(or language?). In order to evoke the negative thrust of our opinion, we will
label these more general structures as "missing-history structures." An example
will provide the reason that we think missing-history structures are descriptively
inadequate. (After articulating the example we will make some remarks that
examine how our negative opinion looks in the context of Tx W, leaving it to
the reader to draw conclusions about Kamp structures, Ockhamist structures,
and bundled trees.)
Let there be a specially interesting radium atom, a, such that as the seconds
tick by after moment mo, the situation is as follows.
As long as a has not yet decayed, (i) a might decay before the
,..,
next tick, and (u) a might not decay before the next tick.
We don't need metrics in order to describe the situation, but it is essential to the
story that the sequence of ticks has no upper bound. The situation is pictured
in Figure 7.2, a diagram that we borrow from Thomason 1984. With reference
to Figure 7.2, we let
p <-> atom a has not (yet) decayed,
so that
200 Foundations of indeterminism
The "can" of (2) is not just mathematical. This "can" is to be taken in its usual
historical-modal sense involving quantification over histories. In other words:
At mo it is a settled fact that if a no-decay chain of length n will come to pass,
then it is possible (but not guaranteed) that a no-decay chain of length n +1
will come to pass.
The truth value of (2) does not depend on whether or not our world is missing
the history, hw; (2) holds whether or not we count hw as a history. The truth
value of the following, however, depends on precisely that.
At mo, it is inevitable that a will decay after a finite number of
(3)
ticks.
This may be restated in various ways.
In (3) and in all of its restatements, the modal words are to be taken as historical
modalities, equivalent to quantification over histories. So it matters whether the
maximal chain of moments, hw, which is determined by the rightmost sequence
of moments, counts as a history.
On our no-missing-history account, hw has to be a history merely in virtue
of being a maximal chain in the tree: (Tree, <> has the histories that it has,
and if (Tree, <> represents our world with its causal order, there is nothing
more to say. Therefore the no-missing-histories account says that (3) is, in
all its versions, false. Suppose, however, that hw does not count as a history,
as is certainly allowed if a missing-histories structure, <Tree, <, History>, can
describe our world. Then all the versions of (3) are true. If you leave out hw
as a history, then no matter what history, the sequence of no-decay terminates
after a finite number of ticks.
Does this quarrel finish in a draw between the missing-histories and no-
missing-histories representations of our world? We don't think so. It seems
to us plain, following an analogous verdict by Thomason 1984, pp. 151-152
(but contrary to the verdict of 0hrstr0m and Hasle 1995, pp. 268-269) that
anyone who asserts both (2) and (3) has contradicted himself. Surely, we say,
it is a real possibility, not to be ruled out by switching "logic," that the atom
may never decay.
There is also an argument against the legitimacy of the missing-histories rep-
resentation that does not depend on intuitions concerning the validity of hard-
to-understand arguments. We introduced moments as representing concrete
possible events, and < as representing the (indeterministic) causal ordering
among them. This gives us <Tree, <> as rooted in objective realityidealized,
of course. The set of all histories is uniquely determined in terms of Tree and <,
so that we take that set as itself objective rather than made up to suit conversa-
tional context, or language, or the like. Let us now consider a missing-histories
structure, (Tree, <, History>, where one or more histories is missing from His-
tory. How can we see as objective the separation of histories (maximal chains)
into those that belong to History and those that do not? Using "chronicle"
where we say "history," 0hrstr0m and Hasle 1995 argue that we must "assume
that not all linear subsets [of Tree] are possible chronicles." But what objective
property of our world could justify treating some maximal chains as real possi-
bilities and others as not? These questions seem to us to have only implausible
answers; consult the following observations.
Observations. (i) Probabilities don't come into it, just possibilities. One
may wonder, however, if a case could be made for missing histories by forging
a conceptual identification of "impossible" with "zero probability." No, for
standard reasons: It is all too likely that in our world, every endless branch
has zero probability, but there is no sanity in letting all of them go missing.
(ii) It is no good subtracting from a tree all those branches containing only
points that belong to other branches. It is indeed true that there is only one
of these in Figure 7.2; it is, however, all too likely that in our world, every
branch contains only points that belong to other branches, ( i n ) Some persons
might think that Omnipotent God can rig things so that both (2) and (3) are
202 Foundations of indeterminism
true. God could arrange things, for instance, so that on the one hand, as long
as the atom has not decayed, it is guaranteed that it has another chance not
to decay, and nevertheless, God forbids that it should never decay. Or maybe
lawful nature can rig things in this way. We don't understand it. How could
even God, or nature, prohibit that the atom will never decay, given that each
stage of nondecay can be prolonged? You can say it of course, but does it make
sense? We doubt it.
Coming back to an earlier point, each of the bundled tree and the Ock-
hamist structure and the Kamp-structure representations, being equivalent to
the moment-history representation, deserves the same missing-histories com-
plaint, as does Tx W. In the case of Tx W and Kamp structures, however,
it is considerably less obvious that there is an unreasonable "omission" rather
than a reasonable "resistance to addition" (A. Zanardo, correspondence). Let
us spell this out. In, for example, Tx W, one might have the infinite collection
of disjoint worlds {w1, ... w5, ...}, as in Figure 7.3. In this figure we represent
a single moment by means of a collection of points, one for each world, that sit
on the same level. A single point then corresponds not to a moment, but to a
moment-history pair. The heavy dots represent branch points, and the topmost
dots, marked "*", represent the last moment of no-decay in that world. The
Tx W diagram of Figure 7.3, with or without ww, is representationally equiva-
lent to the branching-time diagram of Figure 7.2, respectively with or without
hw The point is that if you stare at just the infinite collection of worlds {W0,
...}, without ww,, you may well not feel that there is a "missing world." You
may instead feel that in passing from {w0, ...} to {w0, ...; ww}, an unexpected
world has been added.
We certainly think that such a feeling is justified by the diagram. From that
perspective, it seems an open question as to whether "limit worlds" such as ww
should or should not be added in all cases. You may share in the feeling that
7. Agents and choices in branching time with instants 203
when History is the set of all histories, (Tree, <, History> is somehow unusual
or special. You may therefore prefer to abandon the language of "missing his-
tories" versus "no missing histories" that we have employed because it suggests
that having all the histories is philosophically normal. You may instead pre-
fer to speak of "moment-history structure" versus "complete moment-history
structure," with its suggestion that having all the histories is special. Are we
then back to seeing the quarrel as finishing in a draw? We don't think so. We
think that the Tx W picture is a mere diagrammatic representation: There
is more in the diagram than there is in our world. In particular, there seems
to us no objective truth to all of those disjoint "worlds" in Figure 7.3 except
to the extent that they jointly represent our one objectively real world with
its concrete events in their indeterministic causal order giving rise to a system
of overlapping, branching histories. We recommend not trusting diagrams like
Figure 7.3 when they are not rooted in objective features of our only world. To
put the matter another way, if what binds all the points on a certain level into
the representation of a single concrete event is not sheer identity, then there is
nothing else objective for it to be. "Matching" is a myth.
One may take the fact that Tx W or Kamp diagrams can mislead as an
additional complaint against them: They conceal the truth about the structure
of our world by means of too many henscratches and too much loose play in
their free-floating History parameter. On the other hand, we certainly recognize
that to the extent that we are arguing from premisses, our argument is circular.
Someone who believes that the TxW diagram of Figure 7.3 gives the right
picture of the "facts" of the decay example of Figure 7.2 will draw quite the
opposite conclusion about both.
That is, not only must the histories share m or p, but they must also share
some properly later moment. Now we can say with absolute clarity what it is
for a concrete situation to be deterministic.
There may be many histories through a deterministic past, p, but if so, they
must become "many" after p is past; there is no branching as p itself comes to
a close if p is deterministic. There is another way of saying this that involves
the notion of immediate possibility.
But a modal realist who thinks in the ordinary way that it makes
sense to wonder what the future will bring, and who distinguishes
this from wondering what is already predetermined, will reject
branching in favour of divergence. In divergence also there are many
futures; that is, there are many later segments of worlds that begin
by duplicating initial segments of our world. But in divergence, only
one of these futures is truly ours. The rest belong not to us but to
our other-worldly counterparts. Our future is the one that is part
of the same world as ourselves. It alone is connected to us by the
relationsthe (strictly or analogically) spatiotemporal relations, or
perhaps natural external relations generallythat unify a world. It
alone is influenced causally by what we do and how we are in the
present. We wonder which one is the future that has a special re-
lation to ourselves. We care about it in a way that we do not care
about all the other-worldly futures. Branching, and the limited over-
lap it requires, are to be rejected as making nonsense of the way we
take ourselves to be related to our futures; and divergence without
overlap is to be preferred. (p. 208)
The first sentence refers to "the future," the one that "is truly ours." Branching
time says that only the future of possibilities is uniquely determined by the
moment of utterance, so that "the future" either refers to this, or else is not
history-independent (is open in the history parameter). Branching time says
that if indeterminism be true, then there is no more sense to "the actual future"
than there is to "the actual distant instantaneous present" or to "the odd prime
number."
But what about the future that "is part of the same world as ourselves"?
Assuming indeterminism, there is the following dilemma.
If we read "world" as "history," then it makes no sense to speak of "the
world of which we are part." There are many such possible histories to
which this utterance-event equally belongs. All of them are "connected
to us by the ... spatiotemporal relations ... that unify a world," for there
is, in our opinion, no more fundamental "natural external relation" than
the causal ordering itself. It is to be borne in mind that even wholly
incompatible moments are mediately connected by <; that is exactly the
import of historical connection. It is why (or how) Tree constitutes a
single world, our world.
If we read "world" as Tree in its entirety, then although it would make
sense to speak of "the world of which we are part," it would not make
sense to speak of "the future" history that is part of that world. On the
other hand, invoking a familiar contrast, it would indeed make sense to
ponder "the future of possibilities."
Lewis then gives what are in effect three arguments that, contrary to branch-
ing time, "this very event" picks out a unique future history.
7. Agents and choices in branching time with instants 209
The first argument is that we can define "the future history" as the one
that "alone is influenced causally by what we do and how we are in the
present." This sounds all right, but it is not. It is not merely that we
cannot evaluate this proposal in the absence of an objective theory of
"causal influence" and of "what we do." Nor is it just a matter that the
argument won't tell against a theory such as ours that explicitly holds that
if there is only one possible future history issuing from this present event,
then, the future being settled, there is no "influencing" the future by what
we do in the present. The difficulty is above all that the proposed definition
of "the future history" is akin to defining "the odd prime number" as the
one that "alone is not divisible by two."
The second argument seems to be that the future is the one we wonder
about. As we indicated, however, at the very end of chapter 6, and again
a couple of pages ago, to the extent that wondering is wanting to know,
wondering is similar to other wants. Whether one wants to know what
will happen as the open future becomes determinate, or one wants to have
a dapple gray pony, one must bide one's time.
The third argument is about caring. Certainly branching time shares with
this passage the premiss that we do not care about other-worldly futures.
Indeed the passage suggests a certain mild ad hominem: What we care
about is what alternatives there are for us to choose among. We do not
(much) care what alternatives there are for other-worldly counterparts,
should they exist as Lewis's theory requires. We don't see how counter-
part theory can either make sense of or reject the demand that future
(incompatible) possibilities be for us rather than for our counterparts.
Suppose for example that we are choosing whether or not to start a sea
battle. Surely we care about what will happen if there is a sea battle,
and we also care about what will happen if there is not a sea battle, since
these are possibilities for us. And given that we really do have a choice,
and know that we do, such caring about incompatible alternatives makes
sense. If these histories are (right now) all really possible, then we do now
rightly care what is true on each. We really do care about what happens
on more than one historyas long as the histories are ours.
A is variable only in the history parameter. "Dated" sentences such as "At 4:00
P.M., A," or renditions of "The coin will land heads sometime tomorrow" have
this feature: Their truth depends on the history of evaluation (and in the second
example also on the moment of use), but not on the moment of evaluation.
So suppose A is moment-independent. And suppose that A is in fact settled
true at m. To look for a causal locus, we rely on the following.
Consider now the improper past of m, call it p. It could be that A has been
settled true throughout p, in which case, since settledness propogates forward,
A is "universally true," that is, A is settled true throughout all of branching
time. Suppose, then, that m has not been settled true from all eternity. Then
because of Fact 7-1, we may make a Dedekind cut of p into two nonempty
chains: At every moment in the lower portion, call it copen, the truth of A will
depend on the history parameter; while at every moment in the upper portion,
call it csett, A will be settled true. Thus, where copen draws to a close is the
very "point" at which the status of A changes from "not yet settled to be true"
to "now settled to be true." We put "point" in shudder-quotes because copen
may not end in a moment, so for definiteness we call all of copen the cusp of
causality.
The cusp of causality is where to look for the causal locus of the "effect"
that A is settled true at m. It is well to keep in mind that m, where we have
supposed A to be settled true, may itself minimally upper bound the cusp of
causality. For example, if the A in question is that we are at the restaurant
at 6:00 P.M., Murphy's Law suggests that it may well take right smack up to
6:00 P.M. in order to settle that fact in our favor; in which case the cusp of
causality is the entire set of proper predecessors of the moment of our being at
the restaurant at 6:00 P.M.. Some people think this is a defect in the theory;
we think it is a defect in the world.
Though certainly imperfect in that it is nonrelativistic, the concept of the
cusp of causality is nevertheless an empirical, objective causal concept. The
idea is that nothing can be an effect unless it changes from not-being-settled-
to-be-true to being-settled-to-be-true, and that the locus of causality must be
at the "point" at which this change takes place: The cusp of causality is just
where the effect comes to be settled true.
7C.1 Agents
Nowhere in this book do we offer thoughts that help much when considering
the mental equipment or other aspects of the "real internal constitution" of
agents. The topic is important, and important to us; it is just that our attempt
at progress in this book takes us in a different direction. Alan Ross Anderson
used to say that all progress in philosophy comes by simply assuming certain
problems have been solved even though they haven't, and getting on with the
investigation. We pursue this policy by entering a significantly uninformative
postulate concerning agents.
AGENTS. (BT +I + AC postulate. Reference: Post. 6) Agent is a nonempty
set.
We call the members of Agent agents, and we let lowercase Greek letters a and
(3 range over agents. We intend that the concept of Agent is absolute in the
sense of Bressan 1972 (or a substance sort in the sense of Gupta 1980), which
means that we may "identify" agents across times and histories. In particular,
there is no fission and no fusion of agents as we move from moment to moment.
In this book, however, we do not happen to discuss questions such as de re
versus de dicto ascriptions of agency, nor do we worry about when agents come
to be or pass away. That is why we can get by with the simpler set-theoretical
representation of Agent given by Post. 6. There is a brief discussion of Agent
as an absolute concept in 10C.l, where we are worrying about joint agency,
and a little more in 12F, where we are thinking about generalizing on the agent
position of stit statements. None of this, however, is offered as a serious con-
tribution to solving the problem of personal identity. On the other hand, we
do think that consideration of agents in branching time may deepen that prob-
lem. BT + AC theory, for example when considering strategies, characterizes
the same agent as making some (possible) choices sequentially, and also some
(possible) choices under incompatible circumstances. That the idea of "same
agent" seems essential to the idea of, for example, a strategy is, we think, a
good reason to suppose that a concept of personal identity that does not es-
sentially involve agency is much too partial. A concept of personal identity
that depends on exclusively "passive" notions such as experiential content, or
on only backward-looking ideas such as memory, is by so much inadequate.
7C.2 Choices
The penultimate parameter of a BT + I + AC structure is Choice, which tells
what choices are open for each agent at each moment in Tree.
212 Foundations of indeterminism
The postulate may strike one as unhappy to the extent that it seems to insist
that choosing is localized in a moment, and is therefore instantaneous. The
opposite seems so natural: Choosing takes time. We therefore feel obligated to
explain, in part, our reasons for giving a key role in BT + 1 + AC theory to such
a prima facie counter-intuitive postulate.
Jack has made a deliberate choice to go to the beach. There is, we think,
no harm in postulating that there is a momentary event that entirely precedes
his beginning to deliberate; let it be m1. There is also no harm in postulating
a momentary event that lies thoroughly after his arrival at the beach; let it be
m2. To find a plausible m1 and m2 we don't need to know exactly how Jack's
deliberation relates to its outcome; we only need to know that his deliberation
has a beginning that precedes his arrival at the beach. Having fixed m1 and
m2 in this way, if deliberation is to have point, it must not be decided at
m1, before the deliberation begins, whether or not Jack later arrives at the
beach (Aristotle). And at m2, after Jack arrives, it must certainly be decided
that he arrives (how else?). So now draw a chain of moments from m1 to
m2. Since that chain has members, which we call moments, then by simple
Dedekind analysis, there must either be a last moment of undecidedness or a
first moment of decidedness, or both (a Dedekind jump) or neither (a Dedekind
gap). Our theory of branching time does not say which; it only says that there
is a transition from undecidedness to decidedness. Nevertheless, our thoughts
gowe think harmlesslywith those who think of the flow of our world as
continuous, and hence without jumps or gaps. Furthermore, it is certain that if
our world is continuous, and can be accurately represented by a continuum of
instantaneous moments, then there must be either a last instantaneous moment
of undecidedness or a first such of decidedness. Since epistemology is not likely
to help us choose between these two, we see no objection to our always thinking
of a last instantaneous moment of undecidedness, a last moment at which it is
still not decided whether Jack will arrive at the beach. (Evidently there will
also be such a last moment of undecidedness if our world proceeds by discrete
jumps.) Talk of "deliberation flowing into action" is all right unless it blinds us
to these observations, and especially to the primacy of an objectively more-or-
less localizable transition from undecidedness into decidedness. It is this idea
that is idealized in Post. 7. We theoretically identify "the moment of choice"
as the last moment before the matter is decided, while still thinking of choice
itself as fundamentally a transition from undecidedness to decidedness. (See
2A.2 for discussion.)
Observe that nothing in the postulate denies that deliberating takes time. If
deliberation is mere wheel-spinning, we say nothing about it. If, however, it
involves, for example, a continuous (or discrete) ruling-out of alternatives, we
should then represent it theoretically as a continuum (or succession) of choices.
7. Agents and choices in branching time with instants 213
Finally, we observe that the postulate says nothing about whether choice is
localized in some homunculus in the brain, or whether it is always made by a
neuronic "community," or indeed whether choice is in any way localized. Nor
does any part of our theory say anything about what choice feels like, or whether
every choice is a conscious choice. These are difficult and important questions.
We remind the reader here, however, as we do elsewhere, that our explicitly
and advisedly limited theory concerns only the causal structure of choice, to the
exclusion of its "content." Nor will it hurt to insert our view that those who
deal with the "content" of choice to the exclusion of its (indeterminist) causal
structure can easily be led astray.
The fundamental postulate Post. 7 on choice says that histories are divided
evenly into equivalence classes; we discuss this aspect after we introduce the
equivalence-relation notation for choices, together with some closely related no-
tation that we use with considerable frequency.
of choice, m, lies in the past, then Choiceam (m1) picks out the "projection"
of the uniquely determined choice made by a at TO onto the present instant
or time.6
It is the intent of the postulate Post. 7 on Choice that when agent and
moment are fixed, choice equivalence is an equivalence relation on the histories
passing through the moment: reflexive, symmetric, and transitive. For example,
to say that Choiceam(h1) = Choiceam (h 2 ) is another way of saying that h1 belongs
to Choiceam (h2). The next definition describes the equivalence-relation notation
for choices.
CHOICE EQUIVALENCE. (Definition. Reference: Def. 12)
and we say that h1 and h2 are
choice equivalent for a at m.
We say that
h1 and h2 are choice separated for a at TO.
is defined only when instants are present, and when TO < m1,
We say that m1
is choice equivalent to m2 for a at m.
hi |am h2 is contrary to h1 =am h2. It makes sense to think of a given possible
choice as "separating" each history in the choice from each history not in the
choice. So if we wish to describe separation in a sentence that makes "a" the
grammatical subject, it is all right to say something like "a has a possible choice
at TOO that separates h1 and h2." We should not give in to the temptation to
say that "a can choose between h1 and h2." To put the matter as clearly and
therefore trivially as possible, a can never choose between histories, but can
only choose between choices.
We now consider whether Choice is correctly described as a partition, or, what
is technically equivalent, whether choice equivalence for a at TO is a reflexive,
symmetric, and transitive relation on H( m ).
Reflexivity is perhaps a "throwaway" postulate, except to the extent that it
implies that every history belongs to some possible choice. Perhaps it makes
sense to say that at some moment, if certain things happen a has chosen, but
if other things happen, a has not chosen; J. MacFarlane has suggested some
potential cases, and has urged that we keep our minds open to this possibility.
Nevertheless, it seems to us that speaking causally, and regardless of how people
evaluate a given agent or situation, we tend to side with those who find con-
ceptual difficulty in an alleged situation involving the simultaneous possibility
of choice and no-choice. We do, however, agree with MacFarlane that we know
much too little to be warranted in discouraging the development of alternative
theories along these lines.
6
Choiceam (m1) gives a set of histories and Choiceam (m1) gives a set of co-instantial mo-
ments, which is confusing. The underlining on Choice is a mnemonic intended to help keep
the two concepts apart by calling to mind the horizontal picture of an instant.
7. Agents and choices in branching time with instants 215
The fundamental idea is carried by h1 =am h2, which says that no choice for
a in M separates h1 and h2. "Inseparability" in this context is a geometrically
suggestive rhetorical variation on "choice equivalence" that correctly intimates
the ncmtransitivity of the relation. Its contrary is also conceptually important;
two histories are "separable for a in M," for instance, if somewhere in M there
is a choice for a that keeps one of the histories as a possibility while ruling out
the other as thereafter impossible.
Figure 7.4 provides an example of nontransitivity of inseparability, using a
simple case of a chain, c, consisting of the two moments {W0,w1} represented
by divided rectangles. The division of each rectangle represents that at each
moment there are two possible choices for a. The history, h, that splits from
the chain at the lower moment is choice inseparable for a by c from each of
the histories h1 and h2 that split at the top moment; but obviously h1 and h2
are not choice inseparable for a by c from each other. That is, no choice that
216 Foundations of indeterminism
a makes in the course of c separates h from h1, or h from h2, but there is
evidently a choice (at the top moment) that separates h1 from h2. So choice
inseparability for a at a chain is not and should not be transitive.
There are two further postulates relating to choices by agents. The first
amounts to a "new" principle relating agency to the causal order, while the
second concerns multiple or joint agency.
As reported in, for example, chapter 1, we learned the no choice between un-
divided histories condition from P. Kremer in 1987. All the postulates having
to do with choices are to be found in one form or another even earlier in von
Kutschera 1986.
This postulate is perhaps the most "interesting" of the BT + 1 + AC postu-
lates. As far as we know, the idea of relating choice to brute causal undividedness
has no earlier rigorous expression, even though the relation is and must be of
importance to the theory of action and to moral theory.
An easy consequence is that from the point of view of a properly later moment
m1, what choice an agent made at each properly earlier moment m, is uniquely
determined. Or to say the same thing from the point of view of m, for each
moment m1, properly later than TO, there is a unique possible choice for a at
m that contains all histories passing through m1. For argument, assume m1
is properly later than TO. Then the two moments constitute a (two-member)
chain that by Zorn's lemma can be extended to a maximal chain, that is, a
7. Agents and choices in branching time with instants 217
We don't pretend to know if there are any busy choosers, or even if the idea
makes sense. (We don't think anyone else knows either.) All we do is keep track,
as well as we can, of places in the theory of agents and choices in branching time
at which it makes a difference. In this regard, the following is worth noting:
Sometimes what makes a difference is the existence of an unending forward
series of nonvacuous choices, while sometimes it is the existence of an unending
backward series.
7D Domain
The last entry in the structure (Tree, <, Instant, Agent, Choice, Domain) is
Domain, which we employ as a field of possible denotata, and as the range of
individual variables.
That is, we postulate that moments, histories, instants, and agents shall all
be among "what there is." We enter this postulate partly for some technical
reasons that emerge in chapter 8, and partly to emphasize how harmless and
nonparadoxical it is to include these entities in the domain of quantification.
These entities are not in the least "meta-linguistic," nor does their addition
make the domain "too big" to be a set. Of course Domain must be subject to
some artificial limitation or other in order to avoid paradox, but we think that
this set-theoretical (or even ontological) problem has nothing whatever to do
with agency or the causal structure of our world.
8
Prior chapters of this book have offered bits and pieces of a description of
agency in our indeterministic world. This chapter brings us to the topic of the
semantics of a language taken to be used by those (ourselves) who live in such a
world. Just to have a label, we sometimes call this topic either "the semantics
of indeterminism" or "indeterministic semantics."
Agency constructions we treat in 8G; until then we emphasize branching time
itself. We include (i) quantificational devices, (ii) temporal constructions, (Hi)
historical modalities, (iv) some mixed modalities, and (v) indexicals tied to the
context of use. We include all of these items because they crop up in discussions
of indeterminism, sometimes in a confusing way. We say what we have to say
in the idiom of formal semantics. This amounts to an organized account of the
semantics needed for a language spoken in an indeterminist world. First we go
over and extend some of the foundational and generic semantic ideas broached
in chapter 6. Then we go one by one through a large number of constructions
useful for understanding indeterminism, in each case giving an exact semantic
account. We emphasize points that we take to be important, and we draw
out the analogies among and differences between the semantics of quantifiers,
historical-modal and tense connectives, and indexical connectives, and indicate
how they are to be combined. We briefly review the already-presented semantics
for the achievement stit and the deliberative stit, and we explain how stits might
be witnessed by a chain of choices (instead of a single choice). The chapter ends
with a mention of an alternative agency construction, the transition stit. Let us
note that in compensation for the inevitable tedium of processing henscratches,
the early sections expand on the fundamental ideas that go into the formal
semantics.
220
8. Indexical semantics under indeterminism 221
8A Sources
As we briefly noted at the beginning of 6B, we draw on four sources for the
key ideas needed for the semantics of indeterminism.
Quantification. Tarski's studies in the 1930s provide the foundation for
all compositional semantics, and in particular are the source for our com-
positional understanding of quantifiers.
Tenses. Prior 1957 initiated the compositional understanding of linear
tenses.
Historical modalities. Prior 1967 as made rigorous by Thomason 1970
(see also Thomason 1984) is the source of our understanding of the his-
torical modalities, and also of how tenses should be understood given a
representation of indeterminism by means of branching time.
Context of use. Kaplan 1989, much of which had circulated in typescript
since Kaplan's 1971 lectures, provided the full-scale development of "in-
dexical semantics" based on the idea that compositional semantics must
pay delicate attention to the context of the use of an expression.
The semantics of indeterminism relies heavily on all four of these sources, which
we discuss in turn as ways of following out the fundamental Tarski idea of
relativizing truth to parameters.
Open sentences such as F X 1 X 2 do not have an absolute truth value. They rather
have a truth value only relative to assignments of values to the variables, so that
such assignments are parameters of truth.
Let us be explicit about how we extract the assignment parameters from
Tarski's account so that those familiar with Tarski may understand our stylistic
departure. His fundamental locution is illustrated by
where the members of the sequence such as a and b are individuals drawn from
the domain. To make this work, one must know which member of the sequence,
(a, b, ..., ), goes with which variable. In this example, we have used subscripts
whose sole purpose is implicitly to supply an ordering of the variables. This
order is to be used in conjunction with the order of the sequence, so that we know
that a goes with x\ and b with x2. The ordering of the variables is essential to
the meaningfulness of the Tarski relation of satisfaction by sequence: One must
understand which member of the sequence goes with which variable. As Tarski
observes, however, one does not need for this purpose to suppose a primordial
ordering of the variables such as supplied by subscripts. A more local ordering
will do as well: One could, for example, take the order from "first occurrence in
the sentence under consideration." In this case, a would go with 2 and b with
x\. No matter: In any case, since what "satisfies" a sentence is a sequence of
entities from the domain, one must have some ordering of the variables in order
to be able to say which entity goes with which variable.
The first step in our extraction of the assignment parameters is to observe
that, if we take what satisfies a sentence to be a sequence, there is always a
detour through some sort of imposition of an order on the variables. This detour
has as its only purpose defining which entity in the sequence goes with which
variable. We avoid the detour by letting the second argument of the satisfaction
relation be not a sequence but a function, a, defined on the variables, such
that a directly assigns an individual in the domain to each variable, without
presupposing a sequencing of any kind. So after this first step we have, for
example,
The conceptual aspect of the second step is that in choosing (3) we are empha-
sizing that an expression such as "Fx-^Xi" is grammatically sentential. We know
it is sentential because it is subject to exactly the same embedding operations
as any other sentence: conjunction, negation, and the like. Therefore, for ex-
ample, "Fx^Xi" deserves a sentential semantics, which is to say, an account of
the conditions under which it is true, in this case not absolutely, but relative to
8. Indexical semantics under indeterminism 223
characteristically adds one or two new immobile parameters, the set of "worlds,"
and perhaps a "relative possibility" relation on that set, and also a mobile
parameter, namely, the "world" parameter. The familiar modal connectives are
translocal in the world parameter. This relativization of truth to worlds is not
needed for understanding indeterminism as we conceive it, and so we add no
such parameters. On the other hand, the technical devices and conceptual ideas
of modal logic due especially to Kripke are essential to both plain or linear tense
logic, and to the historical modalities of branching time.
In each case we implicitly allow that Domain might or might not be present,
depending on whether or not quantification is at issue.
from Moment-History into functions from Domainn into Domain; and assigns
to each n-ary predicate letter a function from Moment-History into functions
from Domainn into {T, F}.
Representation of "world" and "language" must fit, and when they do, we
call the combined representation a "model."
You will note that in expressing Observation 8-2, which is derived from Ka-
plan, we use loose language. It is difficult to do otherwise for the following
reason: Although we have in mind that a sentence can be considered either
as stand-alone or as embeddable, the symbolic language (unlike, e.g., English)
makes no such distinction. In the symbolic language, there is no syntactic mark
(such as initial capitalization and final period in written English, or intonation
in the spoken language) that distinguishes sentences taken as stand-alone from
those taken as embeddable. This lack of match between English and the sym-
bolic language makes analysis more difficult. Here is the best we know how to
do (without describing a new kind of symbolic language) by means of a defini-
tion and a policy. (See Green 1998 for a study of "illocutionary-force-indicating
devices," including Frege's sign of assertion.) First the definition, which is a
BT +1 + AC-specific version of the more general concept indicated in Policy
6-2.
The rationale for the first part of this policy is that each utterance should be
conceived as tied to a concrete context, and that such a context determines a
unique causal position, with a definite past and a definite future of possibilities.
We idealize such a position with the moment of use, mc. This moment of use
is the very moment at which we wish to evaluate a stand-alone sentence. That
is why, for stand-alone sentences, we initialize the moment of evaluation with
the moment of use. Keep in mind, however, that the moment of evaluation is
mobile, and can be shifted by tense constructions ingredient in the stand-alone
sentence. And that is the very reason for the second part of Policy 8-3.
Since there is in the symbolic language (and indeed often in the English
examples of philosophers) no syntactic difference between stand-alone and em-
beddable sentences, the definition of "context-initialized point" (Def. 16) and
8. Indexical semantics under indeterminism 231
Policy 8-3 represent the best that we can do. The definition and policy are,
we think, exceptionally useful in discussing tenses and indeterminism, for in
those ventures the failure to observe the distinction between stand-alone and
embeddable sentences is especially harmful.
Here is one way the definition and policy offer immediate progress. Kaplan
(pp. 505-506) asks that in thinking about the "character" or meaning of a sen-
tence, we first fix context, and then ask for the content of the sentence in that
context. There should, however, be' two notions of content-in-a-context, depend-
ing on whether we are thinking of the sentence as stand-alone or embeddable. If
we are thinking of it as stand-alone, then the moment of evaluation is initialized
by the moment of the context. Since one cannot reasonably treat a sentence
with free variables as stand-alone (Assertability thesis 6-7), it is obvious that
for stand-alone sentences the history is the only mobile parameter that is left to
vary when considering a stand-alone sentence. If one correlates time to moment
and world to history, this explains the otherwise puzzling phenomenon noted
by Kaplan on p. 546:
... the truth of a proposition is not usually thought of as dependent
on time as well as a possible world. The time is thought of as fixed
by the context.
That is right for stand-alone sentences: Time (or moment) is fixed, while world
(or history) is not. If, however, we are thinking of the sentence as embeddable
by means of translocal connectives such as tense operators, then for "content
in context" we must let (i) the moment of evaluation diverge from and vary
independently of ( i i ) the moment of the context. This explains why Kaplan
permits content to vary over times as well as "worlds." We shall remain unclear
as to the point of our semantic constructions unless we bear this in mind.
We note that Kaplan is working in a Tx W framework (see 7A.6). It is, we
think, an indication of the relative helpfulness of the moment-history framework
that it explains a phenomenon that from the point of view of Tx W seems just
puzzling.
8-4 QUESTION. (History of the context?) Because we shall have "modal" con-
nectives that are translocal with respect to histories, it follows that there must
be a mobile "history of evaluation" parameter. Furthermore, the mobility of
the history of evaluation plays an essential role in our account of assertion (Se-
mantic account 6-12); but why is there not also a "history of the context," to be
a paired with the mobile history of evaluation, as context parameter? If there
were, for stand-alone sentences we could initialize the history of evaluation with
"the history of the context."
if any facts had been different in any way, even if they are only facts
entirely independent of and isolated from the utterance itself, then
the context of the utterance would, ipso facto, be a different context,
even if the utterance itself remains exactly the same. Salmon concludes that
the structure and interpretation parameters because we wish to have not only
"truth," but also "equivalence," "implication," and "validity."
We want a way of referring to the semantic value of any categorematic expres-
sion, be it term or sentence. Once we have "semantic value" for sentences, we
automatically have truth, falsity, and the idea so important for indeterminism,
settled truth.
SEMANTIC VALUE, DENOTATION, AND TRUTH. (Definition. Reference: Def.
16)
Semantic value. For any categorematic expression, E, be it term or sen-
tence, Val m , m c , a , m / h ( E ) , is "the semantic value of E at the point (937,
mc, a, m/h>." Val m,mc a m / h ( E ) is defined recursively by clauses given
in 8F and 8G. Note that by the earlier clause of Def. 16 that appeared
on p. 229, we may write Valtt (E) in place of
Denotation. Where t is any term, Val m , m c , a , m / h ( t ) E Domain. Valtt(t),
or V a l m , m c , a , m / h (t) is "the denotation of t at the point <m, mc, a,
m/h>." Also, as before, Valtt(t) stands in for Val m, m c , o , m / h ( t ) .
Truth. Val m,mc,a,m//i(A) is "the truth value of A at <m, mc. a, m/h>."
Where A is any sentence,
Alternate much-used notation for truth and falsity:
T.
- m, mc, a, m/h tt= A iff Va/m, m c , 0 ,m/h(A) = Either is read "A
is true at point <m, mc, a, m/h>."
- m, mc, a, m/h A iff Va/gjt,m c ,a,m/h(-<4) = F. Either is read "A
is false at point <m, mc, a, m/h)."
We sharply distinguish settled truth, which is not history dependent, from plain
truth, which is.
SETTLED TRUTH. (Definition. Reference: Def. 17)
A is settled true at a moment m with respect to m, mc, and a iff 971, mc,
a. m/h 1= A for all h H( m )- We may drop h, writing m, mc, a, m t=
A.
A is settled true throughout a set of moments, M, with respect to m, mc,
and a iff m, mc, a, m \= A for all m 6 M. In addition to dropping h, we
may replace m by M, writing m, mc, a, M M A.
We most often write just m, M -= A and m,1, m t= A since we use these
concepts most often when assignment and context are not relevant.
Strong validity is interesting and useful, but perhaps not as intuitive as its in-
context cousin. In-context validity is suitable only for sentences considered as
stand-alone.
is the point that is just like (371, mc, a, m/h) except that the value of the
parameter, Xj, is shifted to be z. (If there is no such point, the notation
is undefined.)
In the rest of this long section we present a plethora of semantically explanatory
clauses for the "stit-free" part of the language. Then in 8G we go over the
stit ideas. We organize the work of the present section on the "stit-free" part
as follows. After the clauses covering the atomic expressions of the language,
we organize our explanations in terms of abstract properties of the functors,
primarily considering whether they are "local" or "translocal" in various mobile
parameters in the sense of Definition 8-1. Also we occasionally need the concept
of "anchoring." Before proceeding, we give a definition of "anchoring" in rough
terms, which suffices for present rough purposes. (The rigorous definition is too
tedious to be helpful here.)
8-7 DEFINITION. (Anchoring) Let $ be a functor.
$ is anchored in a parameter if in passing from E to o (E), you require
the very identity of the current value of that parameter, over and beyond
the pattern of semantic values of E as the parameter varies.
240 Foundations of indeterminism
LXj(A). READING: A definite description: the sole x.j such that A. SE-
MANTICS: Val<w,mc,a,m/h(i<X3(A}) = the sole z such that [z / Xj](9Jl, mc,
a, m/h) \= A if there is exactly one such. If not,
("the non-existing object").
Other variable-binding operators are analogous.
5
That makes it plausible to guess that "absolute" is a good thing to mean by "extensional."
That won't work, however, because, e.g., predicate letters (of the sort considered here) are
thought to carry only extensional meaning, even though they are anchored in the model, m,
and therefore not absolute. One might next guess that "local in all parameters" is a good
meaning for "extensional." Observe, however, that if we were to take this as an explication of
"extensional," then quantifiers would turn out just as "non-extensional" as, e.g., the modali-
ties. Given all this confusion, and since it is difficult to determine what "extensional" means
(van Bentham 1988, p. 109, suggests that "no general satisfactory definition seems to exist"),
we feel that it is better to stick to the ideas of "local" and "absolute," which have clear and
definite meanings that are logical rather than ideological.
242 Foundations of indeterminism
Grammar LX-j(A) is a (categorematic) term, the definite description. Its value when
existence-and-uniqueness fails is a mere throwaway. It corresponds to Kaplan's definite
description. For his "dthat" one may use ix} (Actually1 :A), which is context-dependent,
and independent of the moment and history of evaluation. (We use t. instead of its
inversion for convenience.)
Was:A. READING: It was true that A. Or put the main verb of the
reading of Aif it has oneinto the past tense. SEMANTICS: 977, mc, a,
m/h == Was:A iff 3mi[mi E h and m1 < m and 071, mc, a, m\/h 1= A].
8. Indexical semantics under indeterminism 243
Prior's "P." Shift along the present history, existentially, to earlier moments, and check
them for A (with respect to the current history). Because histories are closed downward
and thus form a chain, the clause "m1 h" is redundant. That the route of backward
travel is uniquely determined, however, should not blind you to the importance of
the nontriviality of keeping to the same (the current) history when you evaluate A.
As Prior explained, Was:A does not imply Sett: Was-Aalthough Was-Sett.A does
indeed imply Sett:Was:A. The point is that although the route traveled is unique, it
is part of many different histories
Will:A. READING: It will be true that A; or put the main verb of the
reading of A, if it has one, into the future tense. SEMANTICS: m, mc, a,
m/h t= Will:A iff 3mi[mi h and m < m\ and 971, mc, a, mi/h N A}.
Prior's "F." Here you shoot forward, existentially, along the current history, checking
each moment along the way. And look: In contrast to Was:, you cannot, in under-
standing Will:, get rid of a reference to the history parameter.
Grammar' t is any term, and A is any sentence At-mstt :A reflects English con-
structions such as "At 4.00 P M. the coin will come up heads." The clause tells you
to shift W, mc, a, m/h by replacing the current value of the moment-of-evaluation
parameter by a new moment, namely, the one (and the only one) in which the current
history intersects the instant specified by the value of t. "Travel up or down the current
history until you hit the instant t; that moment is where you must evaluate A (with
respect to the current history)." This understanding of At-mstt '-A makes it false when
t does not refer to an instant. We manage to live with this awkwardness. You can
see that At-instt ' is translocal in and only in the moment-of-evaluation parameter, for
that is the only parameter that is shifted. Also, as long as t itself is independent of
the moment-of-evaluation parameter, At-mstt :A is bound to be independent of the
moment-of-evaluation parameter. For instance, let t be "4:00 P M.," with some definite
date understood. Adding additional temporal connectives to At-mst^-oopM :A has no
more effect than (is just as vacuous as) nesting one xj-binding operator within another.
For example, "It has always been true that at 4:00 P.M the coin comes up heads" just
reduces to "At 4:00 P.M. the coin comes up heads." So does "At 2:00 P.M. at 4:00 P.M.
the coin comes up heads" (see the discussion of Figure 6.3).
Some philosophical logicians feel that including terms naming times (or instants) in a
formal tensed language is wrong-headed, or at best inelegant. Our excuse for doing so is
that such terms as used in "at" constructions play a role in many philosophical discus-
sions of determinism versus indeterminism. Not everyone is clear on how they should
work under indeterminism, which makes it worthwhile to clarify their logic. Of particu-
lar importance in understanding indeterminism is the fact that At-instfaogp M.'-A is not
in general independent of the history-of-evaluation parameter. "At 4.00 P M. the coin
comes up heads," or even "It has always been true that at 4.00 P.M. the coin comes up
heads," can be just as dependent on the history-of-evaluation parameter as "It will be
true that the coin comes up heads." "At" constructions pose quite the same problems
as does the future tense. These matters are difficult to hear or see in natural language,
244 Foundations of indeterminism
where we can (apparently) say, "At 4:00 P.M. it will be inevitable that Jack at I'OO P.M
might be running at 3:00 P.M. at 5:00 P.M." We don't know if this makes sense or not.
Our point is that those who make philosophical points explicit in a Prior-Thomason
connective form have much less difficultyespecially with regard to scope, which is so
ambiguous in English and so clear when all tense talk is carried by connectives.
Grammar: t is any term and A is any sentence. This construction travels up or down
the current history to the moment denoted by t, and evaluates A just there, still on
the current history. It does not move off the current history (it is local in the his-
tory parameter), and is therefore a true temporal construction. Accordingly, it cannot
be successfully used in connection with moments off the current history. (Contrast:
At-mstt :A is always successful when t denotes an instant, since the current history
intersects every instant.) The At-momt ' connective seems much stranger than the
At-mstt ' connective, chiefly, we suppose, because we don't have names for moments,
but we do have names for instants (or times). The At-momt: construction is never-
theless of great usefulness in untangling "double time references" such as are needed
in order accurately to understand the following scenario. (Our account may sound a
little awkward, because we omit several indexicals that would naturally be used.)
It's 4:00 P.M. At 2:00, Themistocles said, "I promise that Themistocles will, (11)
choose to fight a sea battle."
We are interested in how one can use the semantic content of (12) in order to illuminate
promise keeping. So let it be true at 4:00 that Themistocles has kept his promise; such
is, at 4:00, a settled fact. What does this mean? Here are two candidates that don't
work.
- At 4:00 Themistocles has kept his promise iff (12) is settled true now, at 4:00.
This is wrong. After all, now, at 4:00, what Themistocles promised, namely, (12),
is settled false. The point is that when evaluated at 4:00, the future tense of (12)
would reach forward into times later than 4:00, long after the sea battle. (We
neglect as a distraction the possibility that there be another sea battle.)
At 4:00 Themistocles has kept his promise iff (12) is settled true at the moment
of promising.
This is wrong. At the moment of promising, the sentence, (12), was not settled
true. That is, (12) was, at the moment of promising, true on some histories and
not on othersfor the choosing still lay in the future.
What signifies that the promise has been kept is more complicated.
- At 4:00 Themistocles has kept his promise iff at 4:00 it is settled true that (12)
was true at the moment of promising.
The key is the "double time reference": The "settled" is evaluated later, at 4:00,
while what is settled is the truth of (12) at the moment of promising. See 5c.2.1
for a more extended discussion of double time references.
8. Indexical semantics under indeterminism 245
The upshot is that you cannot do without "the moment of promising," even though
you certainly do not have a name for it. Here is how the analysis could go in indirect
speech, where, as advertised, we can use the At-momt. connective. (Also we shall be
using (12) instead of mentioning it.) We let the moment of promising be mp. At 4.00
Themistocles kept his promise iff
Wrong: At-mstn oo-'(12).
- Wrong. At-mommp:Sett:(12).
- Right: At-mst4 00.Sett:At-mommp:(12)
None of these connectives can be shifted around; and it is essential that there be a
"double time reference."
6
It would be in the spirit of Thomason and Gupta 1980 to add a parameter that makes
sense out of picking a particular history through the moment of use, perhaps a history that
is "very like" the current history, which passes through the current moment. Since, however,
"very like" is not an idea that fits well with the more austere causal notions that we employ,
we forgo adding such a parameter.
8. Indexical semantics under indetenninism 247
As always, Now: and the actuality connectives do very little work at the head of
sentences considered as stand-alone, since anyhow, by Policy 8-3, the moment
of evaluation is already initialized by the moment of use, so that the shifting
called for by Now: or an actuality connective is vacuous. When embedded in
translocal connectives, however, Now: and the actuality connectives really do
shift the moment of evaluation.
This completes our discussion of how various variable-binding, temporal,
history-modal, and indexical functors work in branching time. Next we take
up the stit functors.
The agency condition is needed when the agent position of a stit sentence is available
to every term. In this book we do not often enter the agency condition explicitly, since
usually, such as when we reserve "a" for an agent-term, it is a presupposition that the
term in the agent position denotes an agent. We remark that in normal applications,
one would expect that the term, t, is independent of both history and moment, as is
always true for our use of a. One will usually obtain what one wants by trading in [t
silt: A] for 3xi[xi = t & \xi stit: A}}.
In the semantics of dstit, there is no double temporal reference: The moment of eval-
uation of A and the moment of choice are identified, so that while dstit is an entirely
viable candidate for helping to analyze agentive locutions, its expected properties are
considerably different from those of the achievement stit. If one wishes to think of dstit
as reporting an "action" or even a "choice," it is difficult to say in comfortable English
just when the choice is made. The difficulty is that before or at m it is not yet settled
which choice Val,r(t) makes, while at any later moment the choice has already been
made. The source of the difficulty is that dstit reports an immediate transition, Def.
8, so that there is no room for an "action" qua "event" between initial and outcome.
248 Foundations of indeterminism
The solution is to understand "the action" as being the transition itself. Since a tran-
sition consists in a pair of "events," initial event and outcome event, it is obvious that
a transition cannot have a "simple location" (Whitehead).
We next offer a representation, suitable for use with the achievement stit as
in 9B, of the "all in" ability of Austin 1961 (p. 177), present or absent on
a particular occasion for a particular agent and with respect to a particular
complement. Such ability statements can be tensed either as of the moment of
witness or as of the moment of evaluation of the complement. We use "can" for
the former and "could have" for the latter:
[ii cant2-stit: A}. READING: Agent t1 can see to it that at instant 2, A.
SEMANTICS: 9JI, mc, a, m/h 1= [ti cant2-stit: A] iff Valnfa) Instant,
and there is a moment mi in i >m (i.e., TOI lies on the horizon from m
at z) such that m is a momentary witness to [ti stit: A] at TOI relative to
(9JI, mc, a, m/h).
[t could-have-stit: A}. READING: t could have seen to it that, at the
present time, A. SEMANTICS: SOT, mc, a, m/h t= [t could-have-stit: A] iff,
where a; is a variable not free in [t stit: A], [i(m) I ^K^j fricj ai m/h) 1=
Was:[t canx-stit: A].
13 for some relevant developments via the theory of strategies, and 8G.5 for
an additional suggestion.
We turn to the quest for an understanding of the witness of a stit by a
possibly unending sequence of choices. We use the notion of choice inseparability
explained in Def. 13.
[ti tstit: t-2 =^ A}. READING: See (13). SEMANTICS: 371, mc, a, m/h
N [ti tstit: 2 ==> A] iff, letting a = Val^^i), a 6 Agent; letting TOO =
Val^fe), TOO Tree; letting c = Valv(t3), c is a chain of moments; and
Moment independence of outcome.
Applications of the
achievement stit
This page intentionally left blank
9
The complexity of the connections among actions, moral responsibility, and the
alternatives open to an agent have long tormented philosophers.* Hume, for ex-
ample, the most famous of all compatibilists, claims that universal determinism
is not only consistent with human freedom but necessary for morality (An En-
quiry Concerning Human Knowledge, section 8, part 1). His view is that while
agents are able to choose among alternatives, there is the liberty of voluntary
action. We think that is right. On the other hand, Hume asserts, morality is
without foundation if actions are not fully determined. We think that is wrong.
Our intent, however, is neither to joust with Hume nor directly to engage more
modern entrants in the "free will" tourneys. Instead we recount a tale that
begins with the following proposition, one that might have puzzled even such
as Don Quixote of La Mancha and his squire, Sancho Panza.
If an agent is morally responsible for doing something, then the , .
agent could have done otherwise.
Modern replies to Moore 1912 by Austin 1961 (pp. 153-167) and to Frankfurt
1969 by Van Inwagen 1978 put the issue in just these terms. In an effort to
unravel the complexities of (1), we here cleave it in twain, each conjunct seeming
essential to its meaning. (See also the discussion in 5A.I, which relates the
matter to promising.)
If an agent is morally responsible for doing something, then the , .
agent did it.
If an agent did something, then the agent could have done oth- ,^\
erwise.
The middle term, doing something, is thereby revealed. We vouchsafe that
for the route from moral responsibility to "could have done otherwise" to be
"With the permission of Kluwer Academic Publishers, this chapter draws on Belnap and
Perloff 1992. You will note that here, as in that article, our diction sometimes quixotically
imitates the source of our examples.
255
256 Applications of the achievement stit
accurately charted, heed must be paid this idea, formerly hidden, that stands
at the crossroads. We shall not in this chapter further consider the claim of the
first conjunct (2). We remark, however, that anyone who wishes to argue for
or against the claim that moral responsibility implies "free will" should note at
least the following: There is a deep difference between (2) as here displayed and
its sound-alike,
If an agent is morally responsible for an outcome, then the agent /,..
saw to that outcome.
This sound-alike, (4), is almost certainly false, a fact that in no way impeaches
a claim that (2) holds. Take the following examples.
Sancho Panza was morally responsible for having seen to it that
Quixote's lance was unpolished. Therefore, Sancho Panza did (2x)
see to it that Quixote's lance was unpolished.
Sancho Panza was morally responsible for the fact that Quixote's
lance was unpolished. Therefore, Sancho Panza did see to it that (4x)
Quixote's lance was unpolished.
The example (2x) of (2) rings true. The question of moral or legal responsibility
of an agent for doing something (such as seeing to it that a lance is unpolished)
seems to presuppose that the agent did that something. On the other hand, it
sometimes makes good sense for a moral or legal code to assign responsibility
for a fact (such as the fact that a lance is unpolished) without determining
whether the agent saw to that fact. (If it was Panza's moral job to keep the
lance polished, then it is plausible that we may hold him responsible for its lack
of polish without inquiring into his agency in the matter.) So the example (4x)
of (4) goes wrong. Stit theory helps to avoid confusing (4), which is false, with
its true sound-alike cousin, (2). (See also 5A.l.)
Leaving the first conjunct (2) aside, however, we press attention on the
morality-free claim of the second conjunct, (3), that doing something implies
having been able to do otherwise. There has been little effort to clarify the
second conjunct in isolation from moral considerations; we deem worthwhile
the enterprise of examining the relation between agentive doings and what it
means to say that "an agent could, or might, have done otherwise." Hume's
own account, "if we choose to remain at rest, we may; if we choose to move, we
also may," seems, for example, to suggest a tie between "could" and "might":
An agent could have done otherwise iff there is something else he or she might
have done. Is that right? Our view is that the armory of ordinary language
is inadequate to the task of deciding such questions. We need the weaponry
provided by something like stit theory, a theory that is careful in placing agents
and their doings in relation to the causal structure of our world.
Our preparations having been completed in earlier chapters, we venture forth
to battle a variety of questions in the topic of agency and "could have done
otherwise" as embodied in (3). Our assembled conceptual apparatus, though
weighty, is neither more nor less than is needed to complete the task. On the
9. Could have done otherwise 257
one side we avoid the perils induced by the vagaries of ordinary language, for
example the differences between "might" and "could" in this context, and the
exact target of the anaphoric reference of "otherwise." With stit normal forms
to guide us, we sharpen the contrasts and bring each topic into clear relief. On
the other side, we first render a question or conjecture in the language of stit,
meaning in this chapter the achievement stit, unless we give explicit notice to the
contrary. Then we employ stit theory to render formal judgment (an upshot).
That, in turn, requires clarity about alternatives, about matters temporal and
historical, and about possibilities in branching time, such as we have studied in
earlier chapters.
QUESTION. Is what could have been the same as what might have been?
We hope it is obvious that our suggestion is not empty: "Might have been" and
"could have been" are English, while "Might-have-been:" is formally character-
ized in 8F.6.
CONJECTURE. What an agent might have done is different from what he or she
could have done.
By "the agent might have done it" we intend an impersonal modality, expressed
perchance less ambiguously (and less idiomatically) by "it might have been that
the agent did it."
258 Applications of the achievement stit
Figure 9.1 has a picture. The abstract situation is this. The rectangles are
choices for a, whereas what happens at Wi is not up to the agent. The salient
feature is that while [a stit: Q} at ma has a witness, namely, u>2, that witness
does not stand in the past of mo. So at mo it is true that a might have seen to
it that Q, but it is false that a could have seen to it that Q. In contrast, at mo
it is true that a could have seen to it that R, since w0 witnesses [a stit: R] at
each of mi-m3. Here is an example that fleshes out this abstract description of
Figure 9.1.
At the moment, WQ, that ends his commending, the Knight of the Mournful
Countenance had the choice either to stand down or ride on. Once having
begun his charge, however, there was a slightly later moment, Wi, at which
Rozinante might by chance have collapsed. In the case of no collapse, there
9. Could have done otherwise 259
was then a later moment, W2, at which Quixote had the choice either to swerve
toward or to swerve away from the disastrous confrontation. Let my, as shown
in Figure 9.1 be the moment at which he rammed the windmill. Consider the
history on which, as he finished his commending, the Knight regained his wits
and stood down, and follow that history to a moment, mo, that is co-instantial
with the moment, 7713, at which he rammed the windmill. At that moment, mo,
it would be true to say that he might have attacked the windmill, but false to
say that he could have. What decides the matter is that there is nothing he
could have chosen at the end of his commending (WQwhich is the only choice
point in the past of the moment of non-attack under consideration) that would
guarantee his attack. Both chance (at wi) and the uncertainty of the outcome
of a future choice (at w%) stand in the way of such a guarantee.1
CONJECTURE. If yon fellow sees to some state of affairs, then it might have been
that the state of affairs not obtainat that very instant.
The final phrase accomplishes a task more easily than idiomatic English: Make
sure that the "might" means that the absence of the state of affairs obtains
1
Permit us yet another restatement of methodology: Although we think the distinctions
we are drawing are important for, e.g., moral analysis, we by no means fancy that our chosen
expressions have a perfect fit with ordinary English.
260 Applications of the achievement stit
These stit versions say that if [a stit: Q], then it has not been inevitable (de-
termined) from all eternity that Q should obtain at the instant in question.
UPSHOT. The conjecture is, in its stit versions, true. It is an obvious conse-
quence of the negative condition (8G).
EXAMPLE. If it was inevitable from all eternity that the hog gelder's reed flageo-
let sounded four times while Don Quixote was at his meal, then the hog gelder
did not see to it that his reed flageolet sounded four times while Don Quixote
was at his meal. A hard determinist valiantly endorses the consequent; a soft
determinist becomes cross, changes the topic, and exits the lists.
CONJECTURE. If a does something, then it might have been otherwise; that is,
a might not have done it.
Here let the "otherwise" refer anaphorically to the entire stit sentence, not just
to its complement.
UPSHOT. True.
We belabor the obvious by offering two proofs. First, since [a stit: Q] implies Q
(by the positive condition), so that ~Q implies ~[a stit: Q], this is an immediate
consequence of Upshot 9C. The second and more important proof is this: The
consequent is a truth of logic, so that the implication is vacuous! This is related
to the Triponodo principle of Makinson 1986, except here instead of the "trivial
(legal) power not to do," we have the "trivial possibility of not doing." The
argument that it is logically true is an easy reductio. If [a stit: Q] were settled
true throughout an instant, then by the positive condition, Q would be settled
true throughout that same instantwhich would leave no room for satisfaction
of the negative condition.
9. Could have done otherwise 261
CONJECTURE. "The fact that a person could not have avoided doing something
is a sufficient condition of his having done it" (Frankfurt 1969, p. 150).
STIT VERSION 2. ~[a could-have-sht: ~[a stit: Q]} implies [a stit: Q}.
262 Applications of the achievement stit
EXAMPLE. Since the first version is an easy application of 9D, we illustrate only
stit version 2. On the side of the antecedent, it is evident that not even the great
Mameluke of Persia, either before or after his nine-hundred-year enchantment,
could have refrained from seeing to it that if the golden helmet of Mambrino
was made of brass, then it was made of brass; but on the side of the consequent,
that dignitary certainly did not in fact see to that, nor to any tautology.
CONJECTURE. That we are responsible for some state of affairs implies that it
must have been possible for us to have been responsible for its absence.
UPSHOT. But as all contemporary logicians of action know, the most elementary
story tells us that the conjecture is false. In stit theory, the relevant point
emerges through the negative condition, which requires only that the falsity of
Q be risked, not that its falsity be guaranteed.
EXAMPLE. La Tolosa, the fair cobbler's daughter from Toledo, saw to it that
Don Quixote was girded with his sword; but given the rough company of carriers,
to say nothing of La Molinera, that poor wench was evidently in no position to
see to it that the knight failed to be girded.
2
Perhaps the evident plausibility of the Frankfurt examples derives from the fact that
so many verbs give rise to both agentive and non-agentive readings, a matter that we have
suggested can be at least partly resolved by applying the stit paraphrase thesis, Thesis 3.
9. Could have done otherwise 263
Some think yes. Chisholm 1964a, for instance, says that if some varlet
loosed his firelock, then "there was a moment at which it was true, both
that he could have fired the shot and also that he could have refrained
from firing it." (Observe that Chisholm's "also" is not "could have seen
to it that the shot was not fired"; he does not make the superficial mistake
of supposing that the Conjecture of 9F is true.)
Some think no. Frankfurt 1969 supposes it possible that there should be
such a thing as "the fact that a person who has done something could not
have done otherwise."
On our view this question is not to be happily represented without taking into
consideration the stit analysis of refraining, so that we are not surprised to
find that on those few occasions that the literature notices the existence of the
question, it seems to resort to sheer postulation.
We interpret the question as asking whether or not the fact that [a stit: Q]
is true at m implies that there is a moment in the past of m that stands as
witness to [a stit: ~[a stit: Q}} at a moment co-instantial with TO. That is, is
there some single choice point in the past of TO that, had a different choice been
made, would have guaranteed the agent's failure to stit Ql
UPSHOT FOR STIT. The implication fails, with an easy example, though not
quite so easy as the counterexample to the Conjecture of 9F. In Figure 9.2,
Q is settled true at TOO and TOI, and settled false at 7712, all of which are co-
instantial.
264 Applications of the achievement stit
Abstractly put, each of WA and WB picture a choice for a; you can see from
the diagram that Choice0^,A(mn} {TOO}, so that WA witnesses the truth of [a
stit: Q] at mo (looking to m^ for satisfaction of the negative condition). On the
other hand, Choice^,A(mi) = {mi, 7712}, so that since [a stit: Q] is true at mi,
WA does not satisfy the positive condition for witnessing that [a stit: ~[a stit:
Q}\ is true at m^. The choice point WA is "too soon." (The choice point, WB,
however, does that job.) Therefore, [a could-have-stit: ~[a stit: Q}} fails at m0,
and therewith the implication stated in the stit version.
Here is a concrete example based on Figure 9.2.
DSTIT VERSION. Does [a dstit: Q] imply Can:[a dstit: ~[a dstit: Q}}7
UPSHOT FOR DSTIT. In contrast to the stit version, the implication holds.
As in 8F.4, we take Can:[a dstit: Q] as simply Poss:[a dstit: Q], noting that
the formal countenance of "can" for dstit can be less wrinkled than that of
"can" with the achievement stit, because one need not worry about a double
temporal reference. The implication then comes to this: [a dstit: Q] implies
Poss:[a dstit: ~[a dstit: Q}}. A proof can be found in Horty 2001.
3
We follow Pellicer in identifying the unnamed duke and duchess with Don Carlos de Borja
and Maria Luisa de Aragon, whose ducal descendant celebrated the third centenary of Qmxote
in Pedrola in 1905.
9. Could have done otherwise 265
QUESTION. Suppose that a sees to it that Q; does it follow that a might have
refrained from seeing to it that Q in the sense that there is a co-instantial
alternative at which a refrains from seeing to it that Q?
STIT VERSION. Does [a stit: Q] imply Might-have-been:[a stit: ~[a stit: Q}}?
This is a tricky question. Its answer depends, of all things, on whether or not
there are busy choosers, 7C.5.
UPSHOT WITHOUT BUSY CHOOSERS. If there are no busy choosers, the impli-
cation is valid.
PROOF. Lettered steps are keyed to Figure 9.3. (a) Grant [a stit: Q] true at
mo /ho, and let WQ be the witness in question, (b) Let mi be some moment in
i(m0) a* which [a stit: Q] is settled true, and which has the further feature that
it "has a closest witness" in the sense that there is a witness, w\, to [a stit:
Q] at mi such that between wi and i(mo) there are no further witnesses to the
settled truth of [a stit: Q} at any moment in i(mo)- Because there are no busy
choosers, m1 must exist. By the negative condition, (c) there is a moment, m2,
lying in i(mo) and above wi at which Q is not settled true. We claim that w\
is a witness to [a stit: ~ a stit: Q}} at 7712- The negative condition is easy: [a
stit: Q] at mi is just what is required. Suppose, for reductio, that the positive
condition failed; that is, (d) suppose that [a stit: Q] were true at some moment
m3 Choice^im?.}, with witness at, say, w2. Where could w2 be? Because
both wi and w2 precede ma, by no backward branching, either wi < w2 or w2 ^
wi. ( e i ) The first alternative is impossible, because w\ is a closest witness. (e2)
The second alternative is equally impossible, because then the positive condition
of w2 witnessing [a stit: Q] at ma would conflict with the failure of Q to be
settled true at m2. (/) It cannot therefore be gainsaid that [a stit: ~[a stit:
Q}} is settled true at m2, which is a co-instantial alternative to m0. Therefore,
Might-have-been:[a stit: ~[a stit: Q}] holds at mo/h0.
266 Applications of the achievement stit
EXAMPLE WITHOUT BUSY CHOOSERS. The journey to the ducal couple's castle
portrayed earlier in Figure 9.2 illustrates the subtle difference between "could
have refrained" and "might have refrained." In that adventure the travelers
might have refrained from arriving before sunset, though it was false that they
could have refrained from doing so, since they were not busy choosers. And
thus it is: The moment at which the wayfarers might have departed from point
B is an excellent witness to the truth of [a stit: ~[a stit: Q}} at a moment
co-instantial with the one in question, thus verifying "might have refrained."
Since, however, that moment is not in the past of the moment of their actual
arrival at the castle, it does not help verify "could have refrained."
UPSHOT WITH BUSY CHOOSERS. In the presence of busy choosers and witness
by chains, the implication fails.
are two possible choices, left and right, each of which leads immediately to a
choice point for a of exactly the same kind as before. We suppose the temporal
distance between choices for a is halving, and that each entire denumerable
historical series of choice points for a and (3 approaches a unique member of
i(mo)'i and we assign Q settled false at such members of i(ma)- The moment,
mo, is the one lying above the right side of the first choice for a. We claim
first for Figure 9.4 that each choice point, wi, for a witnesses the settled truth
of [a stit: Q] at the moment, mi, in which the history belonging to the right-
hand possible choice for a at w1 intersects i(mo)- The positive condition is
easy, since we assigned Q settled true at mi and since there is but a single
history contained in that possible choice. The negative condition is satisfied by
our having assigned Q settled false at those members of i(mo) approached by a
denumerable historical series of choice points; one (and indeed many) of those
members of i(mo) must be properly later than w\. As a special case, [a stit: Q]
is true at mo- We claim next that nowhere in i(mo} is it settled true that [a
stit: ~[a stit: Q}}. This is obvious for the members of i(mo), such as mo itself,
that lie above some right-hand possible choice for a. Now suppose for reductio
that [a stit: ~[a stit: Q}] is settled true at some moment m-2 in Z( m o ) that is
approached by a denumerable series of choice points. There must then be a
witness, and since we are allowing witness by chains, 8G.4, let the chain be 02,
as indicated in Figure 9.4. The positive condition implies that ~[a stit: Q] be
settled true at every moment in i(mo) that is choice equivalent to 7712 for a at
C2. Choose some member, W2, of 02. Properly between W2 and m2 there will be
a choice point, w3, for (3, and properly after w3, there will be a choice point, 104,
for a that is not in the past of m,2 (w^ -f. m-i). Let 777.4 be the member of i(mo)
lying above the right-hand possible choice for a at 104. The critical point is
that 771,4 is choice equivalent to 7712 for a at 02, sinceby the no choice between
undivided histories principleno choice for a in c? distinguishes 7774 from 7712-
So, since [a stit: Q] is certainly settled true at 7714, we have a contradiction.
268 Applications of the achievement stit
Thus [a stit: Q] is settled true at mo but Might-have-been:[a stit: ~[a stit: Q}}
is not, so that the implication fails.
9-1 EXAMPLE WITH BUSY CHOOSERS. (The mirror game.) The Knight of
the Mournful Countenance and the Knight of the Mirrors engage in serious
combat. (We thank S. Sterrett for supplying a basic idea of this game.) They
play a busy game that begins at noon and ends at sunsetat which time the
vanquished is to remain entirely at the mercy of the victor. (Thomsen 1990
reminds us that "there is nothing more serious than play," p. 171.) Some plays
of the game consist of infinitely many moves, which our knights-errant manage
by halving the time spent on each successive move. Though busy, it is still a
simple game. Don Quixote has the first move. On his turn the Manchegan
has the following choice: Press on or retire. If he retires then at sunset he is
the vanquished. If he decides to press on, it is the turn of the Knight of the
Mirrors, whose move always consists in selecting a phantasmic replica of one
of two giants for the Manchegan to face: either Pandafilando of the Malignant
Eye, or Briareus with many arms, each later phantasm being, in appearance,
half as tall as its predecessor. Whatever he of the Mirrors selects, the next turn
belongs again to the Knight of the Mournful Countenance. At sunset there are
but two relevant possibilities: Either Don Quixote has retired, in which case he
is the vanquished, or he has succeeded in facing some denumerable sequence of
phantasms, in which case he is the victor. The curious fact to be illustrated
is this: It is possible for Don Quixote to retire from the contest, but it is not
possible for him to refrain from retiring. Contrary to our untutored intuitions,
not even an entire chain of choices to press on, right up to sunset, can witness
that Don Quixote refrains from retiring; for such a chain of choices does not
establish that it was entirely up to him that he persevere. The choices that in
fact were made by the Knight of the Mournful Countenance bestow no hard
information about "what he would have chosen" had the Knight of the Mirrors
confronted him with an unrelenting sequence of replicas of Pandafilando of the
Malignant Eye. Quixote's famous victory, however, does allow him to wrest
from the fallen Knight of the Mirrors the confession that "the torn and dirty
shoe of Lady Dulcinea of El Toboso is better than the ill-combed though clean
beard of Casildea." Trust the concreteness of this fable, we beseech you, only to
the extent that you understand its structural properties. Symmetrically, if you
think our chronicle is wrong, please try to find an alternative rigorous account of
witnessing, refraining, and so on, and not just another picturesque story without
a precisely described structure.
from doing it ([a stit: Q] does not imply Might-have-been:[a stit: ~[a stit: Q}]}-
If you look at any picture, however, it certainly seems as if whenever there is a
witness, w, for [a stit: Q], at that witness a has available a strategy he could
follow that, provided a never deviated from that strategy, would guarantee his
not-doing. That is, [a stit: Q] plausibly implies that there was (in the past)
available to a a strategy guaranteeing ~[a stit: Q}. The intuitive idea is that a
could avoid seeing to it that Q by shooting for a "counter," as called for by the
negative condition, at which Q will not be settled true. This intuition works;
but rigorously establishing the fact (Theorem 13-28) turns out to be less trivial
than one might suppose, requiring as it does much of chapter 13.
In the end we are led to see the sharp difference between (i) something coming
about for which there was a guaranteeing choice by a, and (ii) something coming
about for which there was (in the past) available to a a guaranteeing strategy.
Since past choices are a matter of fact, in case (i) we are entitled to a "because":
The thing came about because of a choice by a. In case (iz), however, we are
not entitled to a "because." since we have said that the strategy was available to
a and that a chose in accord with the strategy, without saying anything about
the agent following the strategy. Even though this book sometimes lapses into
the language of "following," one must recognize that the concepts of the theory
of agents and choices in branching time are too austere to support such loose
talk. But more of this in chapter 13.
9J Summary
The proposition
If an agent is morally responsible for doing something, then the , ,
agent could have done otherwise
conceals complex connections among actions, moral responsibility, and alterna-
tives open to an agent. We simplify by dividing the proposition in two:
If an agent is morally responsible for doing something, then the ,_,
agent did it.
If an agent did something, then the agent could have done oth- /o\
erwise.
This division isolates the idea of doing something, validating the use of a logic
of agency. We use stit theory to clarify proposition (3).
CONJECTURE. Proposition (3) can be disambiguated by means of a logic of
agents who make choices against a background of branching time.
means "could have refrained" then it fails for the achievement stit, but holds
for dstit. If "could have done otherwise" is taken as "might have refrained,"
then without busy choosers the implication holds, but with busy choosers it
fails. If "could have done otherwise" is taken as "had available a strategy for
not doing," then busy choosers or not, the implication holds.
10
The goal of this chapter is to inspect some structural aspects of multiple and
joint agency, a task sufficiently complex to give pause to the three inseparables,
Aramis, Athos, and Porthos.*
First, in 10B, we treat complex nestings of stits involving distinct agents.
The discussion is driven by the logical impossibility of "a sees to it that 0 sees
to it that Q" in the technical sense, even though that makes sense in everyday
language. Of special utility are the concepts of "forced choice," of the creation
of deontic states, and of probabilities. Second, in 10C, joint agency, both
plain and strict (every participant is essential), is given a rigorous treatment
in BT + I + AC theory. A central theorem is that strict joint agency is itself
agentive. In the final section, 10D, we combine these two perspectives, looking
briefly at other-agent joint agentives. Throughout this chapter we use "stit" for
the achievement stit, 8G.3.
As elsewhere in this book, even minimal progress toward the goal of this
chapter has required a variety of simplifications, (i) As in stit theory generally,
we totally avoid the reification of actions, and (ii) we minimize reference to
intention in order to concentrate on causal structure, (in) Of relevant notions,
we omit stits that are based on "witness by chains," 8G.4, and (iv) the deliber-
ative stit, 8G.l. (v) Also, we omit concepts requiring the chapter 13 notion of
strategies, and (vi) we do not consider the evident fact that agents interact in
space-time, a topic yet to be studied. Finally, (vii) for the scope of this chapter,
for the sake of expository simplicity we assume no busy choosers in the sense
of Def. 14. In exchange, although we directly employ only the achievement
stit, we have in this chapter limited ourselves to ideas and applications that, we
think, work equally well for either the achievement or the deliberative dstit.
We refer to Tuomela 1989a and 1989b for an alternative methodology that,
in contrast to stit theory, freely permits one to (i) reify actions and (ii) refer
*With the kind permission of Baltzer Science Publishers, this chapter draws on Belnap and
Perloff 1993 As in the case of chapter 9, you will observe that we sometimes let the source
of our examples influence our mode of speech.
271
272 Applications of the achievement stit
to intentions. Those articles also provide access to some earlier studies of joint
agency.
We omit proofs. One can use Fact 10-1 (w), the sufficient condition for unset-
tledness, as a way of seeing that the following is inconsistent: [a stit: (Q & ~[a
stit: Q})}. The Red Duke is subtle; he is, however, not so subtle that he could
see to it that M. Bonacieux disappeared while at the same time he, the cardinal,
did not see to that fact. (Of course he could so act that that is what people
would say.)1 As a reminder of the imperative content thesis Thesis 4 that we
discussed at length in chapter 4, reflect on the letter from Mme Bonacieux to
d'Artagnan:
Be in Saint-Cloud at ten o'clock tomorrow night, across the street , .
from the bungalow at the corner of Monsieur d'Estrees's house.
Although it might seem that the content of the imperative construction of (1) is a
non-agentive describing d'Artagnan's whereabouts, according to the imperative
content thesis, that is mere appearance. In truth the content of (1) is well-
regimented by the explicit agentive,
[d'Artagnan stit: d'Artagnan is in Saint-Cloud ...].
1
We remark that the locution "Q &: ~[a stit: Q]" plays an important role in the complete-
ness proof of chapter 16.
10. Multiple and joint agency 273
And when d'Artagnan's father says of the old yellow horse "Never sell him,"
the content of his imperative for each moment may appear to be a non-agentive
that merely denies agency to d'Artagnan,
~[d'Artagnan stit: d'Artagnan sells the yellow horse].
The imperative content thesis, however, drives us to take as agentive the content
of the imperative "Never sell him":
[d'Artagnan stit: ^[d'Artagnan stit: d'Artagnan sells the yellow (2)
horse]].
That is, the father charges d'Artagnan to deny himself the agencyto refrain,
where according to stit theory, refraining from seeing to it that Q is always
definable as [a silt: ~[a stit: Q]].
We shall be dealing also with some deontic statements, in connection with
which we remind the reader of the restricted complement thesis, Thesis 5. That
thesis requires (or suggests) that deontic statements have one of the following
forms.
Oblg;[ stit: Q]; is obligated to see to it that Q
Frbn:[ stit: Q]: is forbidden to see to it that Q
Perm:[ stit: Q]: is permitted to see to it that Q.
See 1B.2 and 2B.9, as well as chapter 11 and chapter 12.
PROOF. Assume the following for reductio: (a) [a stit: [0 stit: Q}} is settled true
at mi with w as witness, and with m2 a "counter" as required for the negative
condition, so that (b) [/3 stit: Q} is not settled true at m^- By independence of
agents, there must be an m^ such that both (c) m-i = 7713 and (d) 7713 =^ m?..
By (a), (c), and the second witness-identity lemma (Fact 10-1), it must be that
(e) w is witness for [a stit: [/3 stit: Q}} at m^. By (a) and (c) we must, by the
positive condition, have (/) [0 stit: Q] settled true at 7713let u>i be the witness
for this. From (e), (/), and the witness-identity lemma, we infer (g) w\ < w.
So (d) and (g) imply, by backward monotony (Fact 10-1), that (h) ms =t mg.
But then the second witness-identity lemma with (/) and (h) gives that [0 stit:
Q] must be settled true at m^; which contradicts (b) and completes the proof.
For a picture, think of Figure 10.1 as a diagram of a witnessing moment, mo, in
which columns represent possible choices for a, and rows picture the choices for
0 (see 10C.2 for the attribution of this picture to von Neumann). If [a stit: [0
stit: Q}}, then, where A = [0 stit: Q], A must fill some "choice-column" for a in
10. Multiple and joint agency 275
mo. But then, because A represents a stit by /?, whenever A appears anywhere
in a "choice-row" for /? inTOO,it must fill that row. So A must fill the entire
diagram ofTOO-If so, then A is by definition settled true atTOQ.This contradicts
the principle that, by the negative condition, stit statements are never settled
true at their witness.
by her pronouncement, that an obligation existed where there was none before.
Specifically, she saw to it that d'Artagnan was obligated to retrieve the diamond
tags.
We mean, incidentally, that really giving an order really does create an obli-
gation, not just that the speaker so intends. In making this distinction with
absolute sharpness, we separate ourselves from Searle 1965, who appears indif-
ferently to use the language of intention in his "essential condition" but not to
use it in his "essential rule" for promises. Having read Hamblin 1987, we also
reject the claim of Searle and Vanderveken 1985 that "the point of orders and
commands is to try to get people to do things" (p. 14). Instead the point of an
order is to create an obligation. Nor does advice have the "causing of action" as
its point. Aramis is plainly right that "as a general rule, people ask for advice
only in order not to follow it; or, if they do follow it, in order to have someone to
blame for giving it." What needs telling is a better story of what deontic states
agents really see to when they use not only orders and commands, but also
advice, requests, invitations, promises, ..., and indeed assertions and questions.
Structural features of stit theory accordingly lead us to the following as a
preliminary interpretation of (3).
That, however, won't do, since (3) is a "success" locution. It implies its com-
plement, "d'Artagnan retrieves her diamond tags," whereas the form (10) does
not imply [/3 stit: Q}. In a way that is precisely the point: Anne can stit the
obligation, but not that d'Artagnan carries it out. We need to add that as a
separate conjunct:
Now (11) shows on its face that it involves agency by Anne (as well as agency
by d'Artagnan) without being agentive for Anne. 2 Such cases are many and
important. When in (4) Jussac orders Biscarat to surrender and Biscarat replies,
"You're my commander and I must obey you," he recognizes that his commander
has seen to an obligation, [a stit: [/3 stit: Q}} is not the appropriate reading
here because Jussac did not guarantee Biscarat's surrender. What did Jussac
accomplish with his order? Jussac saw to the creation of an obligation:
Jussac created the obligation to surrender, but it was Biscarat who surrendered.
Regarding (8), [a stit: ~[/3 stit: Q]\ is not appropriate because when Mme
Bonacieux sees to it that d'Artagnan does not follow her she does not prevent
him from following her; rather she sees to it that a prohibition exists where none
existed previously. The form
2
If it were possible for us to use the theory of agents and choices in branching time to
represent that d'Artagnan retrieved the tags because of Anne's order, we would do so. We
cannot, since the "because" in question seems to be an intentional matter, and so falls among
the many aspects of agency for which we offer no theory.
10. Multiple and joint agency 277
One final point before leaving the topic. Notice that if the equivalence, Frbn:[a
stit: Q] <-> Oblg:[a stit: ~[a stit: Q]], is correct, then by substitution in the
complement, (13) is equivalent to
[Mme Bonacieux stit: 06/(?:[d'Artagnan stit: ~[d'Artagnan stit:
(16)
d'Artagnan follows]]].
We think this is right. If so, there is confirmation of the deontic equivalences
worked out in 2B.9.
draper both had a role in the transition to the outcome, l 'M. Bonacieux spies
for the cardinal."
10C.1 Preliminaries
We start with English grammar. Constituent imperatives (see 1C) are embed-
ded imperatives, analogous to embedded declaratives or embedded interroga-
tives. Their content, like that of agentive declaratives, can always be repre-
sented by stit sentences. An imperative, whether stand-alone or constituent,
can have a collective term in subject position, as can an agentive declarative:
M. de Treville announces: "I won't have my musketeers going to f^\
low taverns."
The four friends scraped together nine or ten pistoles. (23)
Example (22) might well be taken "distributively," and as analyzable in terms
of stit sentences with subjects taken to denote a single agent (we call these
"singular stits"), perhaps the subjects being individual variables bound by a
quantifier. On a plausible reading, M. de Treville requires each musketeer to
see to it that he does not go to low taverns. Examples like (23), however, drive
us to widen the grammar of the language of agency. Here it is evidently the four
friends "taken collectively" who succeeded in raising nine or ten pistoles; it is
not something that each of them does. We cannot usefully represent (23) with
only singular stits. We need to add to our formal grammar of singular stits the
category of a "joint stit." Collectives can be represented by mereological sums
as in Massey 1976; here we choose to represent collectives by sets.
This choice limits applicability; the proposed apparatus cannot treat cases
in which collectives change their membership over time (see Parks 1972), nor
cases in which their membership is history-dependent. The limitation is for ex-
pository convenience only, and could be removed by using the language ML" of
Bressan 1972 that we mentioned in 7C.l. We would realize Bressan's "cases"
as moment-history pairs. In that language we would first represent Agent as an
absolute concept, so that Agente would by definition be the extensionalization
of Agent. Collectives of agents would be represented as properties F (possi-
bly extensional, possibly not; possibly contingent, possibly not) such that F C
282 Applications of the achievement stit
one from the other as follows: Given a possible choice for each member of F, we
define a possible choice for F as a whole to be the intersection or "combination"
of all the individual possible choices. The "independence of agents" condition
guarantees that such a combination always exists.
The image we have in mind is due to von Neumann. Let an outcome be
dependent on the choices of two agents a and (3. Von Neumann represents this
graphically as follows. All the outcomes are arranged in a rectangular grid.
Agent a can pick the row and agent (3 can pick the column. What happens is
indicated in the intersection of the row picked by a and the column picked by
B. "Independence of agents" just says that some outcome is indicated at each
intersection of a row and a column. For example, if there are three rows (choices
for a) and four columns (choices for /?), then there are twelve possible outcomes
for their combined choice.
With the help of the concept of choice equivalence for sets of agents, we can
state the truth conditions for [F stit: Q}. We say that
10-4 DEFINITION. (Joint stit) [F stit: Q] is true at m/h iff there is a choice
point wa "witness to [F stit: Q] at TO"satisfying the following conditions
(compare the definition of the achievement stit in 8G.3): Agency:
Agent. Priority: w < m. Positive: Q is settled true at each m\ such that m
Negative: Q is not settled true at some momenta "counter"on the
horizon from w at i(m).
If we apply this definition to (24), it tells us that the raising of the pistoles was
due to a simultaneous antecedent choice of the four friends. By so much the
definition makes (24) a good approximation to (23). Furthermore, Definition
10-4 is good logic:
10-5 FACT. (Carryover from singular to joint stits) Results or analyses con-
cerning singular agents established without the use of the postulate of the in-
dependence of agents also hold for joint agents.
Results not transferring include those expressed with the help of "a ^ /3" when
these rely on the independence of agents. The point is that the possible choices
for a and {3 at w will be independent if a ^ /?, but this is by no means true of
the possible choices for FI and F2 when FI ^ F2. The obvious reason is that
non-identity between the two collectives does not prohibit their having members
in common.
By Fact 10-5 we mean for example that any implication or non-implication
that holds between singular stit sentences with just a also holds between the
joint stit sentences that result when F is substituted throughout for a. For
instance, [F stit: [F stit: Q}\ is equivalent to [F stit: Q], and ~[F stit: Q] is not
in general agentive for F (i.e., it is not in general equivalent to [F stit: ~[F stit:
Q}})-
We can use joint agentives to express the independence of agents, provided
we have the help of the following version of "ability" (this generalizes to a set
of agents the ability concept of 8G.3).
284 Applications of the achievement stit
been, for example, that d' Artagnan was not essential in raising the pistoles in
the sense that
[The three musketeers stit: the four friends have nine or ten pis- (26)
toles],
where, as everyone knows, d' Artagnan E the three musketeers = {Athos, Por-
thos, Aramis}. Here is the easy fact about [F stit: Q] that informs us of this
possibility:
10-8 FACT. (Weakening of ]oint stits) Given F1 C F2 C Agent: if [F1 stit: Q]
then [F2 stit: Q}.
That is, joint stits are closed under "weakening" by the addition of further
agents.
We need to define some related properties of agents in two versions before we
can go further. The first relativizes the concepts to F and Q. The second drops
the F, relativizing only to Q. The point is to be careful as to which concept is
at stake. (We remark that although the terminology to be introduced seems apt
in context, one needs to be sensitive to the considerations mentioned in 10C.l
concerning sequential choices.)
We will soon define "strictly stit" (an expression we introduce for joint agency
when each of the agents is essential) by just this sentence, but first we must
face a difficulty: Only the first part of (27) has an agentive form; the second
conjunct is instead a denial of agency. So the whole may not itself be agentive!
The difficulty is, however, easily overcome. In fact (27) is equivalent to each of
the following.
286 Applications of the achievement stit
The equivalence of (27) and (29) establishes that (27) is agentive for F in spite
of the fact that a conjunct of (27) is a denial of agency. We state this as a
10-11 THEOREM. (No inessential members) The sentences (27), (28), and (29)
are mutually equivalent. In other words, where we let
PROOF. Suppose, for reductio, that O / = F1 C F and [F1 stit: [F sstit: Q]] at
m1/h with witness w and counter m3. Choose any m2 such that m1 ={F1 m2;
then both [F stit: Q] and NIM are settled true at m2- So Q is settled true at
all such m2, and we have the positive condition for [F1 stit: Q] at m1 to be
witnessed by w. If, then, Q is not settled true everywhere on the horizon from
w at i(m1)i we shall have the negative condition as well, and w will witness the
settled truth of [F1 stit: Q] at m1 contrary to the settled truth of NIM there.
We obtain the desired unsettledness of Q from the counter at m3 as follows. We
know that [F sstit: Q] is not settled true at m3, so that either ~[F stit: Q] or
~NIM is settled true at m3. Since both [F stit: Q] and NIM are supposed settled
true at m1, in either case we can use the sufficient condition for unsettledness
of Fact 10-1 to infer that there is a moment on the horizon from w at i ( m 1 ) on
which Q is not settled true, as required.
The following is then an easy calculation.
10-15 FACT. (S4 property for strict stits) [F sstit: Q] is strictly agentive for
F, that is, is equivalent to [F sstit: [F sstit: Q]}.
One direction comes from the fact that quite generally [F sstit: Q] implies Q.
The other direction is a consequence of Theorem 10-11 and Lemma 10-14.
The "S4" property that we just proved of sstit does not give us copious
information about the behavior of strict seeing to it that; although it is doubtless
a beginning, there is much that we do not know.
10-16 QUESTION. (Modal properties of strict stits) Suppose we treat sstit as
a modal operator. What illuminating properties does it have? What about its
modal interactions with plain stit? And so on.
288 Applications of the achievement stit
[The queen stit: Oblg: [the ladies sstit: the ladies bring the diamond tags
from the Louvre to the ball]].
10. Multiple and joint agency 289
Though the matter is uncertain, surely it is plausible that the content of the
queen's order has only the plain form [F stit: Q], so that it is quite consistent
with the content of that order that Mme de Sable should be inessential. If Anne
of Austria really wants all her ladies to be involved, she should explicitly say so,
using something with the content [F sstit: Q}. By Theorem 10-11 she will have
the satisfaction of knowing that the content of her order is indeed agentive.
We think that the outcome is the same for other deontics with joint subjects
such as permissions and prohibitions: Although there is no logical reason not
to permit or forbid a collective to see to it that Q in the strict sense, often the
plain sense is more likely to catch what is wanted. For example,
Although de Treville does not forbid the four friends to spend a
total in excess of 6,000 livres on their equipment for the siege of (32)
La Rochelle, he advises them not to do so.
This example describes a prohibition and some advice. By the restricted com-
plement thesis, each should have an agentive complement. The content of the
advice is that the friends should refrain from spending more that 6,000 livres,
that is, that
The four friends see to it that it is false that the four friends see (33)
to it that the four friends spend in excess of 6,000 livres.
It would follow from the stit deontic equivalences, Ax. Cone. 2, that the content
of the prohibition (the one never issued by Treville), when reconstrued as an
obligation, is exactly the same as the content of the advice, that is, (33). Perhaps
this logical parallelism is why (32) sounds so eminently intelligible.
So now the question is, with what sort of stit should we approximate the
see-to-it-thats that occur in (33)? Let F = the four friends, and let Q <-> the
four friends spend in excess of 6,000 livres. Do we want (i) [F stit: ~[F stit: Q]],
or (ii) [F sstit: ~[F stit: Q}}, or (iii) [F stit: ~[F sstit: Q]], or (iv) [F sstit: ~[F
sstit: Q]]? One can use Lemma 10-14 to show that (iv) is equivalent to (in);
and neither is tolerable. Suppose Porthos in his vanity chooses to spend over
6,000 livres, and thus alone guarantees the truth of the complement, so that he
alone guarantees that the other three friends are inessential. It seems clear that
this behavior counts as not following Treville's advice to refrain, so that (33) is
false on that story; but the candidate (in) is true and so cannot be an accurate
representation of (33). We are left with (i) and (ii). The latter of course mixes
plain and strict stits, but in the absence of a more thorough investigation, both
logical and conceptual, we ought not say more.
Our last example concerns a permission.
The four friends allowed their servants Planchet, Grimaud, Mousqueton,
and Bazin to finish the Beaugency wine.
It seems implausible that the content of this permission should be represented
by a strict stit. Instead
[The four friends stit: Perm: [The servants stit: the servants finish the
Beaugency]]
290 Applications of the achievement stit
seems a more likely normal form. On this version the four friends permit that
the Beaugency is finished by the choice of Planchet, Grimaud, and Mousqueton
alone, Bazin having antecedently gone off to study his theology.
and
[The four friends stit:
(37)
In (36) the obligation is jointly on Planchet and Fourreau as a pair, but the
execution is supposed to be by one of them as an individual. This complex
content, so subtly different from that of (34) and (35), can be clearly expressed
by an other-agent nested joint stit as in (37).
The lesson, easy to miss if you take these examples as little puzzles or tricks, is
that seriously applicable deontic logic needs other-agent nested joint agentives;
and it therefore needs to include a theory with at least the expressive power of
joint stits.
This page intentionally left blank
Part IV
Applications of the
deliberative stit
This page intentionally left blank
11
Conditional obligation,
deontic paradoxes, and stit
which would allow detachment of the obligation Oblg:A in (2), does not belong
to the standard system of deontic logic. Consequently, the argument is invalid.1
*Paul Bartha is the author of this chapter. With the kind permission of Baltzer Science
Publishers, it is based on Bartha 1993.
1
F011esdal and Hilpinen 1971 sets out the axioms for the "standard system of deontic
logic," known as KD, in which Oblg:(B>A) D (BD Oblg:A) does not hold. This proposition
is usually rejected in any system which is based on "deontically perfect world" semantics; see
Hintikka 1971.
295
296 Applications of the deliberative stit
to emphasize analogies with dstit, we instead write [a cstit: A] for that operator.
We could add it as a primitive to our language and, as in 8G.2, provide the
following truth definition:
So defined, [a cstit: A] agrees with the semantic condition (3). One may also
verify that any one of [a cstit: A], Sett:A, and [a dstit: A] could be defined in
terms of the other two operators, since we have as valid sentences:
The Universally: ensures that the implication is strict. We will use something
similar to define the obligation operator. As indicated by the preceding dis-
cussion, however, we will restrict our attention to cases in which A is a dstit
sentence. Three questions arise:
i. What is the appropriate interpretation of the propositional constant S
that is to be added to our language?
ii. What should be the quantificational range of the modal operator?
Hi. How should we deal with the negation of A, where A is a stit sentence?
Ad (i). The intended meaning of S is unambitioussomething such as "there
is wrongdoing," or "there is a violation of the rules." The connection between
obligation and S should be unproblematic. Even though the chapter sometimes
speaks of S as a sanction, we are not entitled to interpret it as punishment or
censure, which has no logical connection to obligation.
So as to avoid confusing the obligations of different agents, S should be in-
dexed by an agent. Then we can interpret Sa as "a does something wrong." We
suppress the subscript, however, since throughout this chapter we will only be
concerned with one agent's obligations at a time. (We might also have added a
second index for the authority (individuals or perhaps institutions) whose rules
are violated. Again, this added complexity is not required at present, but in
11H we briefly discuss the suggestion that obligations should always be thought
of as seen to (as "stitted") by an authority.)
Ad ( i i ) . The scope of the modal operator will not be all of Tree, but only
all histories through a given moment; that is, we will replace Universally: with
Sett:. There are two reasons for this. First, "failing to see to it that A" may
be a case of wrongdoing at some moments, but not at others. We do not want
to limit ourselves to obligations that remain constant for all time. Universally:
is unsuitable. Second, the truth of [a dstit: A] involves consideration of all
histories through a moment (by the negative condition), so it is not unreasonable
to suppose that the truth of Oblg:[a dstit: A] does so as well. (In 11H, we will
discuss some difficulties that arise from the choice of Sett:. Another alternative
is to use the Chellas stit, [a cstit: ]. It turns out that this is not a good
choice, since the resulting definition of obligation makes the sentence [a dstit:
A] D Oblg:[a dstit: A] valid in all models.)
Ad (iii). With regard to the negation of [a dstit: A], there are three evident
possibilities, leading to three alternative definitions of Oblg:[a dstit: A]:
300 Applications of the deliberative stit
(6) and (7) are harsher definitions than (5), in the sense that if Oblg:[a dstit: A]
holds under (5), then the obligation also holds for both (6) and (7). In fact, (6)
and (7) must be rejected as too harsh. On either definition, it turns out that
Oblg:[a dstit: A] holds for all a if A is any tautology or contradiction. In what
follows, we work with (5): a is obligated to see to it that A just in case it is
settled that if a doesn't see to it that A then there is wrongdoing.
The basic picture for Oblg:[a dstit: A], then, is given in Figure 11.1. Although
its interpretation is much like that of the figures for the achievement stit in
2B, for this and later figures featuring the deliberative stit, note the following
conventions, (i) Here, in contrast to, for example, 2B, each sentence in a
diagram represents its truth at a moment-history pair rather than its settled
truth at a moment, ( i i ) Here we allow that different histories coming out of the
same choice box need not necessarily split at m. They may split later. Also
observe that in this chapter we sometimes refer to a possible choice as a "choice
box."
and
Note that by nesting the agentive modality, we avoid the problem of "negated
actions." Since we also have, symmetrically,
3
Recall from 2B.6 that the "refref" equivalence, Ax. Conc. 1, between seeing to it that
and refraining from refraining, unproblematic for dstit, is a more delicate matter for the
achievement stit. See the proofs of Lemma 15-15 and Lemma 15-16.
302 Applications of the deliberative stit
and it is easy to see that (12) is a fact about agency qua dstit (but see 2B.4
for the opposing verdict for the achievement stit).
PROOF. Consult Figure 11.2. To see (12), suppose [a dstit: (ADB)} and [a
dstit: A] hold at m/h on model m. Then ADB and A, and hence B, hold at
all m/h2 for h2 =a h. Further, m, m/h \= [a dstit: A D B ] requires a counter
m/h3 where ADB is false, and thus where B is false. The positive and negative
conditions for [a dstit: B] at m/h are both satisfied.
From (12), (11) follows easily.
PROOF. Suppose Oblg:(a dstit: (ADB)} and Oblg:[a dstit: A] hold at m/h. If
~[a dstit: B] holds at any m/h2, then either ~[a dstit: A] or ~[a dstit: (AD
B)] holds there, by (12). But by definition (5), S must then hold at m/h 2 . This
proves Sett:(~[a dstit: B]DS), that is, Oblg:[a dstit:
Anderson 1956 adopted the axiom that the sanction should be contingent.
We do not believe we can accept the corresponding axiom, ~Sett:S. We are
not entitled to assume that no-good-choice moments can be ruled out a priori.
Genuinely conflicting obligations seem at least possible, especially if the author-
ity issuing obligations is unreasonable. In any case, the valid analog of (KD2),
stated without proof, is
11C Completeness
For this section only, we reduce our language (and our definition of structure,
model, truth in a model, etc.) by eliminating the operators Was: (past) and
Will: (future), and by restricting Agent to contain only one agent, a. The
language still contains the constant S. A completeness result is stated without
proof. Our axiomatization is based on that of 17A; structures and models are
of kind BT + AC.
Recalling abbreviation (4), [a cstit: A] <=^ Sett:A V [a dstit: A], we take as
axioms for a system, SAo, all instances of truth-functional tautologies as well
as the following schemata:
304 Applications of the deliberative stit
In English:
4
The condition is intended only as sufficient. On the side of necessity, Davey 1999 sub-
stantially refines the analysis of this section.
11. Conditional obligation, deontic paradoxes, and stit 305
Therefore, (16)
This says that it's settled that a doesn't see to it that B is false, that is, that
a can't see to it that B is false. Condition (C) is one way to formalize the
assumption that B is a circumstance, for it captures the idea that a cannot
prevent B from being true. (Castaneda's idea of a circumstance is different,
since in his system circumstances sometimes are within the agent's control.)
Assuming (C) makes the argument (15) go through. On the other hand, if (C)
is false, the argument tends in general to fail. (See Davey 1999 for the more
detailed account.)
PROOF. Consult Figure 11.3. We prove (15) with the help of the added as-
sumption (C). We assume (15)(a) Oblg:[a dstit: (B D [a dstit: A ] ) ] , (15) (b)
B, and (C) Sett:(~[a dstit: ~B]) are true at m/h in model m. By the first
assumption, if h2 is any history through m, then by definition (5),
that is, that (15)(c) Oblg:[a dstit: A] is true at m/h. The crucial thing to notice
is that from B and Sett:(~[a dstit: ~B]) at m/h, it follows that for each history
h2 through TO, there is at least one choice equivalent history h3, such that B is
true at m/h3. Less formally, out of each choice box at TO comes at least one
history h3 on which B is true. For if this were not so, then [a dstit: ~B] would
be true at m/h2, with the counter at m/h, where B is true. This would violate
the fact that ~[a dstit: ~B] is settled true. Now if ~[a dstit: A] were true at
m/h2, it would also be true at m/h3. Then the conditional
306 Applications of the deliberative stit
and
In English, "You are obligated to see to it that if you saw to it that you behaved
badly, you see to it that you apologize." Now notice that the antecedent Was: a
dstit: B] satisfies the "circumstance condition,"
Agent a cannot see to the falsity of a statement about his past seeing-to's. The
reason is that either Sett:(Was:[a dstit: B]) or Sett:(~Was:[a dstit: B]) must
hold at any moment-history pair. In the first case the positive condition for [a
dstit: ~Was:[a dstit: B]] fails, and in the second case, the negative condition
fails.
PROOF. Suppose that Was:[a dstit: B] is true at m/h. Then for some m2
< m, [a dstit: B] is true at m2/h. But then [a dstit: B] is true at m2/h2 for
every h2 with h2 =am2 h. Since every history h2 through m passes through m2
and satisfies h2 = am2 h by the no choice between undivided histories condition,
Post. 8, it follows that Was:[a dstit: B] is true at m/h2 for all h2 through
m. Thus, Was:[a dstit: B] at m/h implies Sett:(Was:[a dstit: B] ) at m/h. So
either Was:[a dstit: B] holds at all histories or at none, which is precisely the
"either-or" condition.5
Since Was: [a. dstit: B] is a circumstance, whenever it is true we can detach the
obligation Oblg:[a dstit: A]. It seems to us that many conditional obligations
have the form of (20), even when the antecedent seems to be present tensed.
Consider the obligation: "It's your duty to apologize if you behave badly at
the party." What is the tense of "behaving badly" relative to "apologizing"?
It must be future, present (contemporaneous), or past. Taking the tense as
either future or present is not a reasonable interpretation of the duty, since
any apology given before or at the moment of behaving badly will hardly be
convincing. This leaves us with the same obligation as before:
It's your duty to see to it that you (see to it that you) apologize, if you
have (seen to it that you) behaved badly at the party.
The form (20) of this restatement of (19) is then the same as the form of (19)
itself.
Many conditional obligations whose antecedents express definite actions (ao-
rists in ancient Greek) have the form of (20). An important exception will be
discussed in S11H..
5
As noted in 8F.5, on the Prior-Thomason history-relative semantics for branching time,
it is not generally true that the past is settled in the semantic sense, which requires that for
an arbitrary sentence A, Sett: Was:A or Sett:~ Was:A must hold at each moment-history pair.
For instance, it may be that neither Was: Will:A nor ~ Was: Will:A is settled true. Further,
[a dstit: A] is never settled true, because of the counter. It is only the combination of the
past operator, Was:, with dstit that leads to a form that is bound to be settled true or settled
false. In this special case, the history-relative semantics agrees with the intuition that it is
always true that either Was:A or ~ Was:A is settled.
308 Applications of the deliberative stit
Translated back into English, this reads, "It's your duty to see to it that if Bob
saw to it that he behaved badly, you see to it that you admonish him." The only
difference between (22) and (20) is that a different agent is involved. The same
argument as before shows that Was:[B dstit: B] is a circumstance. Consequently,
it does not really matter if we express the obligation as an OD-statement or as
a D 0-statement.
Expression (22) is also a plausible way to formalize Alchourron's conditional
obligation to "revise the draft." In (15), we symbolized that obligation as
Since "it is raining" is a circumstance, we can detach from (23) the obligation
to bring an umbrella, Oblg:[a dstit: U], if it actually does rain. Once again, the
system S A is indifferent to whether we represent the obligation as (23) or (24),
just as English is indifferent between the corresponding English sentences.
"It is raining" is a present-tensed non-agentive sentence. Can present-tensed
agentive sentences also be circumstantial? It is an interesting fact about dstit
that the answer depends entirely on the agent. It turns out that the following
are true:
The first, (25), says that other agents' present doings are always circumstantial
for agent a. (The proof of Fact 10-2, which relies on the independence of agents,
Post. 9, can easily be adapted to dstit.) Expression (26) says that a's own
doings are never circumstantial for a, except when it is settled that a doesn't
see to something. This becomes important in the discussion of contrary-to-duty
obligations in 11G. Momentarily confining our attention to the case of different
agents, we have the result that SA is indifferent between
and
as ways of putting the obligation to see to it that you (a) show up to the meeting
if your boss (B) shows up to the meeting. In either case, we can detach Oblg:[a
dstit: A] provided [B dstit: B] holds.
We could also provide examples of future-tensed sentences satisfying the cir-
cumstance condition, such as "It will rain." The point is that circumstantiality
is not derivative of temporal ordering. It depends only on what it is possible
for a to see to at a given moment.
(b) But Arthur's doing C entails his doing the act of murdering a man a week
hence.
So, by (a), (b), and (27),
(c) Arthur is obligated to murder a man a week hence.
We want to reject (c); but (27), (a), and (b) all seem acceptable.
People usually attempt to resolve the paradox through analysis of tense,
agency, and the sense of entailment in (27) and (b). We can bring all these
considerations to bear in a precise way by using the dstit semantics. Let M
stand for "Arthur murders a man." Then "Arthur murders a man a week
hence" can be approximately translated as Will:[a dstit: M]. Therefore we can
represent (b) as [a dstit: C] D Will:[a dstit: M]. ("Arthur sees to it that C"
entails that in the future, Arthur sees to it that he murders a man.) As a first
stab at (27), we can try
Then (28) does not apply to (b), since Will:[a dstit: M] is not an agentive
sentence. Instead it is a future tensing of an agentive sentence. So (c) does not
follow.
Furthermore, we should note that (28) is, quite properly, invalid, as Figure
11.4 illustrates. This figure shows a moment m at which [a dstit: A] D [a dstit:
B] at h1, but not at h2 or h3. In such a situation, Oblg:[a dstit: A] D Oblg:[a
dstit: B] fails generally, and so at h1. Indeed, if it is possible to see to it that A
(bandaging a man) without seeing to it that B (killing a man), the conditional
Oblg:[a dstit: A] D Oblg:[a dstit: B] should be false.
When we strengthen the antecedent, we obtain a valid form of (27):
11. Conditional obligation, deontic paradoxes, and stit 311
where D stands for "a man is dead." Premiss (a) is translated as Oblg:[a dstit:
C]. Then Oblg:[a dstit: Will:D] does follow. But to assume (b') is to assume
that Arthur cannot bandage the injured man without seeing to it that he is dead
a week hence. (Perhaps the bandages are coated with poison.) It is doubtful
that Oblg:[a dstit: C] holds under these conditions. The paradox has lost its
sting, since
Oblg:[a dstit: C]
and
Oblg:[a dstit: WillD]
are equally objectionable.
A more reasonable approach is to represent (b) as
Sett:([a dstit: C] D Will:[a dstit: M]) (it's settled that seeing (b")
to it that C entails, in future, murdering a man.)
and to propose yet another version of (27):
(KDb)
11. Conditional obligation, deontic paradoxes, and stit 313
(KDc)
(KDd)
From (KDb), (KDc) and (KDd) together with the valid schema A D B /
Oblg:A D Oblg:B, we can infer Oblg:M, contradicting (KDa).
A possible response to this problem is to try representing the contrary-to-duty
obligation (b) as
(KDb')
This avoids inconsistency, but (KDb') then becomes a redundant premiss. For
Oblg:(M D G), and in fact Oblg:(M D A) for any sentence A, follows from
Oblg:~M. We could equally well add Oblg:(M D ~G). This suggests that
(KDb') is a bad way to represent the contrary-to-duty obligation (b).
The semantics of SA provides a way to represent the statements as a consis-
tent set without redundant obligations:
(SAa)
(SAb)
(SAc)
(SAd)
The first statement expresses a's obligation to refrain from murdering as well
as the fact that a has no obligation to murder. The second statement expresses
the contrary-to-duty obligation. Unlike the situation in the standard system,
(SAb) does not follow from (SAa).
Furthermore, we avoid a contradiction between (SAb) and (SAa) because we
cannot detach the obligation to murder gently, Oblg:[a dstit: G]. The reason is
that [a dstit: M] is not a circumstance in the sense of condition (C). Looking
again at (26), we see that it is never the case, given (SAd), that Sett:(~[a
dstit: ~[a dstit: M]] ). It is not settled that a cannot refrain from murdering;
the counter to (SAd) guarantees a the choice to refrain from murdering. The
obligation to murder gently always remains just a conditional obligation, as it
should.
Figure 11.6 illustrates the situation. On history h1, a violates both obligations
(SAa) and (SAb). On history h2, a violates only the obligation not to murder.
Finally, on h3,(where (SAd) is false) a satisfies both obligations. Clearly, Oblg:[a
dstit: G] does not hold, since it is possible to have ~[a dstit: G] without S (as
at h3).
Chisholm's own example in Chisholm 1963 is the following:
(a) It ought to be that a certain man go to the assistance of his neighbors.
(b) It ought to be that if he does go he tell them he is coming.
(c) If he does not go then he ought not to tell them he is coming.
314 Applications of the deliberative stit
(SAa)
(SAb)
(SAc)
(SAd)
Provided [a dstit: A] is possible (he can assist his neighbors), ~[a dstit: A]
is not a circumstance. So we may not detach the obligation to refrain from
telling his neighbors he will come, and there is no contradiction. There is only
a conditional, not a categorical, obligation not to tell. (If it is impossible for
him to assist his neighbors, ~[a dstit: A] is a circumstance and we can detach a
categorical obligation not to tell. Oblg:[a dstit: T] and Oblg:[a dstit: ~[a dstit:
T]] will conflict in this special case, but it can plausibly be argued that there is
a genuine conflict of obligations.)
(i) The system does not permit different obligations to hold at different his-
tories at the same moment. If an obligation holds at one moment-history pair
m/h, it is settled at m. Yet there are situations in which an agent's choices lead
to different obligations. The most obvious is making promises, as represented
in Figure 11.7. P stands for "a promises to call"; C stands for "a calls." It
seems that Oblg:[a dstit: C] should hold at m/h1, but not at m/h2.
On the proposed semantics, there are two possible ways around this difficulty.
One way is to argue that the obligation to call does not really come into force
until a time after the promise is made, that is, once it is settled that a has
made the promise. What really holds at h1 is the sentence P D Will:(Oblg:[a
dstit: C]). At some later time, the obligation to call will hold over the entire
moment. But it is not certainly satisfying to have to introduce extraneous tem-
poral considerations to resolve the problem. On the other hand, 5C suggests
that causal-temporal concerns may be essential to the idea of promising.
A second approach, following the treatment of contrary-to-duty obligations,
is to treat the obligation to call as the conditional obligation Oblg:[a dstit: ([a
dstit: P] D [a dstit: C] )]. Since promising is not a circumstance, the obligation
to call cannot be detached. The conditional obligation, which holds for the whole
moment, is only violated on histories where the agent makes a promise to call but
fails to call. This solution is also prima facie unconvincing, however, because
we have to introduce a conditional structure into what seems an unconditional
obligation created by the promise.
The difficulty derives from the fact that our definition of obligation quantifies
over all histories through a given moment. If we replace Sett: by [a cstit: ]
in (5). and define
then we allow for obligations that vary depending on a's choices. The problem
with this definition is that [a dstit: A] D Oblg:[a dstit: A] becomes valid because
[a dstit: A] = [a cstit: [a dstit: A]]. Given [a dstit: A], [a cstit: (~[a dstit:
A] D S)] follows classically. There might be some hope if we use a nonclassical
logic (for example, relevance logic), but this suggestion falls outside our present
focus.6
6
Bartha 1999 develops a refinement of the approach of this chapter that allows for obliga-
316 Applications of the deliberative stit
which the semantics proposed here seems unable to handle. A good example
(due to Castaneda in correspondence) is the following:
This fits the given form if a is Mary, A is "Mary does not open the office by
9:00 A.M.," and B is "Mary reports to the manager by 8:45 A.M." Imagine that
it is now 8:45, and Mary is staying home to care for a sick child. Since the
conditional obligation is in force, it seems that we should be able to detach the
obligation to call the manager: Oblg:[a dstit: B]. But we cannot do so on the
given semantics. The antecedent is not a circumstance in the sense of (C), since
Mary might still be able to make it to the office by 9:00.
We mention two stit possibilities that might help with the problem of condi-
tional obligations. The first is to take the content of an obligation to be more
like a strategy, in the sense of chapter 13, than it is like a stit. This suggestion
would be analogous to that of 5C for promising. The second is to consider the
impact of the proposal of Wansing 1998. Just as the analysis of this chapter
makes the obligated agent an intrinsic part of the obligation, Wansing suggests
analyzing obligation even further by making the authority or creating agent an
additional, inseparable part. Thus, two agents are needed for an obligation:
the agent or authority who creates the obligation, and the agent who is obli-
gated. The English normal forms of obligation and prohibition are then given
as follows.
obligates a2 to see to it that A
forbids a2 to see to it that A.
Then Wansing 1998 proposes that these be represented in terms of an agent-
indexed Andersonian sanction on a2, written Sa 2 , such as we mentioned in
11B.1. The representations would then be as follows.
We think that this proposal deserves thorough exploration, especially with re-
gard to the light that it might shed on conditional obligations.
318
12. Marcus and the problem of nested deontic modalities 319
With this example Marcus is challenging all deontic logicians. Deontic logic,
she says, could be a "misleading technical exercise" (p. 582) if it treats (1) as a
simple iteration of an impersonal "ought" and an impersonal "forbidden." That
is, the following symbols (2) won't do for (1):
Marcus, with conclusive arguments that we shall not repeat here, pins the blame
on a failure to keep separate the evaluative modal family headed by "ought to
be" and the prescriptive modal family headed by "ought to do" or "obligatory."
With reference to the prohibition expressed by "forbidden," let us leave aside
the "ought to be," turning instead to the "ought to do." Marcus endorses this
turn. She says
It seems to me that the least problematic reading of the deontic
operators is the one which fits the use of 'obligatory', 'forbidden',
'permitted', as they occur in connection with rule governed conduct.
Such a restriction to prescriptive use has the advantage of allowing
us to be clearer about the semantical interpretation of deontic state-
ments. (Marcus 1966, p. 581)
She is suggesting that we can best use deontic logic if we focus on the "ought
to do." In advancing our understanding of the "ought to do," Marcus, citing K.
Baier, suggests the following:
... we may take 'It is obligatory that A' (where 'A' is an appropriate
statement of the sort 'x does w at t') as meaning that A is entailed
by the set of rules and standards. (Marcus 1966, p. 582)
This is a rich suggestion. Much of the ensuing discussion can be construed as
following the suggestion with the help of the modal logic of agency. In doing
so we are going to concentrate on (1) to the exclusion of any more gripping
examples. If you start to fall asleep, please substitute an illustration filled with
dangerous and dreadful actions of the most excitingly electrifying sort. You
know, something like "buttering the toast in the bathroom."
This suggestion, itself unpretentious, is worth taking seriously and has in prac-
tice guided our work. In fact the suggestion is so helpful that it is easy to
overlook the following: When some philosopher says "x does w at t" we have
no idea over what the variable "w" is supposed to range. In other, more pre-
tentious, words, we don'teven after decades of the most arduous first-order
workhave an "ontology of actions" that is at once useful and rigorous. Nor,
of course, does Marcusthe consummate modal logiciansuppose that we do.
That stit theory is a modal logic is precisely why it can provide an illuminating
framework in which one isn't constantly having to worry about turning actions
into things to name or to quantify over. Stit theory suggests instead that the
grammar of a prescription is this:
It is obligatory that a see to it that Q,
or, in more idiomatic English,
a is obligated to see to it that Q. (5)
Three ideas lie behind this formulation, ideas that we re-mention here in order
to relate them to Marcus. The first is that the place of Q can meaningfully
be filled by any sentence whatsoever (stit complement thesis, Thesis 2). This
is the power of exactly that modal point of view championed and defended by
Marcus against various forms of "first orderism."
The second idea is that the place following "it is obligatory that" can be
occupied only by an agentive, of which the stit sentence is both paradigm and
normal form. This is the restricted complement thesis, Thesis 5. This thesis
gives technical bite to Marcus's distinction between the evaluative and the pre-
scriptive. That is, it forces us to keep the prescriptive "ought-to-do" separate
from the evaluative modal family.
The third idea is this. In contrast to the way everyone waffles when it comes
to an ontology of "actions," we are in firm possession of a clear and useful
semantic account of "a sees to it that Q" (8G.l). This account also shares
convictions with Marcus 1980. Two are paramount. The first is that obligation
and agency are bound up in complex ways with possibility. The second is that
you cannot get away from occasions and times. This is witnessed by the "t" in
(4), and made explicit in passages such as the following.
'Ought' is indexical in the sense that applications of principles on
given occasions project into the future. They concern bringing some-
thing about. (Marcus 1980, p. 135)
Together these ideas help suggest the employment of an independent account
of seeing to it that such as is given by the stit theory of this book. We do
not review these concepts, relying on chapter 8 and the summary in 6. We
do note that our use of branching time as a foundation presupposes answers to
difficult and important questions about actuality and possibility, answers that
are roundly challenged in Marcus 1977 and Marcus 1985/86. Even to begin
discussing any of these matters here, however, would mire us without hope of
12. Marcus and the problem of nested deontic modalities 321
The key point here is that this obvious and elementary truth cannot be said, or
said clearly, without the grammar of dstit. To be able to formulate this striking
definition of a prohibition as an obligation to refrain, you need to be able to
embed a negated dstit inside of a dstit. It is precisely this capability that seems
to be absent from an ontology of "actions." This power of embedding, which
belongs to the very idea of modality, needs to be exploited. You cannot exploit
it if you are ashamed of modalities. That is one of the great themes of Mar-
cus's work: Instead of treating modalities as an embarrassment in the fashion
repeatedly preached by Quine, you should use them for what they are worth,
which is plenty. Use them by themselves, with quantification, with identity, or
with whatever helps.
Combining Ax. Conc. 2 with Definition 12-2, we have the following symbol-
izations and equivalents.
Frbn:[a dstit: Pax]: a is forbidden to park on highway x.
-> Oblg:[a dstit: ~[a dstit: Pax]]: a is obligated to see to it that a does
not park on highway x.
<-> Sett:(~[a dstit: ~[a dstit: Pax]] D Sa): It is settled true that if a
does not see to it that a does not park on highway x, then Sa.
That's a lot of words. One can see further into the proposal by a combination
of cases and pictures. Take a particular concrete moment, TO, of concern to
a and to the rules. There are three major cases, the first two of which are
throwaways.
This is what Marcus urges to be quite another thing. To deny that such a
circumstance could obtain is, as she says, to deny something real, namely, that
a moral dilemma can arise in a particular unfortunate circumstance. So, we
conclude, the two choices are both in accord with Marcus's account, and they are
both right. When all histories are at issue, it is right to assume with Anderson
that the sanction is avoidable, but when we consider only the histories that
are possible historical continuations of a particular moment, then it is right to
allow with Marcus that in some circumstances there may be no way to avoid
the sanction.
You can easily calculate that in a proper no-can-do case, Definition 12-2
neither obligates nor forbids the agent to see to it that Q, which sounds right
when the agent no can do. Of course it follows that in the proper no-can-do
case the agent is permitted to see to it that Q, and also permitted to refrain
from seeing to it that Q, which sounds peculiar given that the agent is powerless
with respect to Q. But it sounds peculiar, we think, only because in ordinary
English we confuse permission and power. (The effort to straighten us out on
this goes back at least to Hohfeld 1919; see especially Makinson 1986.) For
example, we often use the same modal verb, "may," for both. When we reflect,
however, we easily see that in fact there are numerous situations in which an
agent is permitted or authorized to do what in fact is not in their power to
do. The agent a might be permitted to park at any meter in the entire state
of Connecticut, even when a is enjoying Oregon. The proper no-can-do case
highlights this.
The normal case. In the central case of interest, the sanction Sa is false on
at least one history through m, and there is at least one choice on which a sees
to it that Q. We will call this the normal case, even though it certainly does
not happen with much frequency. In the normal case, a can see to it that Q,
and in the normal case, the sanction is not inevitable; see Figure 12.3.
It is a calculation of dstit theory (as opposed to the achievement stit) that in
the normal case, though not at all in general, "seeing to it that" and "refraining
from seeing to it that" are contradictory. (The sanction is irrelevant to this
calculation; one needs only can-do.)
It would be all right to shorten [a dstit: Pax] in (9) to Pax, just because,
according to Postulate 12-1, the latter is agentive, that is, [a dstit: Pax] and
Pax are equivalent. But it is probably better to use the longer "normal form"
in order to remind us that the complement of Frbn: must always be a dstit or
something equivalent. This is in accord with the restricted complement thesis,
Thesis 5. Let us add here that by the restricted complement thesis a third can-
didate, Frbn:Ex[a dstit: Pax], is not on the face of it grammatical. We should
suspect that this third proposed complement of Frbn: might, in the language
of Marcus 1966, describe a mere state of affairs that could not meaningfully
be on a's list of prohibitions. And in fact the coming analysis of the 3x/dstit
transition confirms this suspicion.
Here is a list of plausible equivalences, which we call the "Main Calcula-
tion," that would lead us between (9) and (10). We have indicated on the right
how each entry is tied to the one above it. In gently moving quantifiers from
the outside to the inside we will run into a number of particular transitions:
"QTF" means "quantifier-and-truth-function equivalence." Other transitions
12. Marcus and the problem of nested deontic modalities 327
are applications of previous postulates, as indicated. And the three key transi-
tions "Vx/Sett," "Vx/dstit," and "Ex/dstit" are based on hypothesized quan-
tifier/modality relationships. These three we explicitly discuss, with an eye to
either confirmatopn or disconfirmation.
Main Calculation
Not to put too fine a point on it, we think Postulate 12-3 is good. Here is
why. Sett: at a moment means, close enough, "true in all histories" through
that moment. You can therefore plainly see that Sett: has the intent of an S5
necessity. That means that the permutation of the universal quantifier with
Sett: is precisely the Barcan formula in one of its guises. (From Barcan 1946.
So named in Prior 1957.) The Barcan formula is not only time honored and of
328 Applications of the deliberative stit
12-4 DEFINITION, ([a cstit: Q], the Chellas stit; Poss:; Can:)
[a cstit: Q] is true at a moment-history pair m/h iff Q is true at m/h1 for
every history h1 through m that lies with h in the same possible choice
for a at m. (Note the absence of a "negative condition.")
The advantage here is this: The Chellas modality, having only a positive and not
a negative condition, is an approximation to agency that is easier for certain
analytic purposes than dstit. For example, we can transparently see that a
Barcan formula is appropriate:
12-6 POSTULATE. (Barcan formula for [a cstit: Q]) Vx[a cstit: Qx] <-> [a
cstit: VxQx].
330 Applications of the deliberative stit
Anyone can now see that it is not plausible, even in the Special Case, that the
Right Side (13) should imply the Left Side (12). Nothing is going to make it
plausible that whenever some highway is such that a can park on it, it is also
true that each highway is such that a can park on it. There are just too many
highways, many of them remaining in Connecticut even when agent a is about
to park in Oregon.
As a consequence of the negative condition Poss:~Q, then, we lose the Barcan
formula for Vx/dstit. It is natural to ask whether we should keep the negative
condition at such a high price. We could re-define [a dstit: Q] as [a cstit: Q}.
Then (11) would be valid; indeed, the entire Main Calculation would go through
quite easily. There are, nevertheless, compelling reasons for retaining the neg-
ative condition. For one thing, the simplified semantics would make Oblg:[a
dstit: Q] valid for every tautology Q. More generally, dropping the negative
condition would blur the distinction we have made between the evaluative and
prescriptive modal families. We intended to limit the discussion to the "ought
to do." The negative condition is intended to reflect a basic fact about choice:
An agent can only see to Q if it is possible (at that very moment) for Q to be
false.
What is the upshot? We return to this after discussing the third and last
transition of the Main Calculation.
You should see Ex/dstit as a motion of the existential quantifier between the
inside and outside of the dstit modality. The General Form is certainly to be
rejected in passing from right to left. When in a game of chance one rolls a
couple of ordinary dice, one sees to it that they come up some number or other
(two through twelve), but it is false that there is a number such that one sees to
it that they come up that number. The former is taking one's turn. The latter
is cheating.
We must still, however, look at the Special Case (14). What is most obvious
about the Special Case is that for each value of x, the complement of the dstit
on the Left Side is itself agentive (by Postulate 12-1). Even with this extra
information, however, the equivalence fails. Briefly put, what holds instead is
this. Provided Pax is agentive for every value of x,
We might not have noticed this because the extra conjunct seems so overwhelm-
ingly plausible in the highway-parking example; it seems difficult to imagine a
moment at which it is absolutely settled that there is a highway on which a
parks. It is important to realize, however, that the difficulty of imagination
arises from the example, parking on highways, not from agency itself.
We would belabor this subtlety except for the following: Even though the
General and the Special Cases fail, the Special Case in Context (15) of the
3x/dstit transition is in fact logically valid, given Postulate 12-1. The sur-
rounding context makes the difference. We skip the detailed calculation, which
indeed depends heavily on the logic of dstit (the analysis for the achievement
stit would have to be substantially different) in favor of a mere record:
12-7 FACT. (Special case in context of Ex/dstit) lf Vx[Qx <-> [a dstit: Qx]]
then (a dstit: ~3z[a dstit: Qx]] <- [a dstit: ~[a dstit: E x Q x ] ] .
Here as it turns out we have an even weaker statement, implied by but not
implying (17). As a visual help in sorting this out, here is a definition, intended
only for restricted complements, that highlights a form common to (17) and
(18).
Then (17) and (18) are more transparently written in terms of weak prohibition
respectively as follows:
So now we have four prohibitions, all subtly different, with irreversible implica-
tions as follows:
Inspection of that array seems enough both to explain and to correct our feeling
of equivalence between (7) and (8).
Which of these is wanted when we are writing up the traffic regulations for
the state of Oregon? The following equivalences should guide us in the task:
These equivalences suggest that (9) and (10) are unacceptable interpretations of
the parking prohibition (3). By (21), (9) entails that if there is any highway that
a can't park on, then a has no good choice. Given the abundance of highways,
(9) would perpetually put all agents (including the lawmakers themselves!) in
the no-good-choice situation. Similarly, as (22) shows, (10) entails that if there
is no highway that a can park on (for example, when a is sitting at home),
then a has no good choice. Neither of these objectionable consequences follows
from (19) or (20). The only difference between these two versions of the parking
12. Marcus and the problem of nested deontic modalities 333
prohibition occurs in the (rare) case of a forced choice. For example, suppose
you are moving toward a Y-split between two highways when your engine stops.
You are forced to choose between parking on one or the other highway in the
strongest possible sense: If you don't park on one you cannot escape parking on
the other. In this case, since it is settled true that there is a highway on which
you park, we can see from (23) that (19) implies that you suffer the sanction no
matter what you do. You are in a "legal dilemma." In contrast, (20) does not
provide a sanction in this case of forced choiceeven if you yourself purposely
stopped your engine by turning the key.
It is plausible that the lawmakers in Oregon would want one of the weaker
prohibitions, (19) or (20), but it seems less plausible that they would want the
weakest. For the rest of this essay we will assume that (19) is the most natural
choice.
indeed it surely is correct for the powerful and interesting languages that Mar-
cus has investigated. There are, however, the at least equally powerful and
equally interesting languages championed by Carnap and fully developed by
Bressan and Montague. In these languages singular terms carry "individual
concepts" as their intensional semantic value. These are good and useful lan-
guages. Quine has made delicious fun of them over the years, and Kripke and
others have cooked up their own pungent mockeries, but surely it is high time
to stop dining on proofs by spoof when these are pressed into the service of
intolerance. As Pooh would say, such proofs are not sustaining. Here we are led
to serve up a particularly satisfying portion of advice from Marcus:
... the polemics of modal logic are perhaps best carried out in terms
of some explicit semantical construction. (Marcus 1961, p. 319)
Without fully digesting this advice, however, let us say that such constructions,
namely those of Bressan and Montague, are standing ready in the kitchen. Then
let us abandon the metaphor and just wing it for a while.
Our point is that there is something fundamentally non-extensional about
the agent position of dstit statements. Even for the single evaluation of a single
statement [a dstit: Q] at a single moment-history pair, we need to fix the agent
at that moment independently of histories. We need to be able to say what
choices are available to the agent regardless of what happens next. Further-
more, when we proceed, as we must, to speak of strategies or other concepts
that take us from moment to moment, we shall need to fix the concept of the
agent in an even more severe sense. All of this is carried by saying that the
underlying concept of "agent" must be "absolute" in Bressan's sense. This is
a great deal like saying, in the language of Marcus 1961 (p. 304) that agents
as we need them are "constant objects of reference." There are, however, two
differences. First, the Marcus phrase is informal and metalinguistic, whereas
Bressan's account is carried by a rigorous object-language but second-order def-
inition of the absoluteness of concepts. Second, the Marcus approach, which in
this respect is the same as Kripke's, renders this kind of idea a presupposition
of applying quantified modal logic. In contrast Bressan leaves it open for us
to make it a matter of extra-logical theoryfor example, a matter of serious
empirical physics in the theory of general relativitywhether a concept is or is
not absolute.
There remains another side to the point. As we can learn from Bressan,
even the most non-extensional of concepts have extensional cousins that arise
by "extensionalization" with suitably weak equality relations. Given any Qx,
however fierce, there is always Ex1[(x eq x1) & Qx1]. The predicate of parking
on a highway, and dstit in general, are like that. All we want to add to these
unfortunately cryptic remarks is this: Postulate 12-1 and the various definitions
are intended to hold only for a that fall under the absolute concept of agent
and not for extensional equivalents thereof.
This leaves much that is problematic for us to treat on some other occasion,
perhaps the same occasion on which we can discuss what is needed to cater to
the natural wish to restrict the range of "a" in (24) to some sane set of agents.
12. Marcus and the problem of nested deontic modalities 335
force a search for the "action" of forbidding each agent to park on a highway.
Fortunately, however, no such ontological chase is called for. The thing is that
we already have what is needed to make entirely good modal sense out of one
agent B forbidding another agent a, or as in the present example, forbidding all
agents. To forbid an agent a to park is to see to it that a is forbidden to park.
To forbid all agents to park is to see to it that all agents are forbidden to park.
No tricks, just an appropriate dstit.
Now to say that b ought to forbid parking on highways is just to use language
already at our disposal:
What goes wrong at this point is that we don't much feel like existentially
generalizing this to say that there is some particular agent that ought to do the
forbidding. And the restricted complement thesis prevents us from putting that
existential quantifier inside the Oblg: modality (though we could perfectly well
do that if the modality were impersonal).
There is, however, a natural and proper way in which to express the generality
of the not-very-personal "ought" while still treating the "ought" as prescriptive
and while still hewing to the restricted complement thesis. Instead of "someone"
in the logicians' sense conveying quantification over agents, it would be natural
to say
Here the subject "they" of the English sentence is intended not as a singular
quantifier phrase, but as the name of a class or aggregate in the sense elaborated
by Marcus in Marcus 1963 and Marcus 1974. The obligation is collective, not
distributive. If there were no theory about such constructions this would be
just throwing sand in your eyes, but in fact there is. Take F as some suitable
aggregate of agents, and use the theory of joint agency from 10C. Then (25)
is well symbolized by the following.
This says that there is a joint obligation on F taken collectively to see to it that
parking on highways is henceforth forbidden. If we apply the Anderson/dstit
simplification of 12C to this extended form of obligation, we shall find that
we need to index sanctions not only by single agents a, as in Sa, but also by
groups of agents: SF. That is of course just a formal remark. The concrete
content stands in need of exploration. Our progress here is limited to seeing the
difference between (26) on the one hand and, on the other, either
which is not only unwanted but is also a disagreeable violation of the grammar
required by the restricted complement thesis.
One last complication. The point of view of modality makes us see very
clearly that by one more degree of nesting we could express that whereas (i)
the obligation is joint on the group, nevertheless (ii) the particular dstit is to
be carried out by some member of the group. In this case the dstit is of a
forbidding, so that we have the following:
Again the thing to notice is that the new subtlety requires no tricks. Nothing is
made up to order. All is transparent. We are just using the power of modalities
combined with quantifiers to which we have become entitled by the philosophical
pioneering of Barcan Marcus.
This page intentionally left blank
Part V
Strategies
This page intentionally left blank
13
An austere theory
of strategies
341
342 Strategies
Von Neumann games in extended form are discrete: Every node of the
tree (except the first) is immediately preceded by a unique node, and every
node (except a terminal node) is immediately followed by some accessible
nodes. It is as if the structure of time were discrete. The austere theory
refrains from any such assumption. Nor does it substitute any other as-
sumption about the structure of time, for example that it is continuous
(as in theories of infinite games). The austere theory is truly austere.
The von Neumann theory postulates that every branch on every tree is
not only discrete, but finite in length. The austere theory refrains. Nor
does it substitute an assumption that branches are infinite.
13. An austere theory of strategies 343
The von Neumann theory postulates that at any node there are a finite
number of available choices. The austere theory refrains. Nor does it
substitute an assumption that there are infinitely many available choices.
The von Neumann theory postulates that at each node there is only one
agent that can make a choice. The austere theory refrains.
The von Neumann theory adds as a primitive the notion of "information
sets" to mitigate the "one nodeone agent" restriction just mentioned.
Since the austere theory doesn't start with such a restriction, it refrains
from any such addition.1
The von Neumann theory of games rests on the theory of utility, which pro-
vides a value for each outcome. It is as if sane consideration of strategies
had to treat "strategies" as "strategies for" some outcome (e.g., winning
in chess). The austere theory refrains. Instead it investigates the part
of the theory of strategies that makes sense without values; this part is
"pre-utilitarian," and interesting.
One should think of the present work as more like geometry than anything
else. The proposed concept of "strategy" is analogous to Euclid's concept of
"triangle." Strategies, like triangles, will be tokens, not types; they will be
instances, not kinds. When Euclid proves the Pythagorean theorem about any
right triangle, he proves it about each instance. His proof is only derivatively
about "the right triangle" or "right triangularity." The proof certainly does
not give us information about either the psychological idea of the triangle or
about the word "triangle"these battles have been fought and won since Frege.
It is just so when we prove the existence of, for example, a complete strategy
satisfying certain conditions. We do not mean to prove the existence of a kind,
much less an idea of a kind, much less something with linguistic structure.
Instead we prove the existence of a concrete instance that has as much "location"
in our indeterministic world as does a particular triangle.
Strategies will be objective. Keeping the analogy with triangles in mind
might help one get strategies out of the head and into the world. Strategies, in
the austere sense treated here, are neither linguistic nor mental. Of course one's
philosophy might make everything an idea, triangles too. The philosophical road
now forks. Path A: One who goes on to infer that Euclid's work was therefore
somehow misguided can safely skip this chapter, and indeed this book. Path
B: One who still makes room for Euclid should treat the forthcoming account
1
Information sets are useful in practical cases, including many found in economics, in
which it seems convenient to run together causal and epistemological considerations. They
provide, however, an unsatisfactory foundation. Because information sets conflate causality
and ignorance, their use forever prohibits the game theorist from distinguishing between causal
constraints and epistemic constraints. One who uses them cannot even articulate a theory
about how the two might relate! We remark that this point is not intended as new, nor as
suggesting that the game theoretical community does not worry about the issue. It is only to
suggest that the clearest way to proceed is to throw out information sets and then to start
over with explicit and separate representations of causality and of ignorance.
344 Strategies
of strategies in the same way as the Pythagorean theorem, whatever way that
might be.
What can we expect and what should we not expect? We should not expect
an account of "what it is to have a strategy"; the concepts of the austere theory
are not rich enough to enable such an account. We should expect an account
of what a strategy is in a structural sense. To put the matter tautologically,
a strategy is what one "has" when one "has" a strategy. We should, to use
untrustworthy phrases from the ethical literature, expect an account of what it
is to act "in accord" with a strategy. We should not expect an account of what it
is to act "on" a strategy. This is analogous to the following: Euclid tells us what
it is to be a triangle, but he does not provide an account of what it is to see a
triangle. The present theory, like Euclid's, has nothing interesting to say about
distinctively mentalistic conceptsexcept, perhaps, the following: Strategies
have both psychological/linguistic/cultural and structural/geometrical/causal
aspects, and there is hope of a benefit if we disentangle the two.
The analogy goes a little further. Just as Euclid's definition of "triangle" is by
itself almost without interest, so is our definition of "strategy." It is only when
one passes from the most general concepts to (what Whitehead called) "happy
particularities" that one should hope for enlightenment. We offer several modes
of classification that, with the help of established conceptual connections (which
we call facts, lemmas, or even theorems), aim to deepen our understanding of
strategies.
Not for the theory of strategies itself, but for the application we give in
13E, we need the achievement stit, here written just [a stit: Q], with the
chain witness semantics, 8G.4.
We remind the reader that we take (Tree, \<, Instant, Agent, Choice) sometimes
as a (more or less adequate) idealized representation of how agents and their
choices fit into Our World, in which case the BT + I + AC postulates are "pos-
tulates" ; and sometimes as an abstract structure, in which case the postulates
governing BT + I + AC are clauses of the definition of "BT + I + AC structure."
The foregoing finishes the list of concepts and postulates concerning choices in
branching time that are needed for an austere theory of strategies. No "surprise
postulates" or "commonsense assumptions" are to be expected. Nor will we
add a concept unless it is defined from these according to the standard of rigor
preached and practiced by Frege.
We say that s1 extends so when the second clause can be improved to say
that s1(m) = s0(m) for each
Note the difference in directions of the subset relation. Adding a moment to the
domain strengthens a strategy (we call such an addition an "extension"), while
cutting down the number of options also strengthens it. In this study we shall
be almost exclusively interested in those strengthenings that are extensions.
The reason for this is that in this section we are interested primarily in strict
strategies, and of course a strict strategy can be consistently strengthened only
by extensionsince to cut the number of options below one is to cut them to
zero.
13. An austere theory of strategies 347
One should keep in mind that the technical idea of the "join" of some strategies
represents their common strengthening; so it is more likely to be expressed in
English with "and" than with "or."
The ideas of exclusion and admission are central to the theory of strategies. As
for exclusion, one may start out along an excluded history, but if one chooses in
accord with s, then one cannot forever remain on an excluded history. Eventu-
ally one reaches a moment at which the strategy goes one way and the history
goes another. Nor can one ever reach an excluded moment if one follows a
strategy.
Admission is in one sense a weak notion; for example, any history that con-
tains no member of Dom(s) is vacuously admitted. Apart from this trace of
vacuity, which we remove on a case-by-case basis when we apply the notionfor
example, in the next definitionan admitted history is one that might happen
given that the strategy is faithfully followed. An admitted moment is one that
might be reached provided that the strategy is followed without exception. Ad-
missibility represents what is possible, given that the strategy is followed. It is
therefore natural to define what a strategy "guarantees" in terms of admissibil-
ity:
s guarantees HO iff
s really guarantees HQ iff s guarantees HQ, and furthermor
348 Strategies
Among strategies there is a sharp difference between those that do and those
that do not tell you what to do after you have failed to follow their earlier advice.
We call the former "secondary" and the latter "primary." (Our earlier work on
strategies used the pair of words "redundant" and "irredundant." Horty 2001
uses "lean" where we say "primary." Although the ideas are stable, the best
choice of vocabulary is unsettled.) The thought is that advice given by a strategy
after you have already violated the strategy may be useful, but it is surely
"secondary" to the primary portion of the strategy. So we call the strategy itself
"secondary" or "primary" depending on whether it does or does not contain any
secondary advice.
The strategies (for games in extended form) of von Neumann and Morgenstern
1944 are primary, or mostly so. A typical game theoretical strategy does not tell
you what to do if you do not follow the strategy. Suppose a strategy for chess
begins with advising king's pawn to KP4 as first move. That very strategy
would typically not tell you what to do on your second move if instead you
moved your king's rook pawn on your first move. Why should it? Such advice
would be redundant (it would be secondary advice).
One finds a contrast in the "deontic kinematics" of Thomason 1984, p. 155,
which we discuss in chapter 14. In that scheme, obligation is laid on at every mo-
ment. If we think of these obligations as together making up a strategy (Thoma-
son speaks of plans), the strategy would be secondary. And for good reason;
as Thomason notes, secondary strategies enable consideration of "reparational"
obligations, telling us what to do after we have done wrong. Secondariness can
be important.
13. An austere theory of strategies 349
It could be that a strategy gives advice for every moment in M, or even for
every moment in all of Tree. Such a strategy is said to be "total":
We shall not be interested in total strategies until chapter 14. We are instead
presently interested in notions of "completeness," which are altogether different
from totality. A strategy complete in a set of moments M will tell us what to do
at appropriate moments within a field Mbut by no means all. For instance,
a "complete" strategy need not tell us what to do at moments in the field that
the strategy itself forbids. That is precisely why a field for a strategy must
be distinguished from its domain. Nor is a field uniquely determined by the
domain; that is why we say "field for s" instead of "field of s."
A "conditional" strategy might well not be backward closed: "If you reach
Station A, get off the subway; but for earlier stations, there is no advice." So
strategies that are not backward closed are important. But if a strategy is not
backward closed, then that seems enough to give it a kind of not-yet-available
flavor. Such a strategy gives us advice at a certain moment without giving us
any advice whatsoever as to how to get there. It would in that respect be similar
to an end-game strategy for chess as given in a book on end games, provided we
chose an odd field for it. Namely, suppose we took as field not just the subtree
where the book starts giving us advice. Suppose instead we oddly took as field
the entire tree for chess that starts with the first move. Then that strategy
would not be backward closed in that field. If we want to confine ourselves to
the simple case in which a strategy is available from the very beginning of its
field, we must think about backward closed strategies. This notion is therefore
the third and final part of the concept of simplicity:
So a strategy that is not simple might fail of simplicity in one of three ways,
(i) It might be loose (sometimes giving disjunctive advice), (ii) It might be
secondary (sometimes giving advice at moments it forbids). (Hi) It might not
be backward closed (some parts of its domain might be isolated high up in M).
Simple strategies are not like that.
13C.3 Pre-simplicity
Permit us to say something parenthetical about strategies that are not simple
in virtue of being not backward closed in Mparenthetical in the sense that
later sections do not build on this one. Suppose such a strategy is in fact strict
and primary, and fix the set M of relevant moments. The strategy might have a
simple extension in M, or it might not. There is only one case in which it does
not: There are two moments TOI and m^ in its domain, and they have a lower-
bounding moment m0 in M (but not in Dom(s)) such that every choice for a
atTOOexcludes one of m1 and m2. Were this case to occur, we should not know
how to specify s(mo) while retaining both strictness (at TOO) and primariness
(with respect to each of m1 and m 2 ). This observation points to an alternative
concept of "simplicity" for use when we wish to consider strategies that are not
backward closed. To remind us that the alternative is helpful only when we are
looking forward to simplicity, let us call it "pre-simplicity."
13-12 DEFINITION. (Pre-simplicity) s is pre-simple for a in M iff s is a strategy
for a in M that is strict for a, primary, and such that each two members of its
domain are inseparable for a in M.
EXAMPLE. (Pre-simplicity) Let s be a strategy for you that might be described
in this way: "If you reach Station A or Station B, get off the subway; but for
the earlier moments at which you have a choice about reaching Station A or
Station B, there is no advice." You may not care about completeness at all. If
you do, you may not care about it with respect to a field M for s that includes
those earlier moments. In these cases you will not care about pre-simplicity in
M (though you may still care about it with respect to a smaller field). Suppose,
however, that a field M for s includes at least one of these earlier moments, say
TOO, at which you can choose between reaching Station A and Station B; and
suppose that you care about strengthenings of s that are complete in M. You
may or may not be interested in primariness. If you are not, you can think about
extending the strategy to mo in any way you like. You could then think of some
or all of the "if you reach Station A or Station B" advice as "reparational." You
could think of it as "redundantly" advising you what to do after you have fallen
off the strategy. In this casesee chapter 14pre-simplicity would be of no
interest. Suppose, however, that you are interested in primariness. Even then it
makes sense to extend the strategy toTOO,provided the advice there is not strict
but loose: Head either for Station A or for Station B, whichever you choose.
That would be a primary (though loose) extension. Nothing wrong with that.
But if you are interested in a strategy only if it can be backward closed to a
strict and primary extension, you need to think about pre-simplicity.
352 Strategies
Here are key facts about pre-simplicity. They are put here for help in keeping
our bearings, and are not otherwise used in this chapter.
13-13 FACT. (Pre-simplicity)
The smallest simple extension as guaranteed by (i) might be called the "back-
ward closure" of s. But we won't need the idea, and in any event that phrase
could well be used in other ways when we are interested in loose or inconsistent
or separable strategies. Part ( i i ) gives an equivalent account of pre-simplicity
so as to exhibit that pre-simplicity is a unified idea, not a mere conjunction:
Strictness, primariness, and inseparability are all part of the same package. And
(Hi) tells us that when we know a strategy is backward closed in M, we just
don't have to worry about simplicity versus pre-simplicity. In any event, we
shall nearly always be dealing with simple strategies.
PROOF.
Ad Fact 13-13 (i). Suppose SQ is pre-simple for a in M. Define si so that
Dom(si) = Dom(s0)U{m0: mo M & 3mi[m0 < mi & mi 6 Dom(s0)}};
and for m define
Evidently si is backward closed in M, and si is primary
if SQ is primary. Also pre-simplicity of S0 guarantees that s1 is strict. So s1 is
a simple extension of s0. Conversely, if s0 is not pre-simple, then if it is either
not strict or secondary, no extension can remove that defect. And if its failure
of pre-simplicity comes from having two members m1 and m2 of its domain
that are not choice equivalent at some moment mo in M that lies in the past of
each of m1 and m2, then there is no way to extend s0 backward to m0 without
violating either strictness (at mo) or primariness: If Si(mo) is a single member
of Choicean0, then either m1 or m2 will be excluded, causing s1 to be secondary.
Ad Fact 13-13 (ii), left-to-right. Argue by contraposition. Suppose that s is a
strategy for a in M, but that the right-hand side is false: mi, m2 G Dom(s), hi
S s (mi), hi e s (mi), m0 e M, m0 ^ mi, mo ^ m2, hi _L^ 0 hi. Argue by cases.
If mo = mi = mi, then there is nonstrictness at mo since non-choice equivalent
histories belong to s(m 0 ). If mo = mi and mo < mi, then mo excludes m2,
witnessing that s is secondary. Similarly if m0 < mi and mo = m2. And if m0
< mi and m0 <mi, then there is a violation of inseparability. So in either case
s is not pre-simple.
Ad Fact 13-13 (ii), right-to-left. Argue by contraposition. If s is not pre-
simple, it is either not strict, or secondary, or separable. If it is not strict at
mo then we can find a counterexample to the right-hand side with mo = mi
13, An austere theory of strategies 353
advice, and then sometimes leaves you adviceless even when you follow it, even
when you are still in M.
Complete strategies are analogous to the complete theories known to model-
theoretic logic in the following respect: Although interesting specimens of either
are few and far between, complete strategies, like complete theories, are impor-
tant objects of study because of the limiting role they play.
One can see thatand be a little surprised by the fact thatfor strategies in
M, completeness is converse to primariness:
Let us record that extending a strategy can never make it less complete than
it was.
The next definition and fact articulate the technically useful idea of extending
a strategy by an entire admitted history (more accurately, by that portion of it
in M).
13-17 DEFINITION. (Extension along h) Let s0 be a strategy for a in M.
Define "si is the extension of S0 along h for a in ' as follows:
13. An austere theory of strategies 355
We shall also need some near-the-surface facts about joins of chains of strate-
gies in the sense of the "weaker than" relation of Fact 13-3.
PROOF. Use Zorn's lemma. A violation of any of (i)-(vi) for SE would be finite,
and hence would be a violation of the same property for some member of
13D Favoring
Let us return to the problem mentioned at the beginning of the chapter: As
promised in 91, find a strategy the following of which will guarantee not seeing
to it that Q. The core of an intuitive solution is easy: Shoot for ~Q when-
ever given a choice. To make clear sense out of this suggestion, we introduce
the concept of a strategy "favoring" a set of histories. The root idea comes
from Hamblin 1987, p. 157, where the topic is carrying out (or "extensionally"
satisfying) an imperative:
... the addressee of an imperative would be expected, at least, to act
in such a way as to keep extensional satisfaction within the bounds
of possibility. This means that he must not do any deed d that
would infringethat is, that would ensure dissatisfaction ofthe
imperative.
The next definition aims to capture exactly this idea. In reading it, think of HO
as the set of histories on which Hamblin's iniDerative is satisfied.
13-20 DEFINITION. (Favoring) Let s be a simple strategy for a in M.
s favors H0 iff
PROOF. Well-order H0 with the ordinals less than a limit ordinal A'. (We omit
adapting the argument for finite H0.) For 7 < A', let /i7 be the history in H0
that is indexed by ordinal 7. Furthermore, when the "furthermore" clause of
the lemma is wanted, in the special case 7 = 0, as the "furthermore" hypothesis
we let h-y G H(mQ)<r\HQ^Admh(s). Use this well-ordering to define an ordinal
sequence of strategies SQ, .,., s7, ..., s\i, as follows:
s0 = s.
If s7 admits /i7, then s7 + i is the extension of s7 along /i7 in M. Otherwise,
which mi lies; m( J(m ) t h ) is the moment in which the instant i(mi) intersects
the history h.
The following is the concept of the truth of [a stit: Q] based on witness by
a chain (instead of witness by only a moment); it is a merely verbal variant of
8G.4, with which it may be compared.
13-24 DEFINITION. (Truth of [a stit: Q]) Given that mi hi, we define that
97t, mi/hi 1= [a stit: Q] iff there is a chain of moments CQ, called a witness, such
that the following conditions are satisfied.
Priority-nonemptiness. All moments in CQ must be properly earlier than
mi, and CQ must be nonempty: CQ ^ 0 and Vmo[mo G CQ >TOO< mi].
Positive condition. Q must be settled true at every moment in i( mi ) that
lies on a history intersecting CQ that is inseparable from hi for a in CQ:
We also need a field of choice points for a. On the "stit" side, it must be
large enough to include all the chains that could possibly witness any of the
stits considered in the desired result. On the "strategy" side, since the strategy
is to be "available" at mo 6 CQ, the field must include mo. The field, however,
must not include any moments earlier than mo, which are evidently no longer
"available" at mo. Let us take the set of all moments from mo onward that are
temporally prior to mi, that is, that precede some moment co-instantial with
mi: MI {m: mo ^ m & m < i( mi )}. We record these two abstractions for
later reference.
13. An austere theory of strategies 361
13-25 STIPULATION. (Two critical sets) Suppose that TOO, mi, and Q have
been fixed, and thatTOO< TOI.
Hi = {h: m, m ( l ( m i ) i h ) //i N Q}.
Next notice that /ii doesn't really come into the Definition 13-24 account of
truth for stit. Consequently we can take TOI to come in only as a peg on which
to hang the definitions of H\ and M\. If we resolve the general nonemptiness
condition on the chain witness c in favor of a particular member TOO of c, we
can state the remainder of the truth condition for stit as a relation between TOO,
HI, MI, a, and an arbitrary history ho. We codify this "stand-in" for stit as a
general definition of a five-place relation, stit-stand-in(mo, HQ, M0, a, ho).
13-26 DEFINITION. (Stit-stand-in) For any moment TOO, set of histories HO,
set of moments M0, agent a and history ho, stit-stand-in(m0, H0, M0, a, ho)
iff there is a c such that the following all hold.
Priority-nonemptiness stand-in.TOO
Positive stand-in.
Negative stand-in.
The positive stand-in says this. Let h be any history intersecting c such that c
makes no choice distinction between ho and /i. Then h G HO- That is why we
can say that the choices made in c by a are such as to "guarantee" being in HQ.
Or to put it contrapositively, if h G (H(c) Ho) then 3m[m G cD/ioPI/i & /IQ
_L^j /i]. The negative stand-in says that every moment m in c keeps departure
from HQ as a live possibility. That makes c nonvacuous as a seriesthough it
does not and should not follow that a has a nonvacuous choice at each member
of c. The condition says that at every moment TO in c there is a "live possibility"
of finishing outside of HQ. It does not say that there is a currently available
"live option."
It is important to see that stit-stand-in is a concept entirely in the language of
the theory of strategies. For the upcoming result about stit-stand-in to have its
intended significance, however, we must verify that the concept is satisfactory
on the "stit" side, as follows.
13-27 FACT. (Stit and stit-stand-in) Fix a and TOI, and h\ such that TOI h\.
Assume that 9JT, mi/h\ N [a stit: Q], and let CQ be a witness (Definition 13-24).
Fix TOO G CQ. Define HI and MI as in Stipulation 13-25. Choose any history ho
G H(mo). Then 971, m,(i(m },hQ)/ho ^ [<* stit: Q] iff stit-stand-in(m0, HI, MI, a,
ho).
362 Strategies
PROOF. It is almost straightforward that given the stipulations for HI and MI,
the stand-in conditions are restatements of the truth conditions for [a stit: Q] at
the moment-history pairs in question. There is just one delicacy in the part of
the argument from left to right: The hypothesis promises a witnessing c for [a
stit: Q] at m,(,(m )tho)/ho> but it does not say, as required for stit-stand-in(mo,
HI, MI, a, ho), that mo 6 c. Nevertheless, since mo is assumed to be part of
a witness for [a stit: Q] at m\jhi, we can be sure that if c witnesses [a stit:
Q] at TO(i ,,h 0 )//iO) then so does cUJmo}. The reason is this. We know by
the negative condition for the witnessing of [a stit: Q] at mi jh\ that Q fails
at m(j ( m /i)//i for some h 6 H(moy Therefore, the chain c cannot be entirely
earlier than mo on pain of violating the positive condition for the witnessing of
[a stit: Q] at m(i( }tho)/ho- So some member of c must come properly after
mo. But then adding mo to c keeps the negative condition true, and does not
disturb the positive condition, so that cU{mo} is also a witness to [a stit: Q]
at m(j ( m ),h0)/ho- And of course that set is bound to contain mo, as required
for stit-stand-in(mo, HI, MI, a, h0). (There is a discussion in Belnap 1996a of
what can be added to or subtracted from a chain witness.)
13-28 THEOREM. (Strategy for inaction) Fix mo, MO, HO, and a such that mo
Then there is available at mo a simply complete strategy s for
a in MQ such that stit-stand-in(mo, HQ, MO, a, ho) fails for every history ho
admitted by s.
PROOF. By the second part of Corollary 13-23, let s be a simple strategy for
a in MQ available at mo that is complete in MO and that favors -H(M 0 ) ~Ho m
MQ. Suppose s admits ho (and is therefore complete along /i0 in MO), and, for
contradiction, that stit-stand-in(m0, H0, MO, a, h0) holds (Definition 13-26).
By priority-nonemptiness let mo G c C /i 0 nM 0 . By the negative condition,
so that by favoring, we know that we may let
By the positive condition, So choose
hi (hence in h0r\Mo, therefore in Dom(s)) such that hi ^ ll ho. Here is the
contradiction: ho and hi are both admitted by s, so must both belong to s(mi).
But then s is not after all strict.
13-29 COROLLARY. (Main result) Suppose that 9Jt, mi/hi 1= [a stit: Q], with
witness CQ. Then at each moment mo G CQ, there is available a simply complete
strategy s for a in that really guarantees
PROOF. Start with the hypothesis, and choose m0 CQ. Define MI, #1 in
accord with Stipulation 13-25. Since m0 e MiHljHi, by Theorem 13-28 there
is available atTOOa simply complete strategy s for a in MI such that stit-stand-
OT(TOQ, -Hi, MI, a, ho) fails for every history /i0 admitted by s. By Fact 13-27,
that amounts to saying that s guarantees
And the hypothesis allows us to calculate that s is not vacuous for the set
of histories
Deontic concepts are naturally linked with strategic concepts.* Surely any obli-
gation can be viewed as an obligation to follow a certain strategy in the austere
sense of chapter 13, always making such choices, depending on circumstances,
as conduce to satisfying the obligation. Conversely, to follow a strategy can be
viewed as very like living up to a set of deontic requirements, "doing what the
strategy requires." We do not here directly discuss such common-speech link-
ages. Instead we report a specific and detailed theoretical linkage that allows
reciprocal illumination between certain deontic concepts and certain strategic
concepts. An unexpected result is the connection between the no choice between
undivided histories condition from the austere theory of strategies, and a new
deontic kinematic condition adduced by Thomason to help describe how oughts
fit into branching time.
364
14. Deontic kinematics and austere strategics 365
i. Consistency. O(m)
ii. Locality. 0(m)
Hi. Kinematic.
The admission and exclusion concepts on this list, drawn from Definition 13-4,
are of critical importance for this chapter. Let us start with exclusion, which is
defined (with inescapable confusion) for both moments and histories. A moment
is excluded by a strategy if never violating the strategy implies that you will
never reach that moment. A history is excluded by a strategy if never violating
the strategy implies that eventually you will fall off that history. Since admission
is simply the contradictory of exclusion, a moment is admitted by a strategy
if it is possible to arrive there without violating the strategy, and a history is
admitted by a strategy if staying on that history never requires that you violate
the strategy.
permit the choice H. But what if H is "mixed," containing both some histories
acceptable to 0(m) and others that are unacceptable to 0(m)l There are then
evidently two salient strategic policies available, neither happy. The "weak"
policy says that choosing H, even though mixed, is in accord with the strategy.
The "strong" policy says that choosing a mixed H violates the strategy. There is
also a third policy, itself not entirely happy. The "Hamblin" policy, suggested in
Hamblin 1987, makes it depend upon what other options there are, as follows.
Let H be mixed. If there is at least one possible choice that contains only
histories acceptable to 0(m), then the Hamblin strategy forbids choosing such
a mixed H (like the strong policy). But if there is no possible choice containing
only histories acceptable to 0(m), then the Hamblin strategy permits choosing
mixed H. There is no point moralizing about this here; we simply put the matter
in terms of clear definitions about which something definite can be said.2
We can now intelligibly speak of a deontic tree as in effect one of the strategies
{sweak,a, o, sstrong,a, o, sHamblin,a, o}, and ask what properties it may have.
The consistency condition (z) of Definition 14-1 implies that sweak, a, o and
SHambiin,a,o are consistent. In contrast, the strong cousin s s tr<mg,a,o
may well be inconsistent even though the consistency condition on 0 is
satisfied. Suppose, for example, that the consistency condition is satisfied
at m by means of just a single history h through m that is acceptable
to 0(m). If the possible choice for a at m that contains h also contains
an unacceptable history, then according to sstr0ng,a, o, that choice is not
open to a. The strategy s s t r ong,a,o tells a that at m every option is
forbidden. That is, sstr0ng,a,o 1S inconsistent.
Any member of {sweak,a,o, sstrong,a,o, sHa.mbiin,a,o}, except in trivial
cases, is bound to be secondary. This feature distinguishes any strategy
based on a deontic tree from any simple strategy. The positive role of sec-
ondariness in the deontic-kinematic framework is to permit consideration
of reparational obligations, obligations that come into force after some ear-
lier obligation has been violated. We underline that the feature is neither
good nor bad; but one should bear it in mind. Primariness/secondariness
is the most significant difference between simple strategies and strategies
that are derived from deontic trees.
If S G {sweak,a, O, Sstrong.a, O, SHamblin,a, o}, S is total, 3S W6 Said. So it
is natural to expect that at some moments m, s is close to being undefined;
i.e., s lays on a at m only a vacuous obligation, the set of all histories
containing m: s(m) = H(my This is not primarily a technical point:
Any strategic advice that covers every moment in Tree must be largely
vacuous.
A strategy s {sweak,a,o, sstrong,a,o, SHambiin,a, o} derived from a
deontic tree is therefore never simple: Although s is bound to be backward
closed (in any M), it will never be both strict for a and primary.
Notably missing from this account is any reference to the kinematic condition
(in) of Definition 14-1. What does the kinematic condition mean? Thomason
1984 connects the notion with the idea of a "coherent plan." But there seems
nothing incoherent about a strategy derived from a deontic tree, whether or not
condition (Hi) is satisfied. This suggests that we have not found the best way
of relating austere strategics and deontic kinematics.
So suppose we start at the other end, with a strategy s. One might think
of the histories S(TO) as what a "ought to do" at TO. Suppose we were then to
turn things around and define 0(m) as s(m). We should learn that conditions
(i) and (ii) of Definition 14-1 would be satisfied at least when s is consistent.
We should also find out that condition (lii) is something that does not readily
spring to mind. Mutual illumination seems again absent.
(not admitted) by your strategy to visit your mother. Your strategy therefore
excludes all alternatives through mi. Your strategy is inconsistent at mi.
What has gone wrong? This: One wants for the reparational or secondary
situation a concept that gives a a "fresh start." That is why taking admission
as an absolute character of histories did not work. It is better to invoke the
idea of a history that is "admitted by s from m onward." In other words, the
history may have been excluded earlier, but don't look back: From m onward it
forever after counts as ideal. In still other words, past failures don't count. As
one might expect of a purely deliberative idea, only the future of possibilities is
relevant. Here is the definition.
14-7 DEFINITION. (Admission onward; Admh(s, m)) Let s be a strategy, and
let mo Dom(s).
s admits h from mo onward iff mo and
For any strategy s and moment mo, Admh(s, mo) is the set of histories
admitted by s from mo onward.
In applying this definition, we run into the contrast that strategies can be partial
while deontic trees are always total. For present purposes we therefore choose
to concentrate on total strategies. It is easy to see that from the point of view
of admissibility, nothing is lost. Given any strategy SQ we can always consider
instead the total strategy si obtained by vacuously defining Si(m) = H(m) for
every m & (Tree Dom(so)). (It cannot be inferred, however, that nothing
is changed tout court; obviously total strategies are not simple at all, so that
exclusive concentration on them would put potholes in the road to inquiring
about the properties of simple strategies.)
The upshot is that in the context of reparational obligations, given a total
strategy s, Admh(s, m) seems a good candidate for the concept 0(m) of deontic
kinematics. These are the histories that, having reached m, are and will remain
"ideal" as far as the strategy goes. They are the histories through m on which
(although obligations may have been violated earlier) no more violations occur.
Sounds "ideal," doesn't it? Let us record the suggestion in a definition.
14-8 DEFINITION. (Strategies as candidate deontic trees: Os) Suppose we have
a choice tree (Tree, ^, Choice). Given a total strategy s for a, we define the
candidate deontic tree determined by s as (Tree, ^, Os), where for m e Tree,
Os(m) = Admh(s, m).
Here is a decisive observation: Os is not a strategy and should not be confused
with a strategy. A strategy s picks out choices at m, whereas Os can pick out
some individual history within a possible choice for a at m, while excluding
other histories in that same choice. Such ideal histories are important; but one
must also keep in mind that no agent can choose the set of such histories.
What can we say about the so-induced candidate deontic tree, especially
with respect to conditions (i)-(iii) that Definition 14-1 laid on deontic trees?
14. Deontic kinematics and austere strategics 373
The following sections report that all of these conditions(i) consistency, (ii)
locality, and (Hi) kinematicfall into place given Definition 14-8. There you
have it: Think of deontic kinematics as generated by austere strategics, and each
of the deontic-kinematic conditions is illuminated. This is especially helpful for
appreciating the least obvious of these conditions, the kinematic condition (Hi).
We start at the first of the deontic-kinematic conditions.
It does not appear that the strategy pictured in Figure 14.1 is any more
incoherent than any other more "finite" reparational strategy. The appearance
is more that of a merely set-theoretic construction without intuitive content.
But perhaps that appearance is only due to our reluctance to think about the
possibility of busy choosers; it's hard to be sure.
PROOF. Suppose that we have a choice tree (Tree, <, Choice) with no busy
choosers. Suppose that s is a total consistent strategy, that is, that s(m) is
invariably nonempty. Choose any mo Tree. We show how to find a history
in Os(mo), that is, in Admh(s, mo). Define as follows a countable sequence of
moment-history pairs (mi,ht), 0^1 The first moment m0 is given. At this
and any stage, choose ht as any history in s(m;). It may be that /it (Admh(s,
ml)). In this case the sequence terminates in (mi, hi). Otherwise, {m: m hi &
ml < m & hi. s(m)} ^ 0. Because there are no busy choosers, this set cannot
contain an infinitely descending sequence, that is, it must have a least member.
14. Deontic kinematics and austere strategics 375
Define ml + i as that least member. Now repeat the definition of (ml,hl) unless
the sequence terminates. Either the sequence (mo, ho), (mi, hi), ... terminates
or not. If it terminates in (ml,ht), then ht & Admh(s, mo) (argument omitted).
Suppose it does not terminate. Consider the infinite sequence of moments mo <
mi < . . . . It cannot be upper bounded because there are no busy choosers. So,
since it has no upper bound, that sequence of moments must define a unique
history hw that contains all its members. We argue that hw <E Admh(s, m0).
To this end, suppose mo ^ m and m e hu; we need to show that hu s(m).
There must be an i such that ml ^ m < m,+1. We know by choice of ht that hl
S(TOJ); and since mz + 1 was chosen as the earliest excluder for ht, it must also
be that hl s(m). Also we have that m t +i (hur\h^)\ so, since m < m, + i,
hu =m hl. So by no choice between undivided histories, Post. 8, hu =5^ ht.
Therefore, since we have established that hl 6 s(m), we may conclude that hu
s(m) as well. D
Where are we? Without busy choosers, Os is bound to satisfy the consistency
condition if s does; but if there is a busy chooser, the consistency of s by
no means guarantees the consistency of Os. So we now know all there is to
know about the nonemptiness condition on deontic trees Os (Definition 14-l(z))
induced by a total consistent strategy s (Definition 14-2).
That is, if ro0 < mi, and if H(mi)r\Admh(s, mo) ^ 0, then Admh(s, m\) =
Admh(s, mo)nJf( m i ). What is interesting about the proof is that it depends
essentially on the principle of no choice between undivided histories, Post. 8,
that hi =mo h2 implies hi s^0 h2.
376 Strategies
PROOF. Given the hypotheses of Definition 14-1, it is trivial from the definition
of Admh(s, m) that (Admh(s, m 0 )nH( m i )) C Admh(s, mi). This containment
has nothing to do with anything except the way we explained "ideality": We
quantified over bits of strategic advice about the future. The other direction is
what is interesting: Admh(s, mi) C Admh(s, m 0 )nH( TOl ), under the hypothe-
ses. We are given that mo,mi G ho, that mo < m1, and that ho G Admh(s,
mo). To show the desired subset relation, we suppose hi in Admh(s, mi). We
need to show hi G Admh(s, mo). To show that, pick m? such that mo ^ m2,
with m,2 G (Dom(s)nhi). What is needed is that hi G 5(7712). It is given that
all of mo, mi, m2 lie on hi, with mo < mi and mo ^ 7712- Where is m2 relative
to mi? If mi ^ m2, then the hypothesis that hi G Admh(s, mi) guarantees
that hi 5(7713). So take the other case, m2 < mi. Note the proper less-than.
Now it all falls together. We know that ho (don't forget ho) and hi share a
point, namely mi, that is properly later than m,2. That is to say, ho and hi are
undivided at 7712: ho = TO2 ^1- Therefore, by the no choice between undivided
histories condition, it must be that ho and hi are choice equivalent for a at 7712:
ho = TO2 hi. But we know that ho 5(7712), since ho G Admh(s, mo) and since
mo ^ 77i2. So by the constraint of Definition 14-2 on s that 5(7712) be closed
under choice equivalence for a, we have that hi G 5(7712) as well. D
14D Remarks
It is clear that Os(m) is not very similar to s(m). Each may well have a place,
and they are closely connected, but they are different. The concept s(m) is
strictly local, but Os(m) is'partly global.
It is less clear how the kinematic condition, Definition 14-l(ra), on deontic
trees, and perhaps equivalently the no choice between undivided histories con-
dition on choices in branching histories (Post. 8) are related to the "coherence"
of plans (Thomason 1984, p. 155). Perhaps they are; it's just the "how." Both,
if they are indeed connected in the fashion indicated in 14C.4, have to do
14. Deontic kinematics and austere strategics 377
with plans that involve attending to "can do," which sounds like coherence or
consistency.
But not necessarily in the exact way that one might think. When we prove
that Os satisfies the kinematic condition, we do so for an arbitrary consistent to-
tal strategy. Even for the weirdest such you can imagine. Therefore, satisfaction
by Os of the kinematic condition can never give us a reason to judge one total
consistent strategy better than another. On the other hand, it certainly sounds
as if coherence is a proper ground on which to evaluate a strategy. This evident
tension strongly suggests the existence of deeper issues that need clarification.
Let us add that when one considers primary (nonreparational) strategies, then
the obligation concepts probably go differently. For one thing, there is then no
essential difference between Admh(s, m) and Admh(s). In primary strategies,
ideality does not depend on where you are.
In conclusion, the following evaluative remarks appear warranted. First, deon-
tic kinematics as defined in Thomason 1984 earns additional respect by making
useful connection with the austere theory of strategies. Second, and conversely,
the austere theory of strategies merits our appreciation for its ability to relate
in an interesting way to that very theory of deontic kinematics. And we should
prize both theories, simple as they are, for helping us to get a handle on how
normative ideas fit within the real-world causal order.
This page intentionally left blank
Part VI
Decidability of one-agent
achievement-stit theory
with refref
As presented in earlier chapters, stit theory starts with the stit sentence [a stit:
A] (read "a sees to it that A"), where a is a term for agents and A is any
sentence.*
Based on BT + I + AG structures (2), [a stit: A], as an achievement stit
(8G.3), is interpreted as saying that A is guaranteed true by a previous choice
of a. This chapter considers a subset of the BT + I + AC structures, namely,
those that contain no "busy choice sequences" in the sense of Def. 14:
No busy choice sequences. There is no sequence of infinitely many nonva-
cuous choices for any agent occurring within a finite time; that is, no such
sequence is bounded by moments both above and below.
We say that any BT + I + AC structure with no busy choice sequences is a
BT + I + AC + nbc structure (2). In BT + I + AC + nbc structures, doing is
equivalent to refraining from refraining from doing. (For a discussion of whether
doing is in fact equivalent to refraining from refraining from doing, see 2B.6.)
That is to say, the refref equivalence (Ax. Cone. 1)
(refref)
381
382 Proofs and models
15 A Preliminaries
We are considering achievement-stit theory with only a single agent. So all
sentences are constructed from propositional variables, using truth-functional
connectives ~ (negation) and & (conjunction), and the connective [a stit: ],
with a as the only term for agents in our language. A stit sentence is any sen-
tence of the form [a stit: A}. As usual, V (disjunction), D (material implication),
= (material equivalence), T (truth) and _L (falsity) can be introduced as abbre-
viations. In addition we introduce the following technically useful abbreviation:
(A, but a does not see to it that A).
We will use A, B, C, and so on to range over sentences, and F, S, II, and so on
to range over sets of sentences.
Although for current purposes we admit only one agent term, we present
our semantic account in terms of arbitrary BT + I + AC + nbc structures & =
(Tree, ^, Instant, Agent, Choice), 2, in order to harmonize with the rest of this
book. We use the postulates of 3 and definitions from the rest of the appendix,
adapted to BT + I + AC + nbc structures when necessary, for example, for the
meaning of "model" and the range of the structure variable 6 and the model
variable 9JI. Because of the mathematical emphasis of this chapter, we tighten
up on rigor in the matter of agents: Whereas elsewhere we use "a" sometimes as
ranging over Agent and sometimes as ranging over the special agent-denoting
terms, here we always let a be an agent-denoting term, with a its semantic
value, 3(a). Furthermore, in this chapter we let a (rather than a) range over
Agent.
Let us first present some facts to be used tacitly in subsequent discussions.
15-1 FACT. (Settled truth of achievement stit) For any m and h with m h,
m, m/h \= [a stit: A] iff 9JI, m N [a stit: A].
1
"Lai + rr" is intended as mnemonic, as follows: Logic of the achievement stit with 1
agent and the refref equivalence. See Ax. Cone. 3 for the list of axiomatic systems for which
we employ such mnemonics.
15. Decidability of one-agent achievement-stit theory with refref 383
This second fact is a corollary of Chellas's witness identity lemma, Fact 10-1.
As a special case, this fact can also be proved by applying Lemma 15-4.
2
The idea of a busy choice sequence is listed with a some somewhat more general definition
in Def. 14. A formal study of busy choice sequences can be found in Xu 1995a. Note that
the forward denseness condition is only a particular way in which a chain can be busy. We
only consider forward denseness because when considering the truth values of sentences in our
language, only forward denseness is relevant. One can find, however, other issues in which
backward denseness is more crucial. See, for example, Fact 14-9.
384 Proofs and models
15B Companions
In this section, we will consider some basic semantic features of so-called com-
panions to stit formulas, which we will not define until we obtain the companion
theorem (Theorem 15-11). But it will be helpful to present a description of it
at the beginning so that the reader may have some idea of where our discus-
sion is going. Let 9m be any BT + I + AC model such that m, m t= [a stit: A]
with witness w, and let M = {m1: m' i(m) & m' = m}. Clearly, m, M N
[a stit: A}. We will show in the companion theorem that for each sentence C,
either m, M N [a stit: A&Ca] or m, M ~[a stit: A,Ca\. Whichever is
true together with [a stit: A] in M will be called a "companion" to [a stit: A}.
Note that in this section, "structures" and "models" always mean BT + I+AC
structures and BT + I + AC models, and we do not require that these structures
and models satisfy the property of no busy choice sequences. Let us start with
some simple lemmas.
386 Proofs and models
PROOF. Suppose that w < w'. Consider any TO' 6 i\>w>, that is, m' i and
w' < m'. Assume that m' ^ m. Then by the axiom of choice there are two
histories h and h' such that {w, w', m} C h and {w, w', m'} C h', and hence
h' =w h by definition. This implies, by no choice between undivided histories,
h1 =aw,h and hence m' =, m, that is, m' 6 M. It follows that i| >UJ ' C M. (ii)
Suppose that w' ^ w. By (%), if w' < w, then i\>w C M', and hence M C M'.
If to' = w, then m' =,, m iff m' =, m for every m' i, and hence M = M'.
n
15-5 LEMMA. (Some sufficient conditions for stit) Let SOT be any model in
which w < m, i = i( m j and M = {m": m" e z & m" = m}, and let >1 be any
sentence. Then
PROOF, (i) Suppose that SOT, M N A and for some m' M, SOT, m' N [a stit: .A]
with witness w'. Setting M' = {m": m" 6 i & m" =,/ m'}, we then have
Since m' M, w < m', and hence by no backward branching, either w < w'
or w' ^ u/. If w < w', then by Lemma 15-4(i), i\>w> C M, and hence SOT, i|>u/
N .A, contrary to (1). Hence it must be true that w' ^ w, and then by Lemma
15-4(ii), M C M'. It follows from (1) that 971, M N [a stit: A], (ii) Suppose
that !Ut, i\>w t= yl and for some m' i\>w, 3H, m' N [a stit: A] with witness
w'. Similarly, setting M' = {m": m" i & m" =^, m'}, we also have (1), and
applying no backward branching, we have either w ^ w' or w' < w. By the
transitivity of ^, w ^ w' implies i\>w' C i\>w, which in turn impliesm,i\>w'
t= A since SOT; i\>w N A . It follows from (1) that w' < w, and hence by Lemma
15-4(%), i|>, C M'. Then DOT, i|>tu N [a stit: A] by (1). D
15-6 LEMMA. (Concerning [a stit: A&B]) Let 9H,TON [a stzt: A] with witness
w, and let i = i(m). Suppose that 9JI, M 1= 5, where M = {m': TO' 6 i & m'
= ^ m}. Then SOT, m 1= [a stit: A&B] with the same witness.
PROOF. Suppose that 271, M 1= B. Then by hypothesis, Jt, M *= a&.B and 9Jt,
i|>UI ^ 4, and hence mOT, i|>m J^ ^&5. It follows that 01, m N [a stit: A&B]
with witness w. D
15. Decidability of one-agent achievement-stit theory with refref 387
15-7 LEMMA. (Concerning Aa) Let m be any model in which w < m, i = i(m)
and M = { m / : m ' E i & m ' s ^ m } . Suppose that A is any sentence such that
m, M F= A and M, M F [a stit: A]. Then m, i >w F= Aa. In particular, if m,
m 1= [a stit: A&B] with witness w, and if M, M F [a stit: B], then Wl, i>w F=
Ba and M, M F= [a stit: /4&Ba].
15-8 LEMMA. (01 refraining lemma) Let M be any model, and let A and B be
any sentences such that M, m F= [a stit: A&~[a stit: B]] with witness w, and
let i = i( m ). Suppose that for some m' i|> w , M, m' F= [a stit: B] with witness
w'. Then w < w'.
PROOF. Let M' = {m": m" 6 i & m" =, m'}. Then M, M' F= [a stit: B].
Since m' E i|>w,, w < m' and w' < m'. Then by no backward branching, either
w' < w or w < w'. Suppose for reductio that w' < w. Then i|>w C M' by
Lemma I5-4.(i). It follows that M, m F [a stit: B] since m i\>w. But by
hypothesis we have M, m F= ~[a stit: B], a contradiction. From this reductio
we conclude that w < w'.
15-9 LEMMA. (Concerning Aa and stit) Let M be any model, and let A and
B be any sentences. Then the following hold:
i. If M, m F [a stit: A&B a ] with witness w, then M, m F= [a stit: A] with
the same witness;
ii. If M, m F= [a stit: A&~[a stit: A&B 0 ]] with witness w, then M, m F [a
stit: A] with the same witness.
PROOF. (i) Suppose that M, m F= [a stit: A&.Ba] with witness w. Then, setting
i = i(m) and M = {m': m' i & m' =aw m}, we have M, M F A&5a and M,
i|>w A&Ba. Hence by Lemma 15-7, M, i\>w F= Ba, and hence M, i|>w, F A.
It follows that M, m F= [a stit: A] with witness w. (ii) Suppose that M, m F= [a
stit: A&~[a stit: A&B a ]] with witness w, and let i = i(m) and M = {m': m'
i & m' = awm}. Then
and either M, i|>w F A, or M, i\>w ~[a stit: A&Ba}. Consider the latter
case. There must be an m' i\>w such that M, m' t= [a stit: A&Ba} with
witness w'. By applying Lemma 15-8, we have w < w', and hence i\>w' C i\>w.
388 Proofs and models
Since by (i), M, m' F= [a stit: A] with witness w', it follows that M, i\>w A.
So in both cases, we have M, i\>w A. It follows from (2) that M, m F= [a stit:
A] with the same witness w.
Then there must be an m' i\>w such that M, m' F= [a stit: A&B] with witness
w'. By Lemma 15-8, w < w', and hence, i\>w> C i\>w. Let M' = {m": m"
i & m" = aw, m'}. Then M' C i|>u, and hence by Lemma 15-6, M, m' F [a
stit: A&.Ba] with witness w', and hence by Lemma 15-9(i), M, m' F= [a stit:
A] with witness w', it follows that
We show as follows that M, M F ~[a sizt: A]. Suppose for reductio that there
is an m* C M such that M, m* F [a stit: A] with witness w*. Set M* = {m":
m" C i & m" = aw. m*}. Then M, M* F [astit:A]. Since m* M, w < m*,
and hence either w < w* or w* < w by no backward branching. Now if w <
w*, then by Lemma 15-4(ii), M* i\>w, and hence by (3) and Lemma
15-6, M, M* F= [a stit: A & B ] , contrary to (3). If w* < w, then we have i\>w C
M* by Lemma 15-4(i), and hence, since M, M* F [a stit: A], M, i\>w F= [a stit:
A], contrary to (4). This reductio shows that M, M F= ~[a stit: A]. It follows
from (4), (3) and Lemma 15-6 that M, M F= [a stit: ~[a stit: A]&Ba}.
By Lemma 15-9(i), we have M, m' F [a stait: A] with the same witness w'. This
and our hypothesis imply
15. Decidability of one-agent achievement-stit theory with refref 389
Since m' E M, w < m'. Hence by no backward branching, either w < w', or
w' < w, or w = w'. If w < w', then by Lemma 15-4(i), i\>w> C M, and hence
by (6), m, i\>w> F A, contrary to (7). Similarly, if w' < w, then by Lemma
15-4(i), i\>w C M', and hence by (6), M, i\>w F= A, contrary to (7). We then
conclude that w = w'. Since m'' =aw, m (i.e., m' M), it follows that M = M',
and hence by (5), M, M F [a stit: A&B"]. By the foregoing proof, we know
that if M, M F [a stit: A&Ba], then M, M F ~[a stit: A & B a ] , and hence m,
M F= [a stit: a&~[a stit: A&Ba}} by Lemma 15-6.
15-12 LEMMA. (Companions and witness) Let M be any model in which w <
m, i = i(m) and m' e i >w, and let M, m F= [a stit: A&Ca\ with witness w,
and M, m' (= [a stit: B&~[a stit: B&(Ca]] with witness w'. Then w' < w.
PROOF. Since w < m' and w' < m', either w' < w or w < w' by no backward
branching. Suppose for reductio that w < w'. Then, setting M = {m": m" i
& TO" =aw, m'}, we know that M C i >w. Since m, i\>w F= Ca by hypothesis
and Lemma 15-7, it follows that m, M F Ca. By Lemma 15-9 (ii) we have that
M, m' F= [a stit: B] with witness w', and hence by Lemma 15-6, M, M F= [a
stit: B&C"]. But we know by hypothesis that M, M F ~[a stit: B&Ca], a
contradiction. It follows from this reductio that w' < w.
15-14 COROLLARY. (Validity of A9) For any model 971, 97T N A9.
and let i = i (m) and M = {m': m' i & m' =* m}. Then m, M \= Ca, and
hence by Lemma 15-7, 971, i >w 1= <7Q and 97t, i|>tu ^ ~[a stit; ^4&[a stit; 5&
~[a stit; 5&C"*]]]. Hence there must be an m' 6 i| >U) such that
and by Lemma 15-9 (ii), 971, m' t= [a stit; B] with witness w'. Set M' = {m":
m" 6 i & m" =, m'}. Then 971, M' 1= [a stit; 5]. Applying Lemma 15-12
to (8) and (9), we have w' < w, and hence by Lemma 15-4^, i\>w C M'. It
follows that 971, m E [a stit; B]. D
PROOF. Let M, m \= [a stit: A] with witness w and counter m', and let i =
Since
Suppose for reductio that M, m [a stit: ~[a stit; ~[a stit; A]]]. Then we have
by (10) and Lemma 15-7 that
15. Decidability of one-agent achievement-stit theory with refref 391
15-16 LEMMA. (Refraining from refraining implies doing) Let 971 be any
BT + I+AC + nbc model. Suppose that 971, m t= [a stit: ~[a stit: ~[a stai:
4]]]. Then 971, m N [a sizi; 4].
PROOF. Let 971, m 1= [a stit: ~[a stai: ~[a stit: A]}] with witness w and counter
m*, and let i = i (m ) and M = {m": m" i & TO" = TO}. Then 97t, m* (= [a
siit' ~[a stit: A]] with the witness w* ^ w by Lemma 15-8, and hence
there must be an mo G M such that 97i, mo N [a sfoi: /IJ with witness WQ and
counter m0; for if 071, M 1= ~[a stit: A], then 971, M N (~[a stt .4])Q by (14),
and hence by Lemma 15-7, 971, i\>w t= (~[a stit: A ] ) a , contrary to (13). Since
m0 M, M = {m": m" E i & m" =TOO};and since w < mQ and w0 < TOO,
either w < w0 or w0 ^ w by no backward branching. If w0 ^ w, then, setting
M0 = {m": m" G i & m" =aw0 mo}, we have M C M0 by Lemma 15-4, and
hence, since 971, Mo1= [a stit: A], M, M F= [a stit: A], contrary to the assumption
of our reductio. It follows that w < WQ. Let M0 = {m": m" i & m" =aw0
mo}. In general, we can inductively define two sequences mo, m1, ... and w0,
w1 ... such that for each k > 0, mK + 1 M'k = {m": m" E i & m" =awk mk}
and M, mk + 1 F= [a stit: A] with witness wk + 1 > Wk > w and counter m'k+1.
The proof of this is similar to the earlier proof. But {w, WQ, w\, ...} is a busy
choice sequence, contrary to our hypothesis. D
15-17 THEOREM. (Soundness theorem) For every sentence A, and for every
BT + I + AC + nbc structure 6, h- A in La1 + rr only if 6 f= A.
392 Proofs and models
Note that in this remark, when E contains [a stit: A] and all its F-companions
in $, E is not necessarily a F-companion set for [a stit: A]. This is simply
15. Decidability of one-agent achievement-stit theory with refref 393
and show that F ([a stit: B] = E). On the one hand, we obtain F(E D [a stit:
B]) by A4, A8, and Rl. On the other hand, we obtain F ([a stit: B] D [a stit:
A}) by T7 and T15, and for each i with 1 < i < k, F ([a stit: B} D [a stit: A
&C?]) by applying T15 m times and T7 k l times; and for each j with 1 <
j < m, F ([a stit: B] D ~[a stat/ A&Daj]) by A2. It follows that F [a stit: B]
-D E.
and consequently
PROOF, (i) By Lemma 15-20 and A2, we need only to show one direction of
each equivalence. Suppose that C F and I- [a stit: A&Ca]. Since F [a
stit: B] D Ca by A2, it follows from R3 that E F [a stit: B&C"]. Suppose
next that E E ~[a stit: A & C a ] . Then E T [a stit: A&~[a stit: A&Ca}] by
A8 since E T [a stit: A]. We show as follows that E F [a stit: B&~[o; stit: B
& C a ] ] . By Lemma 15-20, F [a stit: B & < C a ] D [a stit: A&C a] . So E T ~[a
stit: B & C a ] , and hence, T [a stit: B] D ~[a stit: B&C"*]. It follows from R3
that E h [a stit: B&~[a stit: B & C a ] ] . (li) By (i), we have that F [a stit: B]
D [a stit: B&Cai] for all i with 1 < i < K, and that F [a stit: 5] D [a stit: B
&~[a stit: 5 &>"]] for all j with 1 <j < m. Then we can complete the proof
by applying T7, T15, and A4.
A direct consequence of Lemma 15-19 and Lemma 15-21 is the following. Let
E be the F-companion set for [a stit: A] in an MCS $ with C\, ..., Ck, and
DI, ..., Dm to be, respectively, all the F-pos-companion roots and all the F-
neg-companion roots of [a stit: A] w.r.t. $. Let BQ, BI, B%, ... be defined as
follows:
and let = DC([a stit: Bn + i\) for all n > Q. Then E = E0 = EI = .... That
is to say, for each n > 0, E is the F-companion set for [a stit: Bn] in $ (with
[a stit: Bn + i] to be the characteristic sentence of E w.r.t. Bn), and [a stit:
Bn] has the same F-pos-companion roots and the same F-neg-companion roots
w.r.t. E as [a stit: A] has.
Lemma 15-22 (ii) shows that F-companion sets have a certain exhaustive char-
acter. A consequence of Lemma 15-22 (iii) is the following: Suppose that E is
both a F-companion set for [a stit: A] and a F-companion set for [a stit: A'].
Then [a stit: A] and [a stit: A'} have the same F-pos-companion roots and F-
neg-companion roots w.r.t. E, and hence, if [a stit: B} and [a stit: B'] are the
characteristic sentences of E w.r.t. A and A' respectively, then for some k, m >
0,
15-23 LEMMA. (Neg-companion roots and stit sentences in companion sets) Let
E be any F-companion set, and let [a stit: B] be any characteristic sentence
396 Proofs and models
In our proof of the completeness and the finite model property of Lai + rr,
we need to associate each nontheorem A of Lai + rr with a F satisfying the
following property: F is equivalently closed under conjunction iff for every C,
C' F, there is a D e F such that h D = C& C'.
The next three lemmas report combinatorial facts that we use later.
PROOF. Suppose for reductio that there is a sentence C G F such that one of
the following holds:
15-27 LEMMA. (Used for Lemma 15-34 and Lemma 15-35) Let be any set
of sentences, and let B* = 4 & ( & i ^ . , ^ m ~[a siii: 4&.D"]), where A, DI, ...,
Dm are any sentences. Suppose that h ~[a stit: B*]. Then
PROOF. We can simply obtain (i) and (ii) by applying the definition of F-
alternative, Lemma 15-21, and Lemma 15-22 (Hi), and obtain (Hi) by Lemma
15-22 (i). D
PROOF, (i) Let A be any sentence such that E h [a stit: A] and II h [a stit: A],
and let C be any sentence in F. Suppose that E h [a stit: A&, Ca\. Then by A2
and T9, S h A&C&[a stit: A & C} and S h Ca. Thus there is a D e F such
that h D = A&C and hence E h >&[a stit: D]. Then by Lemma 15-28(Hi),
n P Da and II h Ca. Since II h A& C by A2, II h- >&[ stit: D] by Lemma
15-22(ii). It follows that II h [a stit: A&C}&~[a stit: C], and hence II h [a
15. Decidability of one-agent achievement-stit theory with refref 399
stit: j4& Ca] by A6. The other direction of the proof is similar, (ii) Let $ be
any MCS including E, and let ' be the F-companion set for [a stit: A] in $.
By Lemma 15-24, E contains all the F-companions to [a stit: A] in $. Setting
^ to be any MCS including II, we know for the same reason that Ii contains
all F-companions to [a stit: A] in ty. According to (i), each F-companion to [a
stit: A] in $ must be in \? and vice versa. It follows that E' C EnII. D
15-30 LEMMA. (An inclusion relation among companion sets concerning a par-
ticular stit sentence) Let E and E' be F-companion sets in an MCS <J>, let II
be a F-alternative to E such that E' C EnII, E h [a stit: A], II h [a stit: A]
and ' V- [a stit: A], and let [a stit: B] be any characteristic sentence of '.
Suppose that F is equivalently closed under conjunction, A F, and E" is the
F-companion set for [a stit: A&[a stit: B}} in $. Then ' C E" C Enll.
PROOF. Since E" h [a stit: B] by A2, it is then clear that E' C E". To show that
E" C Enll, we need only to show that E and II contain all the F-companions
to [a stit: A&c[a stit: B]} in $. To that end, we first note that since A F and
E h [a stit: A] and II h [a stit: A], it follows from Lemma 15-24 and Lemma
15-29 that
It follows from E' C Enll that E' P [a stit: A] D [a stit: A&Da] (otherwise
we would have E h ~[a stit: A] and II h ~[a stit: A]), and hence by Lemma
15-25, E' y [a stit: B&Da}. Then ' h ~[a stit: B&Da] by Lemma 15-21 ft),
and hence by A5, E' h ~[a stit: [a stit: B]&Da]. It follows from Rl and T4
(by contraposition) that
Hence by (18), we have E h ~[Q stit: A&[a stit: B]&Da] and H h ~[a stit: A
&[a stit: 5]&Z) a ]. It follows from Lemma 15-26 that E and H contain all the
F-neg-companionsand hence all the F-companionsto [a stit: A&[a stit: B]}
in $, that is, E" C EnH. This completes the proof. D
PROOF, (i) Assume that RA(, II). We first show that I I P [a stit: A], which,
by definition, is true when there is no F-neg-companion to [a stit: A] in E.
Assume that D\, ..., Dm (m ^ 1) be all the F-neg-companion roots to [a stit:
A] w.r.t. E. Suppose for reductio that E h [a stit: A}. Then, on the one hand,
we know by definition that II I- V i< s m [a stit: A&D?]. On the other hand,
Lemma 15-29 provides H t- &i^ is^m ~[a stit: A&cD?}. Hence II is inconsistent,
contrary to our assumption of consistency of H. It follows from this reductio
that H Y- [a stit: A}. We next show that H Y- A. Suppose for reductio that H h- A.
Then by Lemma 15-22(ii) and II Y- [a stit: A], Ft h v4&~[a stit: A], and hence E
h ~[a stit: A] by Lemma 15-28 (in), contrary to our assumption of consistency
on E. It follows from this reductio that E V- A. (ii) Suppose that RA(S, II),
~[a stit: A] F and that every C 6 F is a F-pos-companion root of [a stit:
A] w.r.t. E. Then by RA(E, II), II h ~[a stit: A}. Suppose for reductio that II
Y- [a stit: ~[a stit: A}}. Then II h ~[a stit: ~[a siii: 4]] by Lemma 15-22^,
and hence by Lemma 15-28(Hi), E h ~[a stit: A], contrary to our assumption
of consistency on E. It follows from this reductio that II h [a stit: ~[a stit: A]].
It is easy to see that (Hi) follows from (i), (ii) and related definitions.
15F Semi-ref-counters
From our discussion in the last two sections, one can see that F's property of
being equivalently closed under conjunction is important for our proof. We now
show that for every sentence A, we can associate a finite set F that has this
property.
Let A be any sentence, and let I" be the set of all subsentences of A, and let
F* = F'U{~[a stit: C]: [a stit: C] e F'}. We define FA = {5j&...&5 n : BI,
..., Bn are distinct sentences in F*}. In the following lemma, for any set X =
{Bi, ..., Bn} of sentences, &X = Bi&...&B n .
15-32 LEMMA. (Finding a set that has the desired properties) Let A be any
sentence. Then TA is a finite set that is closed under subsentences, closed under
negated stit sentences, and equivalently closed under conjunction.
PROOF. It is easy to see by definition that TA is a finite set that is both closed
under subsentences and closed under negated stit sentences. To show that TA
402 Proofs and models
A direct consequence of Lemma 15-33 is the following: Let n > 0, and let
EQ, > E n and HO, ..., IIn be F-companion sets such that for each i with 0 ^
i ^ n, .R r (j, II,), and for each i with 0 ^ i ^ n 1, IIj C EI+ I. Suppose
that there are n F-neg-companions to [a stit: A] in EQ. Then Hn must be a
F-ref-counter to E n relative to [a stit: A], and there can be no E and II such
that -Rp(E, LT) and nra C E. This feature of F-semi-ref-counters corresponds to
a semantic feature of Lai + rrthere is no busy choice sequence in its models.
The following three lemmas establish some sufficient conditions for F-semi-
ref-counters.
PROOF. We first show that H is a F-companion set for [a stit: ~[a stit: B}} and
is a F-alternative to . To that end, we first note that by T7, II h [a stit: ~[a
stit: B]}. Consider any D = D} with 1 ^ j < m. We know by Lemma 15-21 (ii)
and T16 that
Hence h ~[a stit: ~[a stit: B]&Da} by Rl. It follows from hypothesis and Rl
that
Hence by Lemma 15-19, II is a F-companion set for [a stit: ~[a stit: B}}. It is
easy to check that H is a F-alternative to E. Next, since II h [a stit: ~[a stit:
B]bC], and since h ~[a stit: C} by T6, it follows from Rl that H h [a stit:
~[a stit: B&~[a stit: C]]&.Ca]. Because B is the sentence v4&(7&;(&i^j^ m
~[a stit: A k D f } ) , setting B* = ^ & ( & i < j < m ~[a stit: A&DJ*]), we have II h
[a stti: ~[a siii: B'&^j&C 1 0 ], and hence by T14, U\-[a stit: ~[a stti: 5*]
& (7]. It follows by A2 that II h ~[a sfit: B*], and hence by Lemma 15-27, II
h [a sttt: v4] D V i 5 S j = m ta s^; ^&-0"]- If there is no F-neg-companion to [a
stit: A] in E, then B = A& C, and hence by T6, II h [a stit: ~[a stai: ^4& <7a]
& C10]. It follows from T14 that U\-[a stit: ~[a sttf; ^.] & C"*], and hence III-
[a stit: ~[a stzi: ^4]] by T7. In particular, if [a stit: A] e F, #r(S, II) follows
by definition.
PROOF. Let C"i, ..., CA and Z?i, . . . , D m be, respectively, all the F-pos-
cornpanion roots and all the F-neg-companion roots of [a stit: A & [a stit: B}]
w.r.t. $. We first prove that II' is a F-companion set for [a stit: ~[a stit: B']k
[a stit: B}] and is a F-alternative to E'. By R3, it is sufficient to show that for
each C = Ct with 1 ^ i ^ k, and each D = D m with 1 ^ j ^ m, the following
hold:
Applying (29) and Rl to T16, we have I [a stit: ~[a stit: ']&> Q ]; and
applying (30) and Rl to Til, we have I- ~[a stit: [a stit: ]&>"]. It follows
from T4 and Rl that I- ~[a stit: ~[a stit: B']&[a stit: 5]&D a ]. This com-
pletes our proof of (28). Hence II' is a F-companion set for [a stit: ~[a stit: B'}
&[a stit: B]] and a F-alternative to E'. It is easy to see that E C E'nll' since
E' h [a stit: B} and II' h [a siii: 5].
We show next that #r( s / > n / )- Since A e F and E' h A by A2, ' h [a stai:
A] by Lemma 15-22 (h). Let B* and C1* be the following sentences:
Then by Rl, h [a stit: B1} = [a stit: 5*&[a stit: 5]&C*], and hence by
hypothesis and A2, H' h ~[o; stit: 5*&[a siii; 5]&C7*]&[o! siif; B]. On the
one hand, by applying A4 (by contraposition), we have II' h ~[a stit: B*]V
~[a stit: [a stit: B]&C""]. On the other hand, by applying Lemma \5-2l (ii)
and T10, we have II' h [a stit: [a stit: 5]& C1?] for all i with 1 < i ^ fc (since
IT h [a stit: B } ) , and hence by A4, II' h [a stit: [a stit: B}&C*}. It follows
that II' I- ~[a stit: B*} (when there is no F-neg-companion to [a stit: A] in E',
the same argument here, replacing B* by A, yields II' I- ~[a stit: A}). Hence by
Lemma 15-27 (substituting A&[Q stit: B] for A in the lemma), we know that
15. Decidability of one-agent achievement-stit theory with refref 405
Since II' h [a stit: B] and h [a siz'i: 4]&[a sizi: B\ D [a szi: A&[a stit: B}}
by A3 and A4, it follows that
We also know by Rl, T4, and A5 that for each j with 1 ^ j ^ TO,
But for each such j, h ~[a sz't: 5&.D"] by hypothesis and Lemma 15-21 (ii),
and hence IT h ~[a siif: 5&D] by C n'. It follows from (31), (32), and
(33) that
We have shown that ' h [a sz't: .4], and hence by Lemma 15-24, ' contains
all the F-companions to [a stit: A] in $. Since .K [a stit: A], we know by
Lemma 15-26 that D I , ..., Dm are all the F-neg-companion roots of [a stit: A]
w.r.t. E'. Hence by definition, R$(Z', IT). D
and II' = DC([a stit: ~[a stit: 5']&C"]) is consistent, where C' = fci^j^fc
C? and Ci, ..., Ck are all the F-pos-companion roots of [a stit: A&[a stit: B\]
w.r.t. $. Then #r(E', n') and E C E'nn'.
PROOF. Since A e F and E' h A by A2, it follows from Lemma 15-22(ii) that
' h [a sizi: yl]. By Lemma 15-34 we know that II' is a F-companion set for [a
stit: ~[a stit: B'}} and is a F-alternative to E', and also, setting D I , ..., Dm to
be all the F-neg-companion roots of [a stit: A&.[a stit: B}] w.r.t. <&, we have
by A2, A3, and A4. Consider the sentence C given in the hypothesis. By
Lemma 15-21 (ii), we know that
406 Proofs and models
and hence IT h [a stit: 5] by A9. That is to say, E C II', and hence C E'n
II'. Since H' h [a sii: 5], it follows from Lemma 15-23 and A2 that II' h ~[a
stit: S&.D"] for all j with 1 ^ j < m. Hence (34) and the same argument in
the proof of Lemma 15-35 will yield R (', II'), which completes the proof. D
PROOF. Let [a stit: B] be the characteristic sentence of w.r.t. A, and let Ci,
..., Ck be all the F-pos-companion roots of [a stit: A] w.r.t. , and let II =
DC([a sirt: ~[a stit: B]b.C]), where (7 = &ioo C7f. We show as follows
the consistency of II. By Lemma 15-21 (ii) and T6, h [a stit: B] = [a stit: B&t
Ca], and hence by T20 and T6,
It follows from Lemma 15-37 that [a stit: ~[a stit: 5]&(7], and hence II, is
consistent. Hence we can complete our proof by applying Lemma 15-34.
We first show that (38) holds. Suppose for reductio that (37) holds. By A2 we
have h [a stit: B'} D [a stit: B] and h [a sta!: 5] D Ca. It follows from R3
that E' h [a stit: B'&C01}, and hence by Lemma 15-21(%), S' h [a stit: Ak[a
stit: 5]&C a ], a contradiction. Hence (38) must hold. Let Ci, ..., Ck be all
the F-pos-companion roots of [a stit: A&[a stit: B]} w.r.t. $. As we did in the
proof of Lemma 15-38, we can easily show the consistency of
Setting II' = DC(>*), we have
by Lemma 15-36, which completes the proof.
PROOF. Applying Lemma 15-40, there are E' and H' such that R(E', II') and
Let [a stit: B] be any characteristic sentence of E". From the
proof of Lemma 15-40 we know that we can choose E' to be the F-companion
set for [a stit: A&[a stit: B}] in $. It follows from Lemma 15-30 that
408 Proofs and models
C5 for each m E I, and each w, w' Tree I with w < w' < m, and each &
only
C6 for each w 6 TreeI, each yl F, and each fc = 1, 2, <7fc(w) ^ ^4 only if
there is an m G I such that w < m, Qk(w) C /(m) and ~^4 f ( r n ) .
In our proof of the completeness and the finite model property, we will convert
some element of Kr into a model. Cl C5 are designed to ensure that the model
will satisfy various conditions. The subset I of Tree represents the last instant
in the model, as can been seen from Cl and C2. Each m I is associated with
an MCS f ( m ) representing the set of sentences true at m, and each w S Tree I
is a moment where a has two possible choices, represented by g\(w) and g-z(w)
respectively. C3 and C4 imply that for each w Tree I and each m 6 I with w
< m, exactly one of g\(w) and gi(w) is included in /(m), which represents the
condition that for each w Tree I and each m I with w < m, the maximal
chain passing through w and ending at m belongs to exactly one possible choice
set for a at w. Finally, C5 represents the no choice between undivided histories
condition. An important feature of Kr is given by the following lemma.
PROOF. Let us fix n = |F|. Suppose for reductio that there is a chain w\ < ...
< wn(n + i) < w n ( n +i) + i < m in Tree with m E I. Then for all k with 1 ^ k
^ n(n + l) + l, we have Wk Tree I by C2, and also Rr(gi(wk), 92(wk)) by
15. Decidability of one-agent achievement-stit theory with refref 409
15-43 COROLLARY. (Upper bound of size of models) Let n = |F|. Then for all
(Tree, ^, I,/, ffi, 52} K r ,
PROOF. Since by C2, no element of I has any successor and every element of
Tree I has at most 2n successors, the displayed inequality follows from Lemma
15-42.
We define a proper extension p' = (Tree', ^', I', /', g\, g'2) e Kp of p as follows:
Since k ^ |F|-1 (S h [a sizi: 4]) and n ^ |F|, there are at most 2 x |F| differ-
ent successors of w in Tree'. It can be easily shown that p' satisfies all other
conditions to be an element of Kr; details are omitted.
CASE 2. w < m for some w Tree. Let there be n* elements w' of Tree such
that w' < m. Thus by C2, we have w\ < ... < wn* < m and wi ^ w' for all
w' Tree. We may assume, without loss of generality, that gi(wk>) C f ( m ) for
every k' with 1 ^ fc' ^ n*. Consider wi. For the first subcase, suppose that
gi(wi) I- [a stit: A}. Then 52(^1) l~ ^4> for otherwise [a stit: A] and m would
not constitute a counterexample to C7 in p. It follows from Lemma 15-22 (ii)
and Lemma 15-28(iii) that gi(wi) h [a stit: A]. Applying Lemma 15-39 with
15. Decidability of one-agent achievement-stit theory with refref 411
ffi(wi) = E and g 2 ( w i ) = II, we obtain ' and II' such that #r(', n/ ) and
' 5 i ( w i ) n # 2 ( w i ) . Since wi < w' for all / e TVee, C4 implies that ' C
/(m') for all m' I. Then we repeat step (42) (replacing E and II in (42) by E'
and II' here respectively) and define a proper extension p' of p by setting Tree',
I' and /' the same way as in (43), and setting
Suppose now for the second subcase that 51(101) Y- [a stit; A]. This subcase has
its own two subcases. Consider wn*.
CASE (a). g\(wn-) V- [a stit: A}. Applying Lemma 15-40 with gi(wn*)
E and /(m) = $, we obtain E' and II' such that E' C /(m), #r(', n/ ) and
9i(wrf) C E'nll'. Then we repeat step (42) (replacing E and II by E' and II'
respectively) and define a p' by setting Tree', I' and /' the same way as in (43),
and g\ and g2 as in (44), and setting
CASE (b). g\(wn*} \- [a stit: A]. Then it is easy to see that there is a k* such
that 1 ^ fc* < n*, gi(wk*) Y [a stit: A], g\(wk" + 1) l~ [ex stit: A] and g^(wk* + 1)
h A (or else m and [a stit: A] would not constitute a counterexample to C7 in
p). Then by Lemma 15-22(ii) and Lemma l5-28(iii), g2(w ' + 1) I- [a stit: A].
Then, applying Lemma 15-41 with g\(wk) = E", <?i(wfe- +J = E, g2(wk' + i)
= n and f ( m ) = $, we obtain S' and H' such that R$(S', H'), gi(wk'] C S'n
II' and S' C ffi(wfc' + i)n52(wjfc* + i), and hence by C4, E' C f ( m ' ) for all m' 6
I with Wfc + i < m'. We can thus define a p' just as we did in case (a), except
that we now set
It is easy to see that p', so defined, satisfies all the conditions C1-C6. This
completes the proof. D
This process will eventually provide us with a pk at the end of the sequence,
that is, a p& in which no counterexample to C7 can be found, for otherwise we
would have that for each n ^ 0, pn G K r x and \Treen\ > n, which contradicts
Corollary 15-43. So let p be this pk- This completes the proof.
Now we are ready to prove the completeness and the finite model property of
Lai + rr, both of which are included in the following theorem.
15-46 THEOREM. (Completeness and finite model property) Let A be any con-
sistent sentence. Then there is a finite BT + I + AC structure (which is therefore
a BT + I + AC + nbc structure) & = (Tree, <, Instant, Agent, Choice), and a
model yjl = {&, 3} on 6 such that
i. 9Jt, mo \= A for some mo G Tree;
ii. for each w G Tree, Choice^ contains at most two elements.
PROOF. By Lemma 15-45, there is a sequence (Tree1, <', I, /, gi, <?2) G Kr>i
satisfying C7 and A G /(mo) for some mo G I. In order to construct a structure
6, we may first need to extend (Tree1, ^'} to a (Tree, <} in which each maximal
chain is as long as all the others. Since |7Vee'| is finite, we can fix n to be the
length of the longest chain in Tree', that is, n = max{|ft|: h is a maximal chain
in Tree'}. For each maximal chain h in Tree', we know by C2 that there is a
unique element m G I such that h = {w: w G Tree' & w ^ m}. Let us use m^
to denote such m. If there is any h in Tree' such that \h\ < n, then we choose
some wi, ..., w n _|ft| G W3>ee' and add them into Tree', earlier than m^
and Zoter than all w G ft {m/,}. Finally, we obtain (Tree, ^) such that Tree'
C TVee, ^' C ^, and ^ is still a partial order on Tree subject to the historical
connectedness and no backward branching conditions, and such that for each
maximal chain ft in Tree, \h\ = n. Details are omitted.
For any maximal chain ft in (Tree, ^}, we will still use m^ to denote the
unique m G I such that ft = {w: w G Tree & w ^ m}. Since Tree is clearly
still finite (in fact, the inequality in Lemma 15-44 still holds), there must be
a WQ G Tree such that WQ ^ w for all w G Tree. Now, letting a have an
interpretation denoted by a, we can form a BT + I + AC + nbc structure 6 =
(Tree, ^, Instant, Agent, Choice) by setting the following:
Instant = {i\, ..., in}, where i {WQ}, and for each j with 1 < j ^ n,
i1+ i = {w: w G Tree & 3w'(w' G i} and w is a successor of w')};
Agent
Note that since all histories in (Tree, <) are equal in length, in I, and it
is easy to see that Instant, so defined, satisfies unique intersection and order-
preservation conditions. Let us set i to be the last instant in Instant, that is, i
15. Decidability of one-agent achievement-stit theory with refref 413
= in. Note also that for any w G iL)(Tree Tree'), H^ is a singleton. Note,
finally, that for each w G Tree, either Choice^ is a singleton (when w & iL)(Tree
Tree')), or it contains exactly two elements (when w G Tree' i).
We claim that Choice, so defined, satisfies the no choice between undivided
histories condition, Post. 8. We justify this claim as follows. Let us define
a function g from {(m, w): m G i & w G Tree.1 & w < m} to the set of all
F-companion sets as follows: For each m G i and each w G Tree' such that w <
m, g(m, w) = g i ( w ) if gi(w} C /(m), and g(m, w) = g2(w) if g2(w) C /(m).
Suppose that w G Tree and /i and /i' are distinct histories in H^w) such that w'
G ftnft' for some w' > u>. Then, since ft and ft' are distinct, H^w) and .H"^/)
are not singletons, and hence by an earlier remark, w, w' G Tree'. By definition
g(mh, w) C /(m/i), and hence g(rrih, w) C gi(w')(~\g2(w') by C5. It is also
true by definition that <?(mv, w') Q /(m/^) and <7i(u/)ri<72(w/) c g(m/,<, w').
It follows that
for every TO' 6 i with TO' = m, and C1 ^ /(TO") for some m" z|> w . Hence
by C6, (46) and C4, gi(w) h (7 and 0 2 (w) F (7. Now, if 9l(w) h ~[Q sizf: C1],
then by Lemma 15-28(Hi), gi(w] \~ C, & contradiction. It follows that g i ( w ) V-
~[a stit: C], and hence by Lemma 15-22(H), g\(w) \- [a stit: C]. Since g i ( w )
C /(m), [a stit: C] 6 /(m), which completes the proof. D
Recall the refref equivalence, Ax. Cone. 1, that is, doing is equivalent to re-
fraining from refraining from doing.*
(refref)
It turns out that this equivalence is valid for every BT + AC structure for dstit,
and it is valid for every BT + I + AC structure for astit iff the structure con-
tains no busy choice sequences, and is therefore a BT + I + AC + nbc struc-
ture; see chapter 15 and chapter 18 for details. (BT + AC, BT + I + AC, and
BT + I + AC + nbc structures are defined in 2.) We showed in chapter 15 that
the astit theory with a single agent and the refref equivalence (i.e., the set of all
sentences, with a single agent term, valid for all BT + I + AC + nbc structures) is
decidable. The purpose of this chapter is to remove the no-busy-choice-sequence
restriction and give an axiomatization for the resulting basic logic of astit.
We present an axiomatic system Lai in 16A that is sound and complete
w.r.t. BT + I + AC structures. Soundness is proved in 16B, and completeness
is proved in 16E. Our proof uses the notions of companions and companion
sets introduced and extensively used in chapter 15, and uses facts established
in 15A-15E, which continue to apply even in the absence of refref. In 16E
we also establish some other results, for example that every satisfiable astit
sentence is satisfiable in an at-most-binary model.
16A Preliminaries
We take the language and the basic semantics of Laiexcept for dropping the
requirement of no busy choice sequencesfrom 15A. In particular, we use the
'This chapter draws on Xu 1995b, with the permission of the Association for Symbolic
Logic. All rights reserved. This reproduction is by special permission for this publication
only. (An error in the proof of Xu 1995b has here been corrected.)
415
416 Proofs and models
abbreviation Aa <=dy ,4&~[a stit: A], and take models 9Jt to be pairs {,3}
based on BT + I + AC structures & and interpretations 3. We take over the
recursive definition of truth from 8F, the semantics for the achievement stit
from 8G.3, and the various derivative semantic concepts from 6.
Turning now to axiomatics, let Lai be the axiomatic system which takes, as
axioms, all substitution instances of truth-functional tautologies as well as the
following schemata:
Theorems (h), consistency and maximal consistent sets (MCSs) are defined
w.r.t. Lai as usual. We will use $, \?,and so on to range over MCSs. Deductive
consequences deductive closures are as defined in 15A. Setting KQ to be the
set of all valid sentences and identifying Lai with all its theorems, we show in
the following sections that K0 = Lai. To that end, it is convenient to list the
following theorems and rules in Lai.
lr
The only use of the rule RS in our proof of the completeness theorem is to find a counter
for each F-companion set for [a stit: A}. Actually the following rule serves the same purpose:
Note that A1-A6 and A8 listed in 15A are identical to our Axl-Ax6 and
Ax8 here, whereas we extend the conditional A7 in 15A to the biconditional
Ax7 here. Note also that each of T1-T19 and R1-R4 listed in 15A is either
identical to some of Thl-Th9, DR1 and DR2 listed here or derivable from
them (together with Axl-Ax8). Note, finally, that there is an axiom in 15A,
that is, A9, which is not included in our list of axioms, nor in our list of
theorems; but the only place in chapter 15 where we applied A9 is Lemma
15-36, which is not necessary for the proof we give in this chapter. Hence we
can use here the results established in 15C-15E, where we did not apply A9
or refref.
16B Soundness
It is easy to see that Axl-Ax5 are valid for all BT +1 + A C structures and that
modus ponens and RE are validity preserving. A proof of the validity of Ax6
and Ax8 can be found in 15B. In order to establish the soundness of Lai, it
is therefore sufficient to show that Ax7 is valid for all BT + I + AC structures
and that RS is validity preserving. (In this section, as in chapter 15, we let a
= 3(a) be the agent denoted by a.)
16-1 LEMMA. (Validity of Ax7) Ax7 is valid for all BT + I + AC structures.
PROOF. It has been shown in 15B that
is valid for all BT + I + AC structures. It is therefore
sufficient to suppose that 971, m 1= [a stit: ~[a stit: A}&Ba} for any model 971
with m in it and show that 971, m 1= [a stit: ~{a stit: A&B}&Ba}. Let w be
the witness to [a stit: ~[a stit: A]hBQ] at m. Setting M = {m': m' e i & m'
we know by Lemma 15-7 and Lemma 15-8 that
418 Proofs and models
It follows from Lemma 15-6 that 271, i >w ~[a siz'f: A&B]. Suppose for
reductio that 971, m\ 1= [a sz: A&B] for some mi e M with witness u>i. Let
. Since w < mi and MI < mi, we know by
no backward branching, Post. 3, that either w ^ w\ or w\ < w. If u; ^ wi,
we have by (1) and Lemma 15-6 that 971, mi 1= [a stit: A&Ba], and hence by
Lemma 15-9, 971, mi t= [a stit: A ] , contrary to (1) since TOI e M. So it must
be the case that w\ < w. Since 971, MI N [a sizi: A&B] and 971,
by (1) it follows from mi 6 MI and Lemma 15-7
that 971, MI 1= [a stai: 4&5 Q ], and hence by Lemma 15-9,
contrary to (1). We conclude from this reductio that
and hence, since
It follows from (1) and Lemma 15-6 that
16-2 LEMMA. (Same pos- and neg-companion roots of stit sentences having the
same witness) Let 971 be any BT + I + AC model with mi, m.2 E i. Suppose
that 271, mi t= [a stit: A] and 271, m 2 N [a sirt; S] with the same witness. Then
for any C, 97T,
PROOF. Suppose that 971, mi t= [a stit: A&Ca] with witness w. Then, setting
we know that 97t, M 1= Ca, and hence by Lemma
15-7, Applying Lemma 15-9 we know that 971, mi 1= [a stit: A]
with witness w, and hence by hypothesis, 971, mg N [a siif: 5] with the same
witness. Since {TO: TO i & TO =j m 2 } C z|>, it follows from Lemma 15-6
that 971, m-2 N [a stii: S&C 10 ]. The other direction is symmetrical.
By definition of 3' and the fact that mi M, 271', m^ (= [a stit: q] with witness
w and a counter TOI. Then
by (2) and Lemma 16-2. Hence by (3),
It follows that
sfa't- ^4] D V K J < m[ a s^; q&D"]. Hence RS is validity preserving for .
PROOF. By Lemma 16-1, Lemma 16-3, Lemma 15-7, Lemma 15-9, Corollary
15-10, and Theorem 15-11.
PROOF. Let ' be the F-companion set for [a stit: A] in any MCS $ including
. Then Lemma 15-29 (ii) implies ' C nll. Applying Lemma 16-6, we obtain
a II' which is a F-counter to ' relative to [a stit: A}. D
Then
Since I- ~[a siz'i: [a siz'i: B]&G] D ~[a siz'i: B] V ~[a 5^z'i: G] by Ax4 and
Ax3, we know that h ([a stit: B]&G) Q D [a stit: B]&Ga. Now by Ax5, Ax6,
and (i), we have H [a sizi: B]&G a &[a stti: [a sizi: J5J&G] D [a sfei: 5&G a ]
&~[a stit: B&G"]. It follows from classical logic that I- [a stit: B}&.Ga D ([a
stit: .B]&G) Q , and hence, (4) holds. Now it is easy to see by (ii) and Ax2-
Ax4 that
D^})\ D [a stit: A&[a stit: B}]. Then, setting E = [a stit: B]&G and F} = [a
stit: B}&,D} for each j with 1 ^ j ^ m, we obtain by (4) that h [a stit: q&.Ea
and hence by Lemma
16-5 (substituting E for G, Fj for >.,, and
It follows from (4) that
PROOF. Since [a stit: A], [a stit: B} e $, [a stit: A&[a stit: B}} 6 $ by Ax3
and Ax4. Let E' be the F-companion set for [a stit: A&[a stit: B}] in $. Since
and since Y! \- A and [a stit: .A] 6 $, we have by Lemma 15-22 (ii) that
16. On the basic one-agent achievement-stit theory 421
Let [a stit: B'] be the characteristic sentence of E' w.r.t. A&[a stit: B\. By
hypothesis and Ax2, we know that E' I- [a stit: B]. Thus if (6) holds, Ax2
implies that h [a stit: B'} D Ca, from which, together with DR2 and Lemma
15-21 (%), it follows that E' h [a stit: A&[a stit: B]&Ca], a contradiction. Hence
(7) must hold. Let q be any prepositional variable not occurring in B', and let
Ci,...,Ck and Z?i,..., Z?m be, respectively, all the F-pos-companion roots and
all the F-neg-companion roots of [a stit: A&[a stit: B}} w.r.t. E', and let II'
= DC([a stit: B*]), where B* = qk[a stit: B]&(& 1< z < fc Cf) &(&i< ^ m ~[a
stit: <?&[ stit: B]&D^]). Since E' h [a stit: B} and IT h [a stit: S] by Ax2,
In order to show that II' is a F-counter to S' relative to [a stit:
A], we only need to show, by (5), that II' Y- A. To that end, we first show that
Consider any D3 with 1 ^ j ^ TO. Applying Lemma 15-21 (i) and the fact
that Dj e F, we know that either I- [a stit: B] D [a stit: B&D"} or h [a stit:
B} D ~[a stit: B&D}. Suppose for reductio that h [a stit: B} D [a stit: ?&
Df]. Since E' h [a stit: B], it follows from Lemma 15-23 that E' \~ [a stit: B
&~[a stit: B&D}}, and hence by Ax2, E' \- ~[a stit: B&D}, and hence by
the supposition of our reductio, E' h ~[a stit: B], contrary to the consistency
of E'. From this reductio we conclude that
follows that (8) holds. Next, setting E = &i^ t <; kC", we show that
and C must be Cr for some i with 1 ^ i ^ fc. We thus obtain h [a stit: B&
E] D [a stit: J5&C Q ] by applying Th4 k 1 times, and hence I- [a stit: B] D ~[a
stit: B&E] by (10). It follows from Th8 and DR1 that (9) holds. Finally we
show that II' Y- A. Suppose for reductio that II h [a stit: A]. Then h [a stit: q
&[a stit: J B]&E&(&i s ;^ T n ~[a stit: <?&[a stit: B]&D\)] D [a stot: A], and
hence by Th8, h [a sizi: g&[a stii: B]&E Q &(& 1 < ^ m~[a siit: g&[a sizi: 5]
&>"])] 3 [a sM: ^]. It follows from (8), (9), and Lemma 16-8 that h ~[a
and hence
422 Proofs and models
by Th8 again, h ~[a stit: B'], contrary to the consistency of E'. We conclude
from this reductio that II' V- [a stit: A}. Applying (5), the same argument in
Lemma 16-6 (replacing S and II by E' and II') shows that II' is a F-counter to
E' relative to [a stit: A].
In the following and the rest of this chapter, we assume that F is finite, closed
under subsentences and equivalently closed under conjunction. Applying our
strategy used in 15D, at each step of our construction of the structure, each
history intersects the last instant. Since our construction here could be infinite,
the whole construction could leave us some history h that does not intersect the
last instant i in the structure. In such cases, we have to extend those infinite
chains to meet the condition of unique intersection, which will be accomplished
by applying the following notion of "parameters."
Let 9 C F, and let be the F-companion set for [a stit: A] in <E>. 0 is a
parameter for [a stit: A] in $ if 0 is the set of all pos-companion roots of [a stit:
A] w.r.t. $, 0 is a parameter for in $ if 0 is a parameter for [a stit: A] in
<&, and 0 is a parameter for E if 0 is a parameter for E in some MCS $.
By our discussion in 15D, we know that if E is a companion set in $ for both
[a stit: A] and [a stit: B], they must have the same F-pos-companion roots and
the same F-neg-companion roots w.r.t. E. Furthermore, if 0 is a parameter
for E in an MCS <E>, it is the parameter for . We will therefore speak of a
F-companion set E with parameter 0 to indicate that 0 is the parameter for
E. Applying the definition of F-alternatives (15E), we know that E and II are
F-alternatives iff if they have the same parameter.
(i) We know that E' h Ca. If ' C E, we would have E I- Ca, contrary to
(11). (ii) Let (a stit: A'] 6 <l> with parameter 0'. If E h [a stit: A'], we would
have by Lemma 15-24 that E h Ca since C &', contrary to (11).
(Hi) Consider any B e E n T. Then by Lemma 15-22 (n), either
by Lemma 15-22(i,), and then
Suppose that E h- [a stit: B}. We know that ~[a stif: ?&C a ] 6 $, for otherwise
by Lemma 15-24, and then E h Ca, contrary to (11). It
follows from Ax6 that ~[a stit: B&C] e $. Now let [a stit: A*} be any
characteristic sentence of E'. We know by Lemma 15-21 (i) that [a stit: A*
and hence [a sirt: yl*&C1] e $ by Th4. Since
by Ax4, and then
Because Y is equivalently closed under conjunction, there is a D T such that
h D <-> 5&C and [a stai: A*&D Q ] e $. Then D e 6', and hence E' h B. It
follows that E n T C E' n T. To see that E n T C ' n T, consider C. We know
that C e E' and ~[a stit: C] 6 $. If C 6 E, we would have S h C, and thus by
Lemma 15-22(ii), E h Ca, contrary to (11). It follows that C
PROOF. Let * be any MCS including E', and let A e F such tha
with parameter 0. We show that E h [a stit: A]. Let E* be the F-companion
set for [a stit: A] in *. Since 0 C ', E* OF C E'nF by Lemma 16-11 (Hi), and
hence S' h A. By Lemma 15-22(MJ, either E' h Aa or E' h [a stit: A}. Since
it then follows that E' (- [a stit: A], and hence by
Lemma 15-24, * C S'. Consequently, since E' C $, E* C $, and hence [a stit:
A] e $ with parameter 0. Because E is -maximal in $ w.r.t. F, we have by
definition that
16-13 LEMMA. (Finding suitable T-maximal companion sets) Let E and E'
be F-companion sets in $ with parameters 0 and ' respectively such that
Then there is a F-companion set E* in $ such that E U E' C S* and
maximal in $ w.r.t. F.
PROOF. Let [a stit: B] and [a stit: B'} be characteristic sentences of E and S'
respectively, let AI ,..., An (n ^ 0) be all the sentences in F such that
with parameter Q (I ^ k ^ n), and let E* be the F-companion set for
424 Proofs and models
[a stit: A] in $, where A = [a stit: AI]&. .. &[a stit: An]&,\a stit: B]&[a stit:
B'] ([a stit: A] G $ by Ax3 and Ax4). Clearly, U ' C *. We show that
* has the parameter 6. If C G 6, E h- Ca by Ax2, and then E* h C Q , and
hence by Lemma 15-22(i), [a stit: A&Ca] G $. Suppose that ( 7 ^ 6 . Then
C1 ^ 6', and then by Ax2 and Lemma 15-23, ~[a stit: Ai&cCa],... ,~[a sttt:
A n &C Q ] $, and ~[a stit: S&(7a],~[a stit: 5'&(7Q] G $. Now if [a stit:
A&Ca] G $, [a stit: AI&. .. &A l &5&5'&C m ] e $ by Ax5, and then by
Th7 and DR1, either [a stit: Ai&Ca] 6 $ or ... or [a stit: An&Ca] $ or
[a stit: S&C a ] G $ or [a stit: S'&C"*] e $, contrary to the consistency of $.
It follows that ~[a stit: A&C1"] e $. Hence 0 is the parameter for *, and
then S* is 0-maximal in $ w.r.t. F. D
C4 For each w;^' Tree i Tree and m,m' G i-rvee such that w < w' < m and
w/ < TO', g+(m,w) C <?i(i</) 052(^0 and g+(m,w) = g+(m',w); and for
each to e JVee - ijvee andTOe iiyee, w <m only if there is an m' G zjvee
such that w < m' and g-(m,w) = g+(m',w).
C5 For each to TVee - iTree, if 9i(^) C gz(w) (or ^(w) C 51(10)), there is
an TO G i^ee such that w < rn, g+(m,w) = g\(w] (or 32(w)), and <?i(w)
(or Q2(w}) is 0(^i(ti;))-maximal in /(TO) w.r.t. F.
C6 For each io,u/ G Tree iTree and m G i-rree such that w < w' < m,
g+(m,w) n F = g+(m,w') D F iff g+(m,w) = g+(m,w') iff there is an
m' iTree such that w' < m' and g+(m',w') = g+(m,w'), and both
g+(m,w) and ^+(TO,w') are d(g+(m,u)))-maximal in f(m') w.r.t. F.
16. On the basic one-agent achievement-stit theory 425
We will use p,p', and so on to range over sequences (Tree, iTree, /, 9i, 92, g+,
<?_} satisfying C1-C6. Intuitively, when (Tree, ^, ijvee} satisfies Cl, ijYee is the
"last instant" in Tree. Note that for each p = (Tree, iTree, f , 9i, 92, 9+, 9-},
the following is implied by C4, Lemma 16-10, and Lemma 16-11 (i):
PROOF. If there is no w e Tree with w < TO, we apply Lemma 16-15. Suppose
that w < TO for some u; Tree. Then by Cl, we have a maximal chain w\ <
... < ; < m in Tree. Consider w\. If g+(m,wi) \~ [a stit: A], we apply
Lemma 16-16. Suppose that g+(m,wi) Y- [a stit: A}. We then first consider
wn. If g+(m, wn) Y- [a stit: A], we apply Lemma 16-17. If g+(m,wn) \- [a stit:
A], then there must be a k such that 1 ^ k < n, g+(m,Wk) Y~ [a stit: A] and
g+(m,u>k+i) r- [a stit: A}. Then we apply Lemma 16-18.
in Fg4}. It is easy to verify that for each A, TA is finite, and is both closed under
subsentences and equivalently closed under conjunction (see Lemma 15-32).
16E Completeness
The "preliminary structures" p= (Tree, ^,ziy e e ,/,91,g%,g+,g-) constructed in
the last section provide all conditions we need for the completeness theorem
except the order preservation condition and the following:
Let p = (Tree, =%, iTree, f , 9i,92,9+,9-) satisfy C1-C8, and let h be any history
(maximal <-chain) in Tree, h is a counterexample to C9 in p if /iPUjVee = 0.
Let h be any counterexample to C9 in p. It is easy to see by Cl that h is endless,
that is, for each w h, there is a w* e h such that w < w*, and then by Cl,
there must be an TO 6 irree such that w < w* < TO. It follows from C4 that for
each TO', TO" G iTree, if w < w' < TO' and w < w" < TO" for some w',w" h,
g+(m',w) = g+(m",w). Thus for each counterexample h to C9 in p, we define
Sh to be the function on h such that for each w h, Sh(w] g+(m,w), where
TO is any moment in iTree such that w < w' < m for some w' h (and hence,
Sh(w) = g+(m,w) for every TO iTree such that w < w' < m for some w' h);
and let S be the function on all counterexamples to C9 in p such that for each
such counterexample h, S(h) = {sh(w): w 6 h}. It is easy to see by C4 that for
each counterexample h to C9 in p, S(h} is a chain of F-companion sets.
PROOF. We first observe that by (12), C4 and Lemma 16-10, both S(h) and
{6(sh(w)): w G h} are C-chains. Because F is finite, there is a w* G h such that
9(sh(w*)} = 6(s(w')) for every w1 h with w* < w'. For the same reason, there
must be a w G h such that w* < w and Sh(w) flF = s/ l (w') for every w' h with
w < w', and then by C6 and our definition of Sh, Sh(w) is 0(s/ l (iu*))-maximal
in some MCS w.r.t. F and Sh(w) = SH(W') for every w' h with w < w'. D
and
g_ 0 == g_ U {{(m*,ti>),(?(u>)): to G /i}, where for each w h, g(w) = gi(w)
if Sh(w) = g2.(w], and g(to) = gz(w) if s^(w ; ) = 9i(w}- Then we construct
Po,pi,... the same way as we did in Lemma 16-20, except this time we require
that the following hold for every n ^ 0:
It is easy to see that all TO and w in po satisfy (13). Note that by hypothesis,
for each m G iTreen, w G Treen - iTreen and A G F, if TO and A constitute a
counterexample to C7 in p n , TO ^ Tree; and if w and ^4 constitute a counterex-
ample to C8 in pn, w Tree. We show as follows that at each step n, how to kill
any counterexample to C7 or C8. It is easy to see that if w and A constitute a
counterexample to C8 in pn, we apply Lemma 16-14 and kill the counterexample
by extending pn. So let TO G irreen and A G F constitute a counterexample to
430 Proofs and models
By this lemma, each p = (Tree, <, i-rree, f,9i,92,9+,9-) satisfying C1-C9 can
be converted to an at-most-binary structure (Tree, ^,iTree, Agent, Choice), and
we are very close to getting a BT + I + AC structure. The last thing we need
to do is to make all histories isomorphic. We will sketch how that can be done.
But first let us observe the following fact: When given a consistent sentence
A, we apply Lemma 16-20 to construct a p = (Tree,^,iTree,f,9i,92,9+,9-)
satisfying C1-C8, A f(m) for some m e i-n-ee, and it is easy to see that
Tree is countable, and hence so is each history in it. In extending p to a
p' = (Tree1, ^',iTree',/',9i,92,9+,9-} satisfying C1-C9, applying Lemma 16-
22, we extend each history in (Tree, <} only finitely many times (since there are
only finitely many subsets of F); and at each time, even though we may have
added uncountably many histories, each of these histories is obtained by adding
countably many new moments to a countable chain (a counterexample to C9).
It is then clear that all histories in (Tree1, <'} are still countable.
Let p = (Tree, ^,iTree, f,9i,92,9+,9-) satisfy all C1-C9 and be converted to
an at-most-binary structure (Tree, ^, iTree, Agent, Choice) (where Agent con-
tains a single agent denoted by a), and let each history in {Tree, ^} be countable.
For each history h in (Tree, ^), we can insert (duplicated) real numbers into h
and make the extended history isomorphic to, say, the interval [0,1], and let a
have a vacuous choice at each newly added moment.2 In such a process, one
needs to take care of the common portion of every two histories, which can be
done, for example, by fixing a well ordering of all histories in (Tree, <), say,
/i0, hi,..., / i , . . . ( < K, where K is the cardinal of the set of all histories), and
extending them in the following way. We first extend ho to h'Q such that h'0 is
isomorphic to [0,1]. For each ordinal , assuming that every h^ with < has
been extended to fy, let wnere
h'^ ^ is the initial segment of h1^ up to and including x, and extend h^ P to
an F such that F is isomorphic to (0,1] or [0,1] depending on whether P has
its last element, and finally let
2
It is easy to see from our construction that there must be a ^-smallest moment in Tree, for
each time we extend a structure {Tree!,^',iTree'>/'i9i)92>S+>5-) by adding a new moment
to the past of the earliest one in Tree, it is because we have some [a stit: A] with A G F such
that g^w) \- [a stit: A] and g'2(w) h [a stit: A] for the ^'-smallest member w e Tree'; and
clearly, there can be only finitely many such stit sentences.
16. On the basic one-agent achievement-stit theory 433
When all histories become isomorphic to [0,1], we can easily define (Tree1,
<', Instant, Agent, Choice1): (Tree1,^') is the extended tree-like structure just
described (i.e., Tree! is the union of all extended histories) with all histories iso-
morphic to each other. Instant is a set of instants in (Tree!', ^'), with iTree to be
the "last one" in it, satisfying unique intersection and order preservation. Agent
is the same as before. Choice! is the choice function such that for each w Tree
i-rree (i-G., each "old" moment not in iTree), Choice'(a,w) =
where and for each
w e (Tree' - Tree)Uirree (i-e., each "new" moment and each ("old") moment in
iTree), Choice'(a, w) = {H(w)}, where H(w) is the set of all histories in (Tree1, <'}
passing through w.
Because a has vacuous choice at every newly added moment, and because vac-
uous choice will not affect the truth values of sentences when we evaluate them at
moments in iTree, we will, for convenience, use Lemma 16-24 in a way as if we
get an at-most-binary BT + I + AC structure (Tree, ^, Instant, Agent, Choice)
(instead of (Tree, ^,iTree, Agent, Choice)) with iTree to be the last instant in
Instant. This being the case, we will keep using our notations m^ and hm as
before. The completeness theorem is included in the following theorem.
Let SOT = (&,3) be a model on & such that for every propositional variable
p FA, every h in Tree, and every m i with m h (i.e., m = m^),
(m,h) G 3(p) iff p 6 /(m). Applying ordinary induction, one can show that for
every A e F^0 and every m i, 9JT,m t= ^4 iff A 6 /(m), and consequently,
since ylo /(mo), 9Jt, mo 1= AQ. We omit the induction steps for truth-functional
operators, but provide the step for [a stit: ] as follows.
Let m i and [a stit: A] 6 F^0. Suppose first that [a stit: A] f ( m ) .
Then there is, by C7, & w Tree i such that w < m, g+ (m, w) h [a stit:
A] and g-(m,w) V- A. (15) implies that [a stit: A] /(m') for every m' i
with m' = m, and, since g-(m,w) A, there is by C8 an m" e z >^ with
A f(m"). It follows from induction hypothesis that 971, m' 1= A for all m' i
with m' =^ m, and 9JI, i >w A. Hence 2Jt, m t= [a stit: A}. Suppose next that
there is a w < m such that 9Jl,m' N A for every m' i with m' =j m, and
By induction hypothesis, A 6 /(m-') for every m' G i with
with
434 Proofs and models
m, and A < /(TO") for some m" & i\>w. It follows from C3, C8, and (15) that
g+(m,w) h A and g-(m,w) V- A. Now if g+(m,w) \- ~[a stit: A], we would
have g-(m,w) \- A by C2 and Lemma 15-28(Hi), a contradiction. It follows that
and hence by Lemma 15-22(ii), g+(m,w) h [a stit: A}.
Since by
The following remarks are worth adding to the main completeness result,
which has been established with Theorem 16-25.
REMARK. (At-most-binary BT +1+AC structures) Because the BT + I+AC
structure mentioned in Theorem 16-25 is at most binary, Theorem 16-25 gives
us more than the completeness of Lai. Let Lali be the set of all sentences
valid for all at-most-binary BT + I + AC structures. Identifying Lai as the set
of all its theorems, we have as a consequence of Theorem 16-25 and Theorem
16-4 that Lai = Lali.
REMARK. (About the finite model property) Lai does not have the finite model
property. A simple example is refref, Ax. Cone. 1. Chapter 15 showed that
refref is valid for a BT + I+AC structure iff the structure contains no busy
choice sequence. Thus if Lai had the finite model property, refref would be a
theorem of Lai; but, in fact, it is not.
17
Decidability of many-agent
deliberative-stit theories
In this chapter we give an axiomatization, Ldm, for the basic dstit logic, and
prove its completeness and decidability by way of the finite model property. It
is surely very natural to combine dstit theory with indeterminist tense logics,
especially when we consider deliberative seeing to something to be connected
with what the future will be like. In carrying out some basic technical work in
dstit theory, however, we will use a formal language without tense operators,
though we will use the historical necessity operator Sett:, 8F.4, as a primitive.
In our formal language, we will introduce Chellas's cstit operator, 8G.2,
as an abbreviation, just as in 11 A. The reader familiar with modal logic can
easily see that the Chellas 1992 theory of cstit is decidable, since Chellas did not
propose any condition concerning the relation among different agents. Our proof
actually shows that cstit theory is decidable even if one adds the independence
of agents condition, Post. 9.
How many possible choices does an agent have at a given moment? For a
logic of agency, this is an interesting question. We will show some results about
the expressibility of our formal language in this aspect, that is, for each number
n ^ 1, we have a scheme that corresponds to the condition that every agent
has at most n possible choices at every moment. We will also give, for each
n ^ 1, an axiomatization for the dstit logic with this condition, and prove its
decidability.
17A Preliminaries
The language for Ldm contains as primitive symbols, propositional variables po,
Pi, P2, ..., agent terms QO> c*ij a2, -, &n equation symbol =, truth-functional
operators ~ and &, the historical necessity operator Sett:, and the dstit operator
[ dstit: ]. Sentences are constructed in the usual way, except that a = /3
is a sentence whenever a and /3 are agent terms, and [a dstit: A] is a sentence
435
436 Proofs and models
Recall that the choice equivalence relation, Def. 12, is an equivalence relation,
and hence by the recursion clauses of the truth definition, the operators Sett:,
and [a cstit: ], [/3 cstit: ], and so on are just like the necessity operator in
the modal logic S5, as noted in 11A. As in Def. 20, we define the validity of
A for a model 9JT, 9JT 1= A, as 9JI, m/h 1= A for every m and h in 971 with m e
h, and the validity of A for a structure &, 6 \= A, as 9711= A for every model
97T on 6.
Our logic Ldm takes as axioms all substitution instances of truth-functional
tautologies as well as the following schemata, where a, (3, 7, and /?o, ..., (3k are
any agent terms:
lr
That is to say, for every agent term a, both 3(a) and a denote the agent in Agent that
3 assigns to a. In each situation, we will use whichever is convenient.
17. Decidability of many-agent deliberative-stit theories 437
APC can be called an axiom scheme for possible choices. In the next section
we show that for every BT + AC structure 6 and for every n ^ 1, (3 1= APCn
iff every agent (in 6) has at most n possible choices at every moment (in &).
Other syntactic notions such as theorems in Ldmn (l~Ldin n )) Ldmn-consistency
(consistency w.r.t. Ldm n ), maximal consistent sets w.r.t. Ldmra (Ldm n -MCS),
and so on are defined as usual. It is easy to see that for each i and k with 1
^ i < k, hLdm, APCfc, and hence, identifying Ldmn with all its theorems,
we have Ldm C ... C Ldm3 C Ldni2 C Ldmi. In fact, as the soundness
theorem shows, we have Ldm C ... C Ldm3 C Ldm2 C Ldmi. It is easy to
see by Definition 17-1 that the following are derivable in Ldm (and hence in all
Ldmn):
2
When taking Sett: and cstit as primitive, one needs to introduce
A}&,~Sett:A as an abbreviation, and replace A3 by Sett:A D [et cstit: A}. When taking dstit
and cstit as primitive, one needs to introduce Sett:A fdj [a cstit: A]&~[a dstit: A] for some
particular a, and replace A3 by
438 Proofs and models
Note that although the decidability of Ldm will also give us the decidability of
the dstit theory with a single agent, it would be nice to have an axiomatization
of that theory. It is easy to check, along with our proof, that Al, A2, A3,
and RN will be sufficient for that axiomatization. Note also that A2 does not
directly show the a priori plausibility of the dstit operator. It is easy to see,
however, that the following are derivable in Ldm:
3
Recall from, e.g., 2B.8 that the refref equivalence does not in general hold for astit.
T13 indicates that the stit analysis of refraining coincides with von Wright's analysis of
refrainingnot doing conjoined with the ability of doing (see von Wright 1963). As noted in
Horty 2001, the left-hand side of T13 is only an approximation of von Wright's analysis of
refraining.
17. Decidability of many-agent deliberative-stit theories 439
T14 and T15 give us that doing is equivalent to refraining from refraining.
That is to say, the following refref equivalence holds for dstit:
(refref
17B Soundness
In this section we prove the validity of AIA^ for all k ^ 1 and some results
about the expressibility of the language concerning possible choices, and finally
we establish the soundness theorem for Ldm n with all n ^ 0.
17-4 LEMMA. (Validity of AIA^) For each k ^ 1, AIA^ is valid for every
BT + AC structure.
PROOF. Let 6 be any BT + AC structure, and let 2JI = (&, 3) be any model
on 6, and let k ^ 1. Suppose that 9JI, m/h \= diff(/30, ..., /?fc)&Poss:[/? 0 cstit:
B0] & ... & Poss:[/3k cstit: Bk\ for some m and h in 9Jt with m 6 ft. Then by
truth definition and Definition 17-2, all 3(/?o)> , 3(/?fc) are different agents in
Agent, and there are fto, ..., ftjt such that for each i with 0 ^ i ^ k, OT, m//i
^ [/?, cstit: Bt}. Hence by Fact 17-3,
Although all AIA& with k ^ 1 are valid under the independence of agents
condition, the set {AIA^: k ^ 1} does not fully express this condition. This is
so simply because in a BT + AC structure = (Tree, ^, Agent, Choice}, it could
be true that \Agent\ ^ Ho. 4 Fortunately, the presence of all AIAfc with k ^ 1
is sufficient for our proof of the completeness theorem.
4
Consider the following example: Let Tree = {m, mo, mi, ...}, ^ = {{m', m'}: m' 6
Tree}U{{m, m^}: k ^ 0}, and Agent = {ao> fli, } For each /c ^ 0, we set /i^ = {m,
mjt}. It is easy to see that #( m ) = {ho, h\, ...}. Let us define Choice to be a function on
Agent* Tree such that for each k ^ 0, C/ioicem = {Hk,i, #fc,2} where /f^i = {ho, -, h^},
Hk,2 = {/ifc + l. hk + 2, } Since f l k s o Hk,2 = 0, we know that Choice does not satisfy
the independence of agents postulate. If, however, we treat & = (Tree, ^, Agent, Choice) as
a BT + AC structure, and if we keep all other semantic notions the same, then we will still
have & N AIA& for every k ^ 0. Examples of this kind suggest that there are no sentences
corresponding to the independence of agents condition since we can only talk about finitely
many agents in each sentence. What {AIA^: k ^ 1} corresponds to is the weaker condition
that for each m e Tree, and for each function choicem on any finite subset Agentf of Agent
such that
440 Proofs and models
PROOF. Suppose that \Choiceam\ ^ n for all a Agent and all m 6 IVee. Let
37T be any model on 6. If n = 1, then it is easy to see that 97T 1= Poss:[a cstit:
A] D A. Assume that n > 1 and set a to be any agent term, and A\, ..., An
any sentences. We show as follows that 971 N APC n . Suppose for reductio that
271, m/h APC n for some moment m and some history h with m h. Then
971, m//i t= POSS:[OJ cstz't: AI], 371, m//i 1= Poss:(~^i&[a cstit: A 2 ]), ..., 37T,
m//i t= Poss:(~4i & ... & ~,4 n _i&[a cstz't: A n ]), and 971, m//i 1= ~^i & ...
& ~>l n . Then there are hi, ..., hn such that for each i with 1 ^ i ^ n, m /i,
and
For each i with 1 < z < ra, let us set e% = {h': h' =^ ht}. Since 371, m/h t=
~>li & ... & ~>4 n , we know by (2) and Fact 17-3 that for each i with 1 ^ i ^
n, /i _L^ /i,. That is to say, h e, for all i with 1 ^ i ^ n. Since Choice3^
is a partition of -ff( m ), there must be an e Choice^"' such that h & e. But
obviously e ^ e, for all z with 1 ^ z ^ n. It follows that |(7/iozce^a'| ^ n +
1, contrary to our supposition that \Choice3Jf^ < n. We conclude from this
reductio that 371 N APC n . Next suppose that [Choice^ ^ n + 1 for some a
j4<?en and some m Tree. We show as follows that there is a model 37T on
such that for some h with m h,
Since it is easy to verify that Al A6 are valid for all BT + AC structures and
RN is validity preserving, Lemma 17-4 and Lemma 17-5 give us the following.
17-7 LEMMA. (Properties of the canonical frame for L w.r.t. X) Let L = Ldmra
for any n ^ 0, and let (X, [Po], [Pi], , R[p0], R[0\]i ) be the canonical frame
for L w.r.t. X. Then the following hold:
i. For each w X and each A, Sett:A 6 w iff A w' for all w' X iff
Sett: A e w' for all w' X;
ii. for each w G X and each a and A, [a cstit: A] G w iff A w' for all w'
E X with w'R[a]W iff [a csizt- ^4] w' for all w' X with /.R[Q]w;
iii. for each w & X and each a and A, [a dstit: A] e w iff ^4 6 w' for all w'
X with w'R[a]W, and ~^4 6 w" for some w" 6 X.
PROOF, (i) and fiij are trivial (apply modal logic and T2), and (Hi) can be
easily obtained by Tl, (i) and (ii). D
PROOF. Let / be any function on {[/30], [/?i], ...} with /([/?,]) E[0j for all i
^ 0. We show as follows that f ) t > o /([/?]) ^ 0- We know by modal logic and
Lemma 17-7(i, ii) that there are AQ, AI, A%, ... such that
and that for each i ^ 0, there are Blto, B^i, S t ,2, such that
Since [/?o], [/?i], [,$2], are different =-equivalence classes, we have that
PROOF. Let L = Ldmn for any n ^ 0, and let G be any L-consistent set of
sentences. Then there is a L-MCS w Wi, including Q. Suppose that X is the
^-equivalence class to which w belongs, and let (X, [/30], [/?i], , R[p0], R[0i]>
...} be the canonical frame for L w.r.t. X. We want to convert (X, [/30], [/?i],
..., R[p0], Rtfi], ) into a BT + AC structure 6 = (Tree, ^, Agent, Choice}.
Let us first set
Tree = { X } U X ;
Note that at each w, each agent [/3t] has a vacuous choice. It is easy to see
that for each i ^ 0, Choice % is a partition of H(x), and that the no choice
between undivided histories condition is vacuously satisfied. To see that Choice
satisfies independence of agents, let choicex be any function on Agent such that
choicex ([0z]) Choice fr each i ^ 0. Then by definition of Choice, for each
i ^ 0, there is an e% 6 ^[/g,] such that choice x ([(31}) = {hw: w et}. Let / be
the function on Agent such that for each i ^ 0, /([/?,]) = e, [/?,] Since there
is a one-one correspondence between all w X and all hw in (Tree, <}, it is easy
to see that for any w & X and any i ^ 0, w 6 /([/3Z]) iff hw choicex([/?i]), and
hence by Lemma 17-8, Hoo choicex([/3i}) ^ 0- It follows that Choice satisfies
independence of agents, and thus & is shown to be a BT + AC structure.
When L = Ldmn with n ^ 1, we need to show that each agent in has at
most n possible choices at every moment. Since \Choice^\ = I for every i ^
0 and every w X, we only need to show that \Choicex ^ n fr every i ^
0. But we know by definition of Choice that for each i ^ 0 and each k ^ 1,
It follows from Lemma 17-9 that
n.
444 Proofs and models
Let us define a model 9Jt = (&, 3) such that for each agent term a, 3(a) =
[a]; and for each propositional variable p and each history hw in &, (X, hw)
6 3(p) iff p e 10. It is easy to see by our definition that for each h in 6, h
#(x) iff 3w e X[/i = hw]; and that for each agent term a, and for each w, w'
X, hw = %' hwi iff 3e [Q][w, ;' e] iff wR^w'. Then one can show by
induction that 3JT, X/hw N ^4 iff A w for every ,4 and every w 6 X, applying
Lemma 17-7(i) and Lemma 17-7'(Hi).
Some general remarks are worth making. Let us define a dstit logic as a set
of sentences that contains all truth-functional tautologies, all Al A5 and all
AIAfc for k > 1, and is closed under substitution, modus ponens, and RN. Let
L be any dstit logic. A model for L is any BT + AC model 9Jt such that 271
t= A for all A L. L is complete for its models if for every A L, there is a
model for L such that 371 A. Going over Lemma 17-7, Lemma 17-8, and the
first part of Theorem 17-10, (considering L there as an arbitrary dstit logic),
one can easily see that the following holds.
17-12 REMARK. (Completeness for models) Every dstit logic is complete for
its models.
A general BT + AC structure is any triple 03 = (&, A, Z) such that (i) & =
(Tree, <, Agent, Choice) is a BT + AC structure; (ii) A = {QO, ai, ...}, where
ao, ai, ... are all the agent terms in our language and an Agent for every n
> 0; and (in) Z is a subset of the power set of {(m, h): m h} and is closed
under Boolean operations as well as the following, where z G Z and a A:
PROOF. Let L be any dstit logic, and let A L. We know by Remark 17-12 that
there is a model SDt = (6, 3) for L such that VJl F A. Let us set an = 3(a n )
for every n ^ 0, and A = {a0, a{, ...}; and set \\B\\m = {(m, h): 971, m/h 1=
5} for every sentence 5, and Z = {||B|| : 5 is a sentence}. It is easy to show
by induction that 05 = (S, A, Z) is a general BT + AC structure. We show as
follows that 03 is a general BT + AC structure for L. Suppose for reductio that
B e L but m* B for some B and some model 9JT = (6, 3*} on 03. Let B =
B(p0, ..., pk) (all prepositional variables occurring in B are among p0, ..., pk).
By definition 3*(p 0 ) = I I C o H , > ?*(Pk) = {{Ckf* Consider the sentence B'
= B(C0/po, ..., Ck/Pk), that is, the sentence obtained from B by substituting
Ct for pt for each i ^ k. We claim that 97t B'. To justify this claim, it is
sufficient to observe that for every sentence D(po, ..., p k ) , 9JT, m/h t= D(p0,
..., pk) iff 9JI, m//i N D(Co/po, ..., Ck/Pk), which can be easily established by
induction. Since B e L, and since L is closed under substitution, we know that
B' L, and hence, since OT is a model for L, OK t= B', which is a contradiction.
From this reductio we conclude that B 6 L implies SOT* t= B for every sentence
B and every model 971* on 03, that is, 03 is a general BT + AC structure for L.
It is easy to see that 9JI itself is a model on 03, and hence, since 971 A, 03
A. 0
5
Note that the decidability of dstit theories is not a direct consequence of Gurevich and
Shelah 1985a and Gurevich and Shelah 1985b, since in their result, the theory on tree-like
frames does not contain a quantifier over sets of choice-equivalent histories, which is needed
because of the presence of Choice in BT + AC structures, together with the truth conditions
for dstit.
446 Proofs and models
17-14 LEMMA. (A property of AA) Let L = Ldmn for any n ^ 0, and let A be
any sentence. Then for each j3 occurring in A and each sentence [(3 cstit: B]
A A, there is a sentence B' such that KL [cstit: B'} = [cstit: ~[cstit: B}],
and [ft cstit: B'} e AA.
PROOF. It is easy to see that if [(3 cstit: B] AA, then either [ cstit: B] S0
or [cstit: B] = [cstit: ~[cstit: C]] with [ cstit: C] e E0. Suppose that
[cstit: B] e AA. If [cstit: B} e E0, then, setting B' = ~[/S cstit: B] 6 AA,
we know that [cstit: B'] Si- If [cstit: B] e i-E 0 , then [cstit: B} =
[cstit: ~[/S cstit: C]] for some [cstit: C1] SQ, and hence [cstit: ~[cstit:
5]] = [cstit: ~[cstit: ~[/S cstit: (?]]]. Setting B' = C and applying modal
logic, we have [/S cstit: B1} EI. D
Let L = Ldmn with n ^ 0, let .AT be any ^-equivalence class, and let $ =
(X, [/So], [/Si], ...} be the agent frame for L on ^. Consider any sentence A.
We first define a relation on X as follows: For each w, w' X, w ~ w' if
for every 5 6 11^, B & w iff B w'. w is obviously an equivalence relation.
Since UA is finite, there are only finitely many ^-equivalence classes. Selecting
a representative for each of these w-equivalence classes, we set Y to be the set
of all those representatives. Next, since there are only finitely many agent terms
occurring in A, there must be finitely many =-equivalence classes, say [/So], - ,
[f3k], such that each occurring in A belongs to [/S,] for some i with 0 ^ i ^ k,
and each [/St] with 0 ^ i ^ k contains at least some J5 occurring in A. We may
assume, without loss of generality, that all /So, ..., j3k occur in A.
A filtration of $ through AA and II^ is a sequence (Y, [/So], ..., [/3/t], [/3 0 ],
..., ~[0 fc ]), where F is the set of representatives for ^-equivalence classes, and
for each i with 0 ^ i ^ k, \p,] is the relation on Y such that
and 11^" as if there is only one such filtration. Consider each i with 0 ^ i ^
k. Since Y is finite and [/?,] is an equivalence relation on Y, there are finitely
many [/?,]-equivalence classes. We will use t/[/3,], for each i with 0 ^ i ^ k, to
denote the set of all [/?,]-equivalence classes, and use u, u', and so on to range
over its elements.
17-15 LEMMA. (Properties of filtmtions) Let L = Ldmn for any n > 0, let X
be any ^-equivalence class, and,let $ = (X, [/So], [/Si], } be the agent frame
for L on X. Suppose that A is any sentence and (Y, [/So], - , [/Sfc], [/90], ,
[/3fc]) is the filtration of ^ through A. A and H^. Then for any w y,
i. if Sett:B then Sett:B iff B for all
PROOF. (%) Let 5ett:5 e UA. Suppose that Sett:B w. Then B w' for all
w' e .AT, and hence, B w' for all w' e Y. Suppose that Sett:B w. Then
there is a, w" X such that B w". Since B 11^, and since there is a w'
Y such that w' w", it follows from our definition of fa that 5 ^ /.
(ii) Let [csttt- 5] e A^. Suppose that [cstit: B] w. Let 7 be the
representative of [/?]. Consider any u/ e F with w' \p\ w. Then w' ~[7] w and
[7 csiit: B] w by A5. Since [7 csttf: B] A.A, it follows from the definition
of [7] that [7 cstit: B] G w', and hence by A2, B G w'. It follows that B
w' for all w' Y with w' ~[^j w. Suppose that [csiii: 5] ^ w. Then by A5,
[7 cstit: B] w. Since w Y C X, it follows from Lemma 17-7 (ii) that there
is a, w" X such that B w" and
Since there is a w1 Y such that w' w", it follows from (8) and the definition
of ~[ 7 j that w ~[ 7 j w' (i.e., w c^j w') and B ^ w'.
(Hi) Let [yS dsiii: 5] 6 ZA. By Tl we know that [(3 dstit: B} w iff [cstit:
B}&~Sett:B w. Since, by definition of UA, [cstit: B] AA and ~Sett:B
A A HA, it follows from (i) and fz'z^ that [d5<zi: 5] 6 w iff B w' for all w'
Y with w' ~[0] w, and 5 ^ /' for some w" Y.
(iv) Consider any MQ U[@0], ..., u/t U[pk]. Consider any i with 0 ^ i ^
k. It is easy to show by definition of [/3,j that there are [/3t csz: C"o]i , [/?i
csiii: C,], [8t cstit: D0], ..., [/3, cstit: DA A.A such that
448 Proofs and models
By (9) we know that j + l + 2 ^ |A^|. Since [/? cstit: >0], ..., [, cstit: >,]
Ayi, we know by Lemma 17-14 that there are o, - , EI such that
Selecting some wt e w,, [, cstit: Co] & ... & [/Si cstit: CJ&f/S, cstit: E0] &
... & [0i cstit: EI] e wt by (10) and (12). Now let us set B, = Co & ... & C",
&Eo & ... & -E1. Then by modal logic, [, 0cstit: jB t ] 6 u;,. In general, we can
define BQ, ..., Bk the same way. Selecting w0 MO, ..., Wk E ujt, we have
It follows from modal logic that Poss:[f3o cstit: BQ] & ... & Poss:[k cstit: Bk]
to for some (actually, all) w e Y. Since [0], ..., [Pk\ are different =-equivalence
classes, it follows from the definition of = that diff(0, ..., k)&Poss:[(30 cstit:
B0] & ... & Poss:[j3k cstit: Bk] w, and hence by AIAfc,
Recall that for each i with 0 ^ i < k, Bt = C0 & ... & Cj&E0 & ... & ;
with [t cstit: C"0], ..., [/?t cstit: Cj], [t cstit: 0], ..., [/? cstit: jBj] e AA and
j + i + 2 < |AA Hence Poss:([/?0 cstit: 50] & ... & [/?fc cstit: B fc ]) 6 HA by the
definition of IIA. It follows from (14) and (i) that
and hence by A2, [, cstit: CQ] &...&[cstit: C f J ]&~[ i cstit: D0] & ... &
It follows from (9)-(11) that w' w t. Thus we can show
in general that w' MO, , w' M& the same way. It follows that
PROOF. Suppose for reductio that |^[/3,]| ^ n+1 for some i with 0 ^ z ^ fc.
Let a = @t, let UQ, ..., un G C/[Q] such that UQ, ..., w n are all different, and let
Wo UQ, ..., wn 6 w n . It follows from Lemma 17-15(ii) that there are [a cstit:
AI], ..., [a cstit: An] A^ such that for each i with 1 < i < n, [a cstii: .Aj]
Wi and ^4j ^ lu^ for all j with 0 ^ j ^ n and i ^ j. That is to say, ~(^4i V
...VAn) e w0, [a cstit: AI] e wi, ~^i&[a cstit: A2] & io2, ..., ~^4i & ... &
It follows from Lemma 17-7(i) that
PROOF. Let L = Ldmra for any n ^ 0, and let A be any L-consistent sentence.
Set X to be any ./^-equivalence class such that A w for some w G X, and J
= (X, [/?o], [/3i], ...) to be the agent frame for L on X, and (F, [/30], , [/?n],
^[/30], - . - , [/3 t ]} to be the filtration of $ through A^ and 11^. We first construct
a structure 6 = (Tree, ^, Agent, Choice} as follows. Let us set
Tree = {Y}UY,
Since Y is finite, Tree is clearly finite. Similar to our proof of Theorem 17-10,
for each w Tree, we set hw to be the unique history in (Tree, ^) to which
w belongs, that is, hw = {Y, w}. Note that there is a one-one correspondence
between all w Y and all hw in (Tree, ^}, that is, for every h in (Tree, <},
Finally we set
The refref equivalence for the achievement stit, Ax. Cone. 1, was discussed in
2B.6, 2B.8 and chapter 15.* We repeat it here for convenience:
(refref)
With [a stit: A] and [a stit: ~[a stit: A}} representing doing and refraining from
doing respectively, this equivalence asserts that doing is equivalent to refraining
from refraining from doing. This assertion holds if the BT + I + AC structure
contains no "busy choice sequences" (sequences of infinitely many nonvacuous
choices for an agent within a finite time). That is to say,
ques
in every BT + I + AC Structure containing no busy choice se-
quences, doing implies, and is implied by, refraining from re-
fraining from doing
Figure 2.12 suggests a partial justification of (1), and we provided a proof in
15C. We now raise the converse question. Let BT +1 + AC + be be the set of
all B T +1 + A C structures containing at least one busy choice sequence. We ask
whether or not the refref equivalence fails in every BT + I + AC+bc structure.
We know that the refref equivalence fails in some particular BT + I+AC+bc
structures (see Figure 2.13). Here we prove that
in every BT + I + AC + bc structure, doing neither implies nor is
implied by refraining from refraining from doing.
We can then establish the following as a consequence of (1) and (2).
The refref equivalence holds in a BT + I + AC structure iff the structure
contains no busy choice sequences.
Our first section contains some preliminaries, and clarifies the concept of a busy
choice sequence. The second section presents the proof of (2).
*With the permission of Kluwer Academic Publishers, this chapter draws on Xu 1994c.
451
452 Proofs and models
ISA Preliminaries
In this chapter we deal only with the achievement-stit, 8G.3, and BT + I + AC
structures, 2. Semantic clauses for prepositional variables, ~, D, and the
achievement stit are as listed in 8, while other semantic concepts such as an
interpretation 3 and a model 97t on a BT + 1 + AC structure are listed in 6. A
past is defined, as in Def. 6, as a nonempty subset p of Tree such that 3m[m
Tree & Vu>(u) S p > w ^ m)] and VwVw'[w 6 p & w/ ^ w / p]
(intuitively, any cftazn of moments in Tree that closed in both past and future).
We compare moments and instants as in Def. 9 by defining w < i iff w < m
for some m e i, p < m iff Vw[w p > w < m], and p < i iff p < m for some
m i. As in Def. 9, we set i\>w = {m: m i & w < m} (tfte horizon from
w at i), and we also set i\>p = {m: m & i & p < m} (the horizon from p at
i). It should be noted that p is used for pasts (as in Def. 6) and q is used for
prepositional variables.
We use m =, m' (m and m' are choice equivalent for a at w), and m
m' (m and m' are choice separated for a at w), as listed in Def. 12. Note that
both m =, m' and m J_, m' imply w < m and w < m'.
The following lemma is useful in later discussions.
PROOF. By the axiom of choice there are two histories h and h' such that {w',
and {w', w, m'} C ft'. Since by definition, which
implies h' =,/ ft for every agent a by no choice between undivided histories
Post. 8, and hence m' =,, m for every agent a. In particular, if m' =, m,
then w < m and w < m'. Hence w' ^ w and m' =, m implies
It is the fact that Lemma 18-2 implies the uniqueness of w in the displayed
right-hand clause that entitles us to call call w the witness to [a stit: A] at m;
see Fact 15-2. Similarly we will call any m" satisfying the listed conditions a
18. Doing and refraining from refraining 453
counter to [a stit: A] at m. This being the case, we will allow ourselves to speak
of "971. m 1= [a stit: A] with witness w and counter m"."
Semantic definitions of the various uses of "!=", for example 9H, M\= A (settled
truth of A throughout the set of moments M), are listed in 6. As special cases
of settled truth throughout a set, here we use 971, i 1= A, 971, i\>w N A and 971,
i\>p 1= A, and so on. The following easily verified facts prove useful.
ii. A implies
Recall that although for each h, 9Jt, m/h t= [a stit: A] iff 971, m N [a stit: A]
(that is, settled truth and truth relative to a history are the same w.r.t. stit
sentences), 971, m/h 1= A is in general different from 971, m t= A (that is, settled
truth is not the same as truth relative to a history w.r.t. all sentences).
As listed in Def. 14, a chain c is a busy choice sequence for a iff (i) c is upper-
and lower-bounded in Tree, and (M) c is an infinite sequence of (nontrivial)
choice points for a. What turns out to be essential for the present chapter
is this: For any agent a, a busy a-choice sequence in a BT + I + AC structure
& (or in a BT + I + AC model 971) is an upper- and lower-bounded chain of
moments that does not terminate in the forward direction, and at each of those
moments a has nonvacuous choice. That is to say, a busy a-choice sequence is
a nonempty chain BC of moments such that
We will fix on a single agent for our discussion, and therefore we will say "busy
choice sequence" instead of "busy a-choice sequence." We let BC, BC', and so
on range over busy choice sequences, and for any busy choice sequence BC, we
will use pBC to denote the smallest past including BC, that is, PBC {w: 3w/[u/
G BC & w < w'}}. We define BC < i iff A w [ w e BC - w < i}. Whenever we
speak of a model 371 = ( , 3 ) and a busy choice sequence BC in it, we assume
that 3(a) = a where BC is a busy a-choice sequence.
i. Suppose that p = pBC for some BC < i in 6. Then for all m, m' 6 i\>p,
all w p and all a 6 Agent, m => m'. (This is a consequence of Lemma
18-2.)
454 Proofs and models
m. Suppose that BC1 is any busy choice sequence. Then there is a countable
subset BC' = {WQ, Wi, W2, } of BC that is still a busy choice sequence.
18-5 LEMMA. (Sufficient condition for ~[a stit: A}) Let 97T be any BT + I + AC
model in which p < i, and let A be any sentence. Suppose that
and that for every w G p, there is a w' G p with w < w' and
Then
PROOF. Suppose for reductio that 371, m t= [a stit: A] for some m G i|> p with
witness w and counter m'. Then by no backward branching, we have either p
< w or w G p. But this is impossible. For if p < w, then m' G i|> p , contrary to
hypothesis; and if w G p, then by Lemma 18-2, for each w' p with w < w',
m =^, m' for all m" 6 i|>u/, and hence 971, i >w> t= 4, contrary to hypothesis.
It follows that
18-6 LEMMA. (Sufficient condition for ~[a stit: .A],) Let 971 be any BT + I + AC
model in which p < m* G z, and let
. Suppose that 971, M 1= A and that for eve
Then
PROOF. Suppose for reductio that 971, m N [a stit: .A] for some m e M with
witness w. Two cases.
CASE, w p. Since m i >p, there is a w' p such that w' ^ m. It follows
that w < w', and hence by the hypothesis of this lemma, there is an m' i
such that w' < m' and 97T, m' A. But Lemma 18-2 and w' < m* imply that
m' =2, m*, and so m' =j m since m M; hence 971, m' 1= ^4, a contradiction.
CASE, w ^ p. Consider any counter m" to [a stit: B] at m. We show that
m" M, which contradicts our hypothesis. First, since w < m" and w p, we
have the following by no backward branching:
Then by Lemma 18-2, for every w' p with w' < m", m" =^, m since w <
m and w < m", and hence TO" =^, TO* since TO =^, TO* (m 6 M). Next, if
m" G i|> P , we would have p < m", and hence by (3), p < w which implies m
G i|> p , contrary to our assumption that m G M. Hence, m" ^ i[ >p . It follows
that m" G M. From this reductio we conclude that 971, M \= ~[a stit: .A], which
completes the proof. D
18. Doing and refraining from refraining 455
PROOF. Let p = pBc and fix m" i\>p. We define a model 971 in such a way
that for all m and all /i in 6 with m h, (m, /z) ,3(g) iff m e M = {m: m
z - i >p & 3w[w e p & w < r a & T O _ L m*]}. Clearly, z = i| > p UMUM',
where M' = {m: m e z i|> p & Vw[w p & w < m > TO =, w*]}. By our
definition of 3 and Remark 18-3 (ii), we have
Let m' be any moment in i with TO' =2, TO. Since m _L^ TO*, m' _Lj m*. We
know by m ^ z| >p that there is a w' e p with w' m, and hence w < w'.
Since m' _L^, TO*, we know by Lemma 18-2 that w' m', and hence m' i >p.
Hence m' G M, and it follows that (5) holds. Hence we have by 97t, M t= q that
971, m' N 9 for all m' e i with m' =2, TO. Since w e p, 971, ?!>, ^ g by (4) and
Remark 18-1, and hence 971, m 1= [a stit: q] by Remark 18-3 (^. It then follows
from Remark 18-3 (ii) that
18-8 THEOREM. (Doing does not imply refraining from refraining from doing)
Let & be any BT + I + AC structure in which there is a busy choice sequence.
Then
PROOF. By Lemma 18-7, there is a model 971 on & such that for some i, 971, i
and 971, i 1= ~[Q stit: ~[a stit: q}}. and hence 971,
for some m e i and 97?, i t= ~[a stit: ~[a stit: ~[a stit: q}}}. It follows that 6
456 Proofs and models
Note that if S' is a substructure of 6, (Tree1, ^') must satisfy the historical
connection condition Post. 4 and must be closed forward w.r.t. (Tree, ^}, and
that every pair {Tree', <') (Tree' C Tree and <' = ^n (Tree' x Tree')) satisfying
these two conditions determines a unique substructure. Note also that if 6' is
a substructure of 6, then for any a 6 Agent' and any w e Tree', m' = , m iff
Tree, where m' =', m means that in 6', m' and m
are choice equivalent for a at w.
A model 931' = (6', 3') is a submodel of a BT + I + AC model SJt =
3) w.r.L an instant i' in fH' if 6' is a substructure of 6 and for each agent
term a, 3'(a) = 3(a); and for each m 6 i', each A in 6 with m & h, and each
prepositional variable 9, {m, /i) 3(g) iff (m, /i'} e 3'(g), where /i' is the history
in 6' such that h1 = hn Tree'.
18-9 LEMMA. (Submodel lemma) Let OT' be a submodel of 9Jt w.r.t. n instant
i' in 9Jt', and let F be any set of sentences closed under subsentences. Suppose
that for each agent term a, and each sentence A, [a stit: A] 6 F only if SOT', i'
A . Then for every sentence A e F, every m g z', and every /i in 931 with m
Given our induction hypothesis and an earlier note about the relation between
m' ='w m and m' =, m, we need only to assume (9) and show that w stated in
(9) must also be in Tree'. Let us suppose for reductio that w Tree'. Consider
any m' 6 i'. If m' = m, then 931, m" t= A by induction hypothesis. If m' j^ m,
then, w' <' m' and w' <' m for some w' Tree' by historical connection. It
18. Doing and refraining from refraining 457
follows from no backward branching that w < w'. and then m' = m by Lemma
18-2. Hence OT, m' 1= A by (9); and so 9JI, m" t= ,4 by induction hypothesis. It
follows that 9JI, ' 1= A, contrary to hypothesis
18-10 THEOREM. (Doing is not implied by refraining from refraining from do-
ing) Let 6 be any BT + I + AC + bc structure. Then
PROOF. By Remark 18-4(iii), let {w*, WQ, u>i, ...} be any busy choice sequence
earlier than i in 6. Then BC = {w0, wi, ...} is still a busy choice sequence in
6 and BC < i. Set c = {w: w* < w ^ w0}, Tree' =
G c & w' ^ w]} and <' = ^n(Tree' x Tree'). It is easy to see that (Tree1, ^'}
satisfies historical connection and closed forward w.r.t. (Tree, ^}, and hence it
determines a unique substructure 6' of (3. Since BC C Tree', 5(7 is a busy
choice sequence in 6' and BC < i' = id Tree'. So by Lemma 18-7, there is a
model 9JT = (6', 3') such that
We can thus define a BT + I+ AC model 5Jt on 6 in such a way that for every
m i and every h with m h, if m e i', then (m, /z) 3(<?') iff {m, /i'} 3'(g')
for every prepositional variable g', where /i' = /ifl Tree'; and if m i i1, then
(m, /i) ^ 3(g). Clearly OT' is a submodel of 9Jt w.r.t. z' in OT'. It follows from
(10) and Lemma 18-9 (setting F to be the set of all subsentences of ~[a stit:
~fd! sizi: (/ll) that
Let us set M =
by (12). Since 9JI, M h ~g, we know by Remark
18-3 (ii) that
Suppose for reductio that for someTOO M, 971, mo N [a stit: ~[a stit: 9]] with
witness w'. Since w* < mo and w' < mo, either u>' ^ w* or w* < w' by no
backward branching. If w' ^ w*, then by (12) and Lemma 18-2, m0 =, m for
all m e i', and hence, 9Jt, i' N ~[a stit: 5], contrary to (11). Thus it must be
the case that w* < w'. Consider any counter mi to [a stit: ~[a stit: q}} at mg.
Since w* < w' < mo and w' < mi, m-i ^ mo by Lemma 18-2, and hence mi
m* since mo M. It follows that either mi i' or mi M. But mi e
458 Proofs and models
M contradicts (13) because OT, m\ t= [a stit: q}. Hence mi e i'. Since w* < w'
< mi, it follows from no backward branching that w' e Tree', and hence mo 6
i', contrary to the assumption that mo 6 M. From this reductio we conclude
that 9Jt, M t= ~[a siii: ~[a sM: <?]], and hence,
Consider M' = {m: m e i - i ' & w * < m & m X^,, m*}. By our definition of
m and Remark IS-Sfnj, OT, M' 1= ~[a sh<: gj. It is easy to see by (12) that for
each m' e M', m" e M' for all m" e i with m" =,, m', and hence by (11)
and Remark 18-3(i), SEJl, M' t= [a 5^ii: ~[a sizi: <?]]. Since w* is an element of
a busy choice sequence, Choice^, ^ -H^(w), and so M' ^ 0, and hence by (14)
and Remark l8-3(iii), SOT, m N [a stit: ~[a sizi/ ~[a sizi: q}}] for all m e i with
It follows from (12) that 271, i' N [a siif: ~[a stit: ~[a sizi: g]]].
We can then complete the proof by applying (11).
Appendix: Lists
for reference
Here we list in one place many of the items to which we repeatedly refer. Cross-
references to this appendix, either to its sections or to one of its numbered
statements, are indicated by the use of one of the boldface forms occurring in
the following display.
1. Stit theses Thesis 1-Thesis 6.
2. Structures of various kinds.
3. BT + I + AC postulates Post. 1-Post. 10.
4. Definitions of important BT + I concepts Def. 1-Def. 9.
5. Definitions of concepts involving choice Def. 10-Def. 14.
6. Fundamental semantic concepts Def. 15-Def. 16.
7. Derivative semantic concepts Def. 17-Def. 20.
8. The grammar of the mini-language we introduce.
9. A few concepts from the axiomatics of stit theory, Ax. Conc. 1,
Ax. Conc. 2, and Ax. Conc. 3.
We do not list recursive semantic clauses in this appendix; instead, we refer to
the proper sections of chapter 8.
459
460 Appendix: Lists for reference
We remind the reader that we offer these theses as worthwhile heuristics, but
not as pronouncements to be taken strictly. In some cases, indeed, we call
explicit attention to how one or more can usefully be modified, while nevertheless
continuing to advance them as excellent rough approximations.
2 Structures
We list the most important classes of structures that we treat. In each case, if
quantifiers are not being considered, Domain may be missing. (Note. We use
U
BT" for example, both as a singular term naming a class of structures and as
an adjective modifying "structure" or "model.")
Discussed in 7A.
That concludes the list of postulates for BT + I + AC theory, that is, the
theory of agents and choices in branching time with instants.
Appendix: Lists for reference 463
4 Branching-time-with-instants
definitions: Def. 1-Def. 9
We list key definitions of BT +1 + AC concepts. Most are intended as revelatory.
In each case we suppose that we are given a BT + I+AC structure, (Tree, ^,
Instant, Agent, Choice).
Discussed in 7A.
i >m = {WQ: m < mo & mo e z}. We say that i|>m is i/ie horizon from
moment m ai instant i.
Where i\ and 12 are instants, we may induce a linear time order (not a
causal order!) by defining i\ ^ i^ iff m\ ^ m^ for some moment m\ in ii
and some moment 7712 in ^2. Instants can also be temporally (not causally)
compared with moments, m: i\ < m iff mi < m for some moment m\ in
i\\ and m < 12 iff m < m? for some moment m2 in i% .
Discussed in 57A.5.
v. Choice^ (mi) is defined only when instants are present, and when m\ is
properly future to m. Recall that i( mi ) is the instant on which TOI lies.
Then Choice1^ (mi) is defined as the set of all moments on the instant i(mi)
that also lie on some history in CViozce^(mi). In symbols, Choice^L(m\]
= *(mi) n (JChoice^(mi). 1
Discussed in 7C.2.
Discussed in 7C.2.
Discussed in 8E.
Discussed in 8E.
Discussed in 8E.
8 Grammar
Part of our project involves speaking of a mini-language that we offer as illumi-
nating. Here we indicate our ways of speaking of its grammar, and, within the
general semantic framework outlined in 6 and 7, we point to chapter 8 for
the semantics of each feature of the language.
Base clauses. Typically any one of our discussions draws on only some of the
following items, which we introduce by simultaneously describing how we speak
of the language that is our target. See 8F.l for their semantics.
p (and sometimes q) ranges over prepositional variables.
u ranges over individual constants, including two sorts of special terms:
(i) a ranges over agent terms (and frequently over the agents themselves),
and (M) f is a term that artificially denotes "the non-existing object," to
be available as a throwaway value of definite descriptions when existence
or uniqueness fails.
Xj ranges over individual variables.
/ ranges over operator letters.
F ranges over predicate letters.
Stit-free functors.
Vx3A and tXj(A). See 8F.3.
Sett:A, Poss:A, and Can:A. See 8F.4.
Prior tenses Was:A, Will:A, Was-always:A, and Will-always:A, and tem-
poral operators At-instt:A and At-momt:A. See 8F.5.
Stit functors.
The achievement stit, [a astit: A}. See 8G.3 (witness by moments) and
8G.4 (witness by chains).
The Brown stit, [a bstit: Q], is mentioned in 1D and discussed in Horty
2001.
The deliberative stit, [a dstit: A]. See 8G.l.
The Chellas stit, [a cstit: A}. See 8G.2.
The strict joint stit, [F sstit: Q}. See jjlOC.3.
The transition stit, [a tstit: m =^4> A]. See 8G.5.
Plain stit. [a stit: A]. Used both for the achievement stit when there is
no ambiguity, and for the general stit idea.
Lai is the Logic for the basic achievement stit with 1 agent, without the refref
equivalence (hence with the possibility of busy choice sequences). The
language of Lai contains truth functions together with- stit sentences for
just a single agent, a. See chapter 16.
Lai + rr is the Logic for the achievement stit with 1 agent and the refref
equivalence (hence with no busy choice sequences). The language of
Lai + rr is the same as the language of Lai. See chapter 15.
Ldl is the Logic for deliberative stit with 1 agent. In addition to truth
functions and stit sentences, the language of Ldl includes Sett:. See
chapter 17.
Ldm is the Logic for the deliberative stit with many agents. In addition to
truth functions and stit sentences, the language of Ldm includes identity,
and hence can express multiplicity of agents. See chapter 17.
Ldmn (for n ^ 1) is the Logic for the deliberative stit with many agents,
where at each moment each agent is limited to at most n + 1 choices. The
language of each Ldmn is the same as the language of Ldm. Also Ldmo
is defined as Ldm. See chapter 17.
SA is the chapter 11 combination of Stit theory with Andersonian devices; or
the Sanction with a logic of Agency. SAo is the fragment of SA described
in 11C.
KD is the standard system of deontic logic from F011esdal and Hilpinen 1971.
S4 and S5 are the standard modal logics of C. I. Lewis.
This page intentionally left blank
Bibliography
Almog, J., Perry, J., and Wettstein, H., eds. 1989, Themes from Kaplan, Oxford
University Press, Oxford.
Anderson, A. R. 1956, The formal analysis of normative systems, Technical re-
port 2, Contract no. SAR-NONR-609(16), Office of Naval Research, New Haven.
Reprinted in Rescher 1967, pp. 147-213.
Anderson, A. R. 1962, "Logic, norms, and roles," Ratio 4, 36-49.
Anderson, A. R. 1970, "The logic of Hohfeldian propositions," Logique et analyse
49-50, 231-242. Reprinted in the University of Pittsburgh law review 33 1971,
29-38.
Anscombe, G. E. M. 1957, Intention, Basil Blackwell, Oxford.
Aqvist, L. 1972, Performatives and venfiabihty by the use of language, Philosophical
Studies No. 14, Uppsala (Philosophical Society and the Department of Philoso-
phy, University of Uppsala).
Aqvist, L. 1974, "A new approach to the logical theory of actions and causality," in
Stenlund 1974, pp. 73-91.
Aqvist, L. 1978, "An analysis of action sentences based on a 'tree' system of modal
tense logic," in C. Rohrer, ed., Papers on tense, aspect, and verb classification,
TBL Verlag Gunter Narr, Tubingen, pp. 111-161.
Aqvist, L. 1984, "Deontic logic," in Gabbay and Guenthner 1984, pp. 605-714.
Aqvist, L., and Mullock, P. 1989, Causing harm: A logico-legal study, Walter de
Gruyter, Berlin.
Austin, J. L. 1961, Philosophical papers, J. Urmson and G. Warnock, eds., Oxford
University Press, Oxford. Essays from the 1940s and 1950s.
Austin, J. L. 1975, How to do things with words, J. Urmson and M. Sbisa, eds., Harvard
University Press, Cambridge, MA. The William James Lectures delivered at
Harvard University in 1955.
Badecker, W. 1987, Formal grammars and the analysis of infinitives, Indiana Univer-
sity Linguistics Club, Bloomington.
Barcan, R. 1946, "A functional calculus of first order based on strict implication,"
Journal of symbolic logic 11, 1-16. See also entries under Marcus.
Barcellan, B., and Zanardo, A. 1999, Actual futures in Peircean branching-time logic.
Unpublished manuscript.
Bartha, P. 1993, "Conditional obligation, deontic paradoxes, and the logic of agency,"
Annals of mathematics and artificial intelligence 9, 1-23.
Bartha, P. 1999, "Moral preference, contrary-to-duty obligation and defeasible
oughts," in P. McNamara and H. Prakken, eds., Norms, logics, and informa-
tion systems: New studies in deontic logic and computer science, IOS Press,
Amsterdam, pp. 93-108.
Belnap, N. 1991, "Backwards and forwards in the modal logic of agency," Philosophy
and phenomenological research 51, 777-807.
475
476 Bibliography
Brown, M. 1992, "Normal bimodal logics of ability and action," Studio, logica 51, 519-
532.
Burgess, J. 1978, "The unreal future," Theoria 44, 157-174.
Burgess, J. 1979, "Logic and time," Journal of symbolic logic 44, 556-582.
Burgess, J. 1980, "Decidability for branching time," Studio, logica 39, 203-218.
Burgess, J. 1984, "Basic tense logic," in Gabbay and Guenthner 1984, pp. 89-133.
Castaiieda, H.-N. 1954, "La logica general de las normas y la etica," Universidad de
San Carlos, Guatemala, vol. 30, pp. 129-196.
Castaneda, H.-N. 1974, The structure of morality, Charles C. Thomas, Springfield, IL.
Castafieda, H.-N. 1975, Thinking and doing, D. Reidel Publishing Company, Dor-
drecht.
Castaneda, H.-N. 1976, "The twofold structure and the unity of practical thinking,"
in M. Brand and D. Walton, eds., Action theory, D. Reidel Publishing Company,
Dordrecht, pp. 105-130.
Castaneda, H.-N. 1981, "The paradoxes of deontic logic: The simplest solution to all
of them in one fell swoop," in Hilpinen 1981, pp. 37-86.
Chellas, B. F. 1969, The logical form of imperatives, Perry Lane Press, Stanford, CA.
Chellas, B. F. 1980, Modal logic: An introduction, Cambridge University Press, Cam-
bridge.
Chellas, B. F. 1992, "Time and modality in the logic of agency," Studia logica 51, 485-
518.
Chisholm, R. 1963, "Supererogation and offence: A conceptual scheme for ethics,"
Ratio 5, 1-14.
Chisholm, R. 1964a, "Human freedom and the self," The Lindley Lecture. Reprinted
in K. Lehrer, ed., Freedom and determinism, Random House, New York, 1966,
pp. 11-44; and in R. Chisholm, ed., On metaphysics, University of Minnesota
Press, Minneapolis, 1989, pp. 5-15.
Chisholm, R. 1964b, "Symposium: Action, motive, and desire. The descriptive element
in the concept of action," Journal of philosophy 90, 613-625.
Chisholm, R. 1966, "Freedom and action," in K. Lehrer, ed., Freedom and determinism,
Random House, New York, pp. 11-44.
Chisholm, R. 1969, "Some puzzles about agency," in K. Lambert, ed., The logical way
of doing things, Yale University Press, New Haven, pp. 199-217.
Chisholm, R. 1971, "Reflections on human agency," Idealistic studies 1, 36-46.
Danto, A. 1973, Analytical philosophy of action, Cambridge University Press, Cam-
bridge.
Davey, K. 1999, Obligation and the conditional in stit theory. Unpublished manuscript.
Davidson, D. 1966, "The logical form of action sentences," in Rescher 1966, pp. 81-95.
Davidson, D. 1980, Essays on actions and events, Oxford University Press, London.
Dazeley, H., and Gombocz, W. 1979, "Interpreting Anselm as logician," Synthe.se.
40, 71-96.
Di Maio, M. C., and Zanardo, A. 1998, "A Gabbay-rule-free axiomatization of T x W
validity," Journal of philosophical logic 27, 435-487.
Dummett, M. 1991, The logical basis of metaphysics, Harvard University Press, Cam-
bridge, MA.
Elgesem, D. 1989, Intentions, actions and routines: A problem in Krister Segerberg's
theory of action. Unpublished manuscript.
Fitch, F. B. 1963, "A logical analysis of some value concepts," Journal of symbolic
logic 28, 135-142.
F011esdal, D., and Hilpinen, R. 1971, "Deontic logic: An introduction," in Hilpinen
1971, pp. 1-36.
478 Bibliography
Horty, J. F. 2001, Agency and deontic logic, Oxford University Press, Oxford.
Horty, J. F., and Belnap, N. 1995, "The deliberative stit: A study of action, omission,
ability and obligation," Journal of philosophical logic 24, 583-644. Reprinted in
The philosopher's annual, vol. 18, ed. P. Grim, Ridgeview Publishing Company,
Atascadero, CA, 1997.
Humberstone, I. L. 1976, An alternative account of bringing about. Unpublished
manuscript; read to a meeting of the Australasian Association for Logic in 1976.
Humberstone, I. L. 1977, "An alternative account of bringing about (with a pinch
of relevance)," Bulletin of the section of logic 6, 144-145. Reprinted in Rel-
evance logic newsletter 2, 1977, 107-108. This is an abstract of the unpub-
lished Humberstone 1976.
Kane, R. 1998, The significance of free will, Oxford University Press, Oxford.
Kanger, S. 1957, New foundations for ethical theory, Stockholm. Reprinted in R.
Hilpinen, ed., Deontic logic: Introductory and systematic readings, D. Reidel
Publishing Company, Dordrecht, 1971, pp. 36-58.
Kanger, S. 1972, "Law and logic," Theoria 38, 105-132.
Kanger, S., and Kanger, H. 1966, "Rights and parliamentarism," Theoria 6, 85-115.
Kaplan, D. 1989, "Demonstratives: An essay on the semantics, logic, metaphysics,
and epistemology of demonstratives and other indexicals"; and "Afterthoughts,"
in Almog, Perry,- and Wettstein 1989, pp. 481-563; 565-614.
Kenny, A. 1963, Action, emotion, and will, Routledge and Kegan Paul, London.
Kripke, S. 1959, "A completeness theorem in modal logic," Journal of symbolic logic
24, 1-15.
Kripke, S. 1963, "Semantical considerations in modal logic," Acta philosophica fennica
16, 83-94.
Lemmon, E. J. 1966, "Comments on D. Davidson's 'The logical form of action sen-
tences' ," in Rescher 1966, pp. 96-103.
Lewis, D. 1970, "Anselm and actuality," Nous 4, 75-88.
Lewis, D. 1986, On the plurality of worlds, Basil Blackwell, Oxford.
Lindahl, L. 1977, Position and change: A study in law and logic, D. Reidel Publishing
Company, Dordrecht.
Lukasiewicz, J. 1920, "Three-valued logic," Ruch Filozoficzny 5. Reprinted in S. Mc-
Call, ed., Polish logic, 1920-1939, Oxford University Press, Oxford, 1976.
MacFarlane, J. 2000, What does it mean to say that logic is formal? Ph.D. diss.,
University of Pittsburgh.
Makinson, D. 1986, "On the formal representation of rights relations," Journal of
philosophical logic 15, 403-425.
Marcus, R. B. 1960, "Extensionality," Mind 69, 55-62.
Marcus, R. B. 1961, "Modalities and intensional languages," Synthese 13, 303-322.
Marcus, R. B. 1963, "Classes and attributes in extended modal systems," Acta philo-
sophica fennica 16, 123-136.
Marcus, R. B. 1966, "Iterated deontic modalities," Mind 75, 580-582.
Marcus, R. B. 1972, "Quantification and ontology," Nous 6, 240-250.
Marcus, R. B. 1974, "Classes, collections, and individuals," American philosophical
quarterly 11, 227-232.
Marcus, R. B. 1977, "Dispensing with possibilia," in Proceedings of the American
Philosophical Association, pp. 39-51. Presidential Address, Western Division,
American Philosophical Association, delivered April 30, 1976.
Marcus, R. B. 1980, "Moral dilemmas and consistency," Journal of philosophy 77, 121-
136.
480 Bibliography
483
484 Index
definitions w.r.t. agents and choices, and refraining from refraining, 50, 52,
466-468 451
definitions concerning BT, 462-465 and strategies, 349
general structure, 444 bystander, mere, 285, 290
postulates for agents and choices, listed,
462 c
postulates for BT, listed, 461 a chain, 181, 463
SAo complete w.r.t., 304 a context of use, 145
SAo sound w.r.t., 304 Calvin, J., 163
structure, 35 Can:, 242, 329
theory, 28 canonical frame, 441
BT + AC with max. n choices cant (can see to it at i), 249
Ldmn complete w.r.t., 440-445 with multiple agents, 284
Ldmn sound w.r.t., 439-440 caring about the future, 209
BT + I + AC (agents and choices in Carnap, R., 328, 334
branching time with instants) Castaneda, 22, 74-78, 316, 328
at-most-binary structure, 434 1954, 21
definitions w.r.t. agents and choices, 1974, 12, 74, 75
466-468 1975, 16, 74, 75
definitions concerning BT, 462-465 1976, 75, 77
definitions concerning instants, 465-466 1981, 21, 77, 295, 296, 309
Lai sound w.r.t., 417-419 categorematic expressions, 234
model, 35, 228, 468 causal future of possibilities, 180, 463
point, 228, 469 causal past, 180, 182, 463, 465
postulates for agents and choices, listed, causal-order relation, 139, 179, 180, 463
462 postulates for, 180
postulates for BT, listed, 461 causality, 251
postulate for Domain, listed, 462 in branching time, 209-210
postulates for instants, listed, 461-462 cusp of, 210
structure, 35, 226 chains
theory, 28 definition of, 181, 463
BT + I + AC+bc (BT + I + AC with lower bound of, 182, 464
busy choosers), 451 upper bound of, 182, 464
refref invalid in, 454-458 witness by, 49, 249-250
BT + I + AC+nbc (BT + I + AC with no characteristic sentence, 393
busy choosers), 381 Chellas, 5n, 85-87
Lai + rr sound w.r.t., 390-391 1969, 22, 85, 197, 297n
Lai + rr complete w.r.t., 407-414 1992, 86, 86n, 248, 274, 275, 297, 297,
bundled trees, 198-199, 202 298n, 329
Burgess Chellas stit, 23n, 86, 297, 436
1978, 159, 198, 205, 207 definition of, 298
1979, 198 semantics for, 248
1980, 198 Chisholm, 5n, 22, 65-68
1984, 224 1963, 312, 313
busy choice sequence, 218, 250, 383, 453, 1964a, 19, 21, 263
468 1969, 65, 68
busy chooser, 49, 219, 265-268, 468 Davidson's objections to, 80
and consistency condition for deontic choice box, 300
trees, 373, 374 choice
definition of, 50 equivalence, 36, 214, 248, 467
Index 487
names for, 243, 244 1989, 141, 145, 149, 151, 221, 225-226,
postulates for, 194, 462 229-234
role in stit theory, 194 KD, 295n, 300, 301
theory of, 194-196 Kenny, 5n, 20, 68-74
and times, 194 1963, 21, 22, 87
instructions. See imperatives kinematic condition (on deontic trees),
intensional predication, 328n 366, 375-377
intention, history-dependence of, 56n knowledge
intentionality history-independence of, 56
and action, 33, 79 implies truth, 56
and promises, 119 Kremer, P., 216
interdefinability of stit operators, 298, Kripke, 139, 151, 195, 224, 334
298n 1959, 179
interpretation, 227, 468. See also 6- 1963, 328
interpretation
settled, 434 L, 141, 227
interpretation parameter(s), 144, 227 Lai (astit with 1 agent)
immobility of, 144, 228 axiomatics of, 416-417
terminology, 144 completeness of, 428-434
intervals, 179 no finite model property for, 434
invitations, 93 soundness w.r.t. BT + I + AC, 417-419
inviting, 111 Lali, 434
Lai + rr (Lai +refref), 383-385
joint agency, 281-291 and at-most-binary structures, 414
strict, alternative concept of, 113 completeness of w.r.t. BT + I + AC
joint stit + nbc, 407-414
applications of, 288-290 decidability of, 414
carry-over from singular stit, 283 finite model property for, 407-414
essential and inessential agents for, 286 soundness w.r.t. BT + I+AC
essential and inessential for, 285 + nbc, 390-391
grammar of, 282 later. See causal-order relation
other-agent nested, 290-291 laws, 245
plain, 282-284 Ldrn (dstit with many agents)
semantics for, 283 axiomatics of, 436-439
strict, 284-287 Ldm0 (= Ldm), 437
strict, alternative account, 113 Ldmn (Ldm + APCn), 437
weakening of, 285 completeness of w.r.t. BT + AC with
max. n choices, 440-445
K (class of structures), 237, 471 decidability of, 450
-valid, 237, 471 finite model property for, 445-450
K-axiom, 301 soundness of w.r.t. BT + AC with max.
Kamp structures, 198, 202 n choices, 439-440
Kamp, H., 246 least upper bound, 182, 464
Kane 1998, 28, 204 Lee-Hamilton, E., 183
Kanger, H., 5n legal sequence (of parameters), 142
Kanger, S., 5n Leibniz, G. W., 163
1957, 20, 22 Lemmon 1966, 81
1972, 23 Lewis, 151, 164, 195, 246
and Kanger 1966, 21, 22 1970, 163
Kaplan, 242, 246 1986, 170, 179, 188, 196, 205-209
Index 493