Godfrey-Smith. Triviality Arguments Against Functionalism (Article)

Triviality Arguments against Functionalism
Author(s): Peter Godfrey-Smith

Source: Philosophical Studies: An International Journal for Philosophy in the Analytic Tradition
, Vol. 145, No. 2 (Aug., 2009), pp. 273-295
Published by: Springer
Stable URL: http://www.jstor.org/stable/27734479
Accessed: 30-12-2015 21:51 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.
Springer is collaborating with JSTOR to digitize, preserve and extend access to Philosophical Studies: An International
Journal for Philosophy in the Analytic Tradition.
http://www.jstor.org
This content downloaded from 130.91.117.231 on Wed, 30 Dec 2015 21:51:02 UTC
All use subject to JSTOR Terms and Conditions
Philos
DOI
Stud (2009) 145:273-295

10.1007/sl
1098-008-9231-3
arguments
Triviality
against
functionalism
Peter Godfrey-Smith
Published
?
online:
26 April 2008
Science+Business
Media
Springer
B.V.
2008
Abstract
in the philosophy of mind
"Triviality arguments" against functionalism
hold that the claim that some complex physical system exhibits a given functional
is either trivial or has much less content than is usually supposed. I
organization
several
survey
earlier
some limitations
but
tional
to revise
functionalists
requires
of
arguments
this kind,
and
present
in the earlier arguments. Resisting

popular
views
a new
one
overcomes
that
is possible,
triviality arguments
about
the
of
"autonomy"
func
description.
Functionalism
Keywords
Mind
Computation
Realization
Causal
role
1 Introduction
Functionalism
have
them
in the philosophy
in virtue
of
the
of mind holds
systems'
functional
that systems with mental

and
organization,
properties
particular
mental
states such as pains and hopes are functionally individuated internal states. "Triviality
as I will call them, hold that the claim that some
arguments against functionalism,"
complex
functional
physical
system
state,
is either
exhibits
trivial
a
functional
given
organization,
or has much
less content
than
or
is in a particular
is usually
supposed.
I group in this category a family of arguments with different ambitions. An early

version is attributed to Ian Hinckfuss,
in discussion in the 1970s.1 The "Hinckfuss'
1
to Cleland
in a 1978 discussion
of computation
(2002), Hinckfuss'
According
argument was presented
at the Australasian
Association
of Philosophy.
Lycan (personal communication)
says the discussion was
of an early version of Lycan (1981) at the conference,
a paper which then appeared
during presentation
with a presentation
of the Hinckfuss
argument. Lycan treats the argument as something different from a
in my sense, however; Lycan says the bucket of water might, by chance, come to
triviality argument
over some interval.
realize a human's functional organization
P. Godfrey-Smith
(El)
of Philosophy,
Department
e-mail: pgs@fas.harvard.edu
Harvard
University,
Cambridge,
MA
02138,
USA
4y Springer
274
P. Godfrey-Smith
is often described by saying that a bucket of water
pail" argument
so much
has
to realize
taken
or any
these
other
program.
computer
asserting
nor
Hinckfuss
gave
that
as running
of interpreting his wall

Neither
(1990)
be
there
the program
Searle
gave
proofs
claims.
Putnam
two
gave
(1988)
functionalism
the
computationalism
legitimate way
it can
states,
Searle
the mind,
about
sitting in the sun

of
categorization
a human
agent.
of
organization
against
be some
Wordstar,
functional
a suitable
via
that,
complexity
the
argument
triviality
would
of
causal
collapses
input-output
more
The
arguments.
associated
properties
some
with
we know that it can be interpreted as having

Putnam did offer proofs. Chalmers
(1996)
several
developed
modifications
of
but
conclusions,
surprising
them.
denies
He
that
functional
particular
that
claims
that a system has

organization,
that functional organization as well.

criticized Putnam's
arguments but
takes
some
they
endanger
of
these
to have
arguments
or
computationalism
(1996) gave a formal treatment of Searle's
about the mind. Copeland
functionalism
one
important
Once we know
into logical behaviorism.
argument.
some
So
on
of
these
take
arguments
are
Some
computationalism.
as
functionalism
with
given
their
while
proofs,
target,
while
others
rely
others
on
focus
intuitions.
Some conclude that a functional or computational description

is entirely trivial,
when applied to a sufficiently complex system, while others conclude only that
functional descriptions collapse into behavioral descriptions.
Given the threat that such arguments pose, it is surprising how little they have
been
discussed,
accounts
do
problems
schematic
Here
as
especially
of the realization
not
arise.
the mainstream
accounts
Many
functionalist
of
realization
used
examine
the
problems
such
by
posed
use
not
does
it clear that triviality

are
functionalists
by
so
are handled.
that it is uncertain how these problems

I
literature
structures that make
of functional
for
arguments
mainstream
functionalism
in philosophy of mind. The discussion covers functionalist accounts
of both "folk" and scientific psychological
but the analysis of
properties,
as
is regarded
computation
a separate
I also
question.2
not
do
consider
ideological
versions of functionalism
that understand function in a rich biological
sense.3 The
is restricted to what might be called "dry" functionalism,
discussion
of the kind
seen in Fodor (1981), Stich (1983), Braddon-Mitchell
and Jackson (1996), Crane
(1995),
and
framework
physical
After
(Sects.
They
behavioral
treatment
The
with
are
arguments
mainstream
also
functionalism;
within
presented
employ
larger
2-4).
are
about
description.
of
the
stronger
Two
the
The
claims
are modifications
content"
"surplus
third
of
associated
arguments
that
functional
which
argument,
with
seen
Hinckfuss
is
new,
and
three
in Putnam
description
is a more
Searle.
of
picture
systems and a set of idealizations associated with that project.

an initial discussion
of functional
I present
characterization,
arguments
Chalmers.
others.
many
associated
and
has
I then
over
precise
argue
to say exactly what computationalism

it is difficult
about the mind
is committed
to, it is
Although
intended to be a stronger claim than functionalism
is
(Smith 2002; Piccinini
2004). Computationalism
of the functionally
characterized
supposed to involve a claim about particular characteristics
operations
that comprise cognition.
3
Lycan (1981) treats this as part of the answer
to Hinckfuss'
pail.
Springer
arguments
Triviality
(in Sect. 5) that while

it is not
fatal.
These
the threat raised by these problems has been underestimated,

the
Avoiding
include
the
275
functionalism
against
revision
is possible,
problems
of
popular
but
about
claims
has
the
costs.
and
consequences
"autonomy"
of
functionally
characterized
functional
description.
of an FSA
2 Realization
There
are
states
and
two
of
ways
their
in a precise
talking
realization.
specifies
a functional
understands
realization
functionalism,
entities,
and
structures. The other, deriving

similar.
something
hypotheses
roles.
specify
The
tools
used
to
According
set of roles,
in this
with
and
second
various
are
paper
transitions."
Initially
borrowed
set
of
and
inner
inner
state
outputs,
inputs,
look
of present
states
and
from
functional
input
the
are
related
inner
set
of
occupants
of
computation,
roles are described
state
rules
and
using
in a
to be
is taken
system
by
of
those
(In the final Section
state automaton
to a new
sentence
interlocking
discussions
either approach.
where
that
be
may
objects
at cases
an
approach,
state at each time. A finite
single total functional

combination
we
(1972), uses a Ramsey
from Lewis
the
but the arguments are supposed to work within

I discuss the relations between the two.)
My starting point is a framework in which
"state
the
of
computationalist
style
as a set of relations
between
abstract
profile
in terms
of a mapping
between
abstract
and
physical
or
about
way
associated
One,
(FSA) is a finite
that map
each
This
output.4
is
one way of formalizing the kind of functional characterization often envisaged in

the days of "machine functionalism"
(Putnam 1960; Block and Fodor 1972). I will
use
the
2.1
Coke
following
machine
I begin with
modifies
of
example
state
which
corresponds
to a simple
coke
machine.
transitions
(S1,I1)^(S2,01)
(S^^^Oi)
(S2,Ii)^(S3,Oi)
(S3,Ii)-+(Si,02)
(S2,I2)^(Si,02)
(S3J2)-^(Su03)
the following
a formulation
an FSA,
due
account of the realization
of this kind of structure, which
to Chalmers.
FSA: A physical system realizes a given FSA during a time interval iff there is
mapping M from states of the physical system onto states of the FSA, and
from inputs and outputs of the physical system onto inputs and outputs of the
FSA, such that: for every state-transition (S, I)
(S', O) of the FSA, if the
physical system were to be in state P and received input I* such thatM(P) = S
as the outputs are associated with

this is a "Mealy machine,"
not a "Moore machine,"
Technically,
transitions rather than states. Some early discussions
of functionalism
focused on Turing machines.
I take
to be an unpromising model for the mind, though important for in principle
themselves
Turing machines
discussions
of the mechanization
of intelligence. The CSA framework, discussed below, can be used to
as Chalmers
(1996) notes.
represent Turing machines,
4y Springer
276
P. Godfrey-Smith
and M(I*) = I during this time interval, then itwould transition to state P' and
would emit output O* such thatM(P') = S' and M(0*) = O.
criterion for realization.
This will be called a "simple mapping"
two
The FSA itself treats inputs and outputs abstractly. It merely distinguishes
and
this
number
of
possible
inputs
inputs and three outputs. Any system with
outputs could, if appropriately organized, realize the FSA. When specifications of
this kind are used in philosophy of mind, it is natural to require that a system's
inputs and outputs be of a specific kind. Just as a coke machine has to be able to
and
money
accept
out
give
an
cokes,
agent,
intelligent
has
perhaps,
to be
to
able
track and act on the world in particular ways. If this is right, then an FSA understood
as a mathematical
object only specifies the formal backbone of a functional
structure in the sense relevant to philosophy of mind (Block 1978).
So where necessary, I will distinguish between an FSA in a broad sense and in a
narrow sense. An FSA in the broad sense includes specification of particular inputs
and outputs; the FSA in the narrow sense is just the formal backbone, with inputs
and outputs treated abstractly. Specifying the coke machine FSA in the broad sense
includes giving both the state transitions above, and the following specification of
century pricing).
inputs and outputs (complete with anachronistic mid-twentieth
Coke machine
Ii
in terms
I2
a
causes
This
inputs and outputs
5 cents;
of
10 cents;
that
null;
preserves
between
relations
them
between
relations
physical
O]
02
in discussions
complication
mapping
mathematical
or other
of
coke;
one
them
on
there
must
and
side,
the
coke
and
If realization
realization.
relations,
on
03
other.
be
physical
a
But
abstract
objects
mapping
5 cents.
is understood
objects
with
with
causal
criterion
for
realization can then only be applied directly to the narrow sense FSA. Showing the
realization of a broad sense FSA then involves two steps. One is showing a mapping
between the narrow sense FSA and the physical system. The other is showing that
the inputs and outputs of the physical system are of the right kind.5
In general,
below,
sense
broad
FSAs
and the symbols
concrete,
physical
except
will
be more
"I" and "O" will

where
important
than
narrow
sense
FSAs
be used for inputs and outputs
in a
indicated.
otherwise
The first triviality argument aims to show that any sufficiently complex system
with the input-output dispositions associated with a given FSA (broad sense) is also
a realization of the FSA. My presentation uses graphical methods,
and modifies
to
I
Putnam
FSAs
due
and
Chalmers
(1996).
(1988)
represent
arguments
using what
I will
call
contingency
trees.
The
tree
for
the coke
machine
is represented,
over
three
rounds of input and output, in Fig. 1.

The same kind of tree can be used to represent the dispositional properties of a
physical object. If we specify a set of inputs and outputs, any physical system
starting
in a particular
state
will
have
some
set of
dispositions
in response
to
those
5
of mind as involving specific
If a functionalist
does not see the functional roles relevant to philosophy
kinds of inputs and outputs, perhaps because of cases of humans with unusual interfaces with the world,
can be used on its own. This makes Hinckfuss-type
then the mapping
arguments more
approach
in Sect. 4.
threatening. This issue will be discussed
4y Springer
arguments
Triviality
Fig. 1 Contingency
coke machine
FSA
277
functionalism
against
tree for the
inputs (if it can interact with them at all), and itmay

relevant
kind.
dispositional
Such
properties,
world,
all
will
usually
can
we
though
all the influences
interval, representing
and
tree
the ways
it would
only
imagine
(ormay not) emit outputs of the

some
represent
a "full"
tree
the
object's
an
over
object
receive from the rest of the
the system might

its environment
affect
of
an
for
in response.
Suppose we have a physical object whose input-output dispositions over some

time interval are the same as those associated with a particular FSA. The object goes
through some particular sequence of states, emitting the right outputs in response to
a series of inputs, and was also disposed to emit suitable outputs if its input history
are the product of the
had been different. Assume
also that these dispositions
to being mediated
internal structure of the object, as opposed
by an external
trees for the FSA and the physical object can then be
controller. The contingency
superimposed on each other. (Imagine doing this with transparent slides.) The only
differences are found in the inner states represented at each point. Provided that the
there will be a mapping between
physical system has sufficient overall complexity,
physical states and FSA states such that the physical system is a realization of the
FSA.
sense
relevant
The
total
physical
particular
state,
of
state
at
Pj,
and
"sufficient
every
overall
point
each
input
on
the
it may
The
is as follows.
complexity"
system's
tree
starts
is unique.
The
in a
system
receive
unique state, but also affect the system's physical

only
As
ways.
microphysical
a consequence,
system down a path comprising

sequence
of
Putnam
inputs.6
physical
offers
every
will
not
response
only
send
a new
it into
to later inputs, perhaps in
sequence
of
inputs
will
send
the
states that cannot be reached by any other
reasons
why
actual-world
systems
physical
should satisfy a principle of this kind. Instead I treat this as a condition inmy claims
about the realization of functional structures. If a physical system has this kind of
overall
The
complexity
claim
uniqueness
into triviality.
collapsing
(1998).
and
has
the
same
input-output
properties
as a
given
then
FSA,
here is intended to apply to intrinsic properties

of the system,
I assume an account of intrinsicness
along the lines of Langton
to avoid
and Lewis
4y
Springer
it
it
P. Godfrey-Smith
278
is a realization of that FSA. Below all the physical states labeled as P? are assumed
to be distinct from each other.
From here it is simple to show the existence of a mapping of the relevant kind.
two
the
Using
physical
whose
trees,
superimposed
we
on
positions
tree
the physical
each
map
FSA
a disjunction
state that is specified with
to occurrences
correspond
to a coarse-grained
state
inner
states
of all the unique physical

of
that
state
inner
in
tree.
the FSA
More
formally, we inspect the FSA tree and note all locations at which a
particular state, perhaps Si, appears. Each location can be identified independently
of the inner state that appears there, by its place in the space of possible input
sequences. Call the set of locations at which Si appears Zj. We then inspect the
physical tree, and note all the physical states that appear in locations in E {.This set
of physical states is described with a disjunction, labeled Qt. We do the same for the
other
states.
FSA
physical
can
We
where
system,
construct
then
each
on
node
another
yet
tree
the
tree
contingency
is characterized
an
by
for
the
a
and
output
states created by the method

outlined
above. The
of physical
in this tree is a weakening or coarse-graining of that contained in the
original physical tree. Each Q? will appear in all and only the locations occupied by
Si in the FSA tree. As the transition properties of any FSA state are expressed
entirely by its set of locations in the tree, the Qi's can be seen to have the same
disjunction
information
transition
The
as
properties
claim
that
the
that
that
Si's
the Qi's
have
are
they
the
to.
mapped
same
"transition
as
properties"
the
is
Si's
I
subject to a qualification discussed below. Before addressing those complications
will work through the coke machine example to illustrate the procedure.
The contingency tree for the coke machine FSA was given in Fig. 1.A complex
physical
system,
arbitrarily
chosen,
with
same
the
input-output
will
properties,
have
tree pictured in Fig. 2.

the contingency
Behavioral criteria are used to "align" the FSA with the physical system at the
start of the interval. This alignment may yield any FSA inner state as the initial
state, though here I assume the initial state is S i.We superimpose the trees and map
each formal state to a disjunction of physical states as follows.
Let:
Qx =PivP5vP6vP7vP8vP9
Q2=P2vP10vP12vP14
Q3=P3VP4vPi1vPi3VPi5
Then for all i e {1, 2, 3}, M(S?) = Q?

The
tree,
Qi's
a
This
can
fragment
procedure
also
be
used
of which
can
can
used
be
to generate
seen
in
be
to
show
a
Fig.
both
coarser-grained
physical
contingency
3.
"good"
and
"bad"
realizations
of
an
FSA by a physical system. What is the difference? A first response might be that the
"good" cases are those where the physical states disjoined to produce a Q that maps
to some
formal
we
state
are
need not suppose

realizations of an FSA,
similar
in some
non-trivial
sense.
In a real-world
system,
states are identical. But in natural

that these physical
similar
the system should be doing something physically
4y Springer
arguments
Triviality
against
279
functionalism
tree for a
2 Contingency
system with the
physical
behavioral properties of the coke
machine
Fig.
P4,0
p2,o,
p5,o2
P6,02
P3,Oi
p7,o3
when
it reaches
the
states
various
that
each
to
map
some
Sj.
In
the
coke
machine
case, S3 is especially useful to think about in this connection. The FSA has two
different ways of reaching S3 within a single cycle, depending on the inputs
received. Do the physical states the system reaches via these two different paths
have anything in common? If not, the FSA representation of that device seems at
least somewhat misleading. Explanations of the behavior of the machine in terms of
the role of S3 in its functional economy would imply a spurious unity across the
in which that state is involved. This response to the problem, which I
processes
think is indeed along the right lines, will be fleshed out in more detail in Sect. 5.
Before moving on I will discuss some complications
and possible objections.
These
involve
satisfies
First,
aspects.
restricted
the question
of whether
the conditionals
as
number
"terminal"
the
treatment
of
physical
above
procedure
really
that
shows
the
system
used in the criterion for realization. The problem has two
inputs
states,
assumes
above
can
be
states
obtaining
at
time
particular
each
received,
the
Qi
end
interval
as defined
of
the
above
interval.
over
which
will
contain
The
analysis
given does not suppose that we know how the system will respond to inputs when in
these states. So when it is claimed that the system is causally disposed to transition
from
one
to another,
in a way
simplest way of understanding
that
corresponds
these dynamical
to
the FSA
state
relations between
transitions,
the
the Q states is not
3 Coarser-grained
tree for the physical
contingency
system in Fig. 2
Fig.
4y Springer
280
P. Godfrey-Smith
tree has it
available. That simplest option would require that, if the coarse-grained
that if the system is in Qi and receives input 1 itwill transition to Qj, then itmust be
the case that for every state disjoined in Qj, if the system is in that state and receives
input 1 then it will transition to some state in Qj. This does not apply to terminal
states in Qi. So the analysis above requires that conditionals describing how the
system is disposed to move from one Q state to another describe the system's
over
dispositions
I will
the
argue
even
interval
that
these
when
the
antecedent
are
conditionals
true,
include
Q's
an
under
states.
terminal
that
interpretation
is
appropriate for functionalism. The point can be made by imagining a long interval
and a system undergoing a definite aging process over the interval, though the point
applies generally. The Q's are then very long disjunctions, containing physical states
that would
when
encountered
naturally
the
is at various
system
different
ages.
Each
Q includes a terminal state, one that would only be reached (if it is reached at all) at
the end of the interval. Suppose we have an FSA conditional that says if the system is
in Si and receives I1? itwill transition to S2. Then if the system realizes the FSA over
the interval, Q{ maps to Si, and Q2 maps to S2, itmust be true that if the system were
to be inQ! at, for example, the start of the interval, then if it received Ii itwould have
to transition to Q2. What would it involve for the system to be in Q] at the start of the
interval? It would involve the system being in an "age-appropriate" physical state
If
within the disjunction. In possible-worlds
jargon, these are the nearest Qrworlds.
the system would transition appropriately from that physical state at that stage in the
time interval, the conditional is satisfied. To assess whether a system of this kind
realizes an FSA over a specific time interval we do not have to know how the system
would behave if itwere, at the beginning of the interval, in a physical state that could
is intended to capture systems that
only arise late in the interval. So if functionalism
or
is one in
the
relevant
age,
interpretation of the conditionals
physically develop
which
antecedents
and
consequents
are
in a way
assessed
that
location of physical states in the history of the system.7

There is also a second problem. Although a contingency
various
specifies
non-actual
paths
as well
as
the
actual
one,
the
respects
temporal
tree of the kind above

some
states
of
an FSA
might not be reachable from the state the system was in at the start of the interval.
So it will be impossible to assess whether the system respects transitions involving
those FSA states. Consequently,
showing the realization of an FSA in those cases
the
physical system with a number of different contingency
requires describing
trees, each beginning with different initial states. This does not create a problem for
the methods used above. Provided that the physical states at all the places on all the
trees
way
are
distinct
from
each
other,
the
disjunctions
can
be
in the
constructed
same
as before.
Before moving
Chalmers
physical
(1996).
on Iwill compare this handling of the problems

Chalmers
uses
a more
realizer for an FSA. He assumes
elaborate
specification
above with that in

of
an
arbitrary
that the physical device keeps a record of
to transition
that a system might have the dispositions
has the consequence
to S2 at one time step and (also given suitable input) from S2 to S3 at that
to transition from Si to S2 and then to S3, given those inputs in
disposed
of
But this result is appropriate, as itmay well be that one consequence
is to disable the system with respect to further transitions.
input at the first time-step
As a referee pointed out, this

(given suitable input) from Si
same time-step, without being
series over multiple
time-steps.
receiving
either
4y Springer
arguments
Triviality
its "input history,"
the complete
state.
assumes
also
Chalmers
at
value
particular
281
functionalism
against
the
start
the
a "dial"
has
system
any run.
of
from some initial
of inputs it receives
sequence
a device
Such
can
that
be
the
has
set to
permanently
an
to
enter
capacity
indefinitely large number of distinct inner states, each specifiable in advance in a

way that pairs it with a particular input history and a particular initial state. The
"dial" feature enables us to describe how a given input history could be experienced
from many different initial states. If the system has the right behavioral dispositions
as it traverses all of these possible physical states, indefinitely long disjunctions can
be
for use
constructed
in a different
interval
of
in mappings
about what would
be specified
state,
above.
discussed
a state
including
a more
achieves
method
Chalmers'
the kind
in the
that was,
actual
treatment
straightforward
can
Counterfactuals
if the system had begun
have happened
terminal.
world,
of
the
the time
conditionals
linking Q states, by making richer assumptions about both the internal structure and
behavioral dispositions of the physical realizer. On the behavioral side, Chalmers
responses over
imagines a realizing system that exhibits the right behavioral
indefinitely long series of inputs. I assumed only that a system has the right
behavioral dispositions over an interval which begins with either the actual initial
or
state
one
as
either
the
of
an
table,
set
of
relevant
to
alternative
or
behavioral
I keep
alternatives.
the
argument
simpler
structural
Chalmers'
Chalmers
assumptions
on
procedure
above.
given
I do
But
are
makes
not
the
think
for
necessary
If a
the construction of a triviality argument that raises problems for functionalism.
a
a
recorder of its input history and dial, of the kind Chalmers describes,
system has
we
then
time
can
in advance
know
interval
under
that
the
will
consideration,
system
do so
enter
must
for
both
states
unique
actual
and
the
during
non-actual
input
histories, and would do so from relevantly different initial conditions. But a system
need not have those features in order for it to be true that it is disposed to enter
unique
states
physical
under
all
those
circumstances.
All
that
the extra
features
give
us is the possibility of labeling the states compactly and in advance. As long as a

physical system has such a range of states (and the behavioral assumptions are met),
there is the possibility
in principle of constructing a mapping
that shows it is a
realization of the FSA.
3 Realization
of a CSA
The
argument
preceding
not
does
a threat
constitute
to contemporary
functionalism.
The only position threatened so far is a form of the now-outdated

"machine
functionalism." An FSA is in a single internal state at any time, and it is usually
thought that a viable view in philosophy of mind must recognize that a cognitive
is in more
system
interactions
moods...),
a
than
along
Putnam-style
with
triviality
The
total
inner
state
of
at a time.
input.
Chalmers'
for
Behavior
mental
present
simultaneously
argument
this
1996
case.
discussion
I will
present
is the
states
includes
an
of
consequence
desires,
(beliefs,
a sketch
of
of
the
argument
form.
the
introduces
state
mental
sensory
same kind in a different

Chalmers
one
several
between
"CSA"
a system
formalism
is now
(for
represented
state
combinatorial
as a vector,
or
list,
automaton).
of
substates.
4y Springer
282
P. Godfrey-Smith
have the form: ((Cn,C2i,C3i),

Transitions
Oi). Though
I,)
((Ci2,C2i,C3i),
Chalmers uses this category in an analysis of computation, I take the CSA concept
to be a good description of the kind of structure that the mainstream
functionalist
literature
often
in mind.
has
criterion for realization of a CSA is different from that for an FSA, as

something in the physical system must be mapped to every substate that figures in
the state transitions. An initial account of realization might be given as follows:
The
CSA: A physical system realizes a given CSA during a time interval iff there is
a mapping from states P of the physical system onto substates C of the CSA,
and from inputs and outputs of the physical system onto inputs and outputs of
the
I)
such
CSA,
that:
for
((Ci,C2,..
a combination
of
states
state-transition
((Ci,C2,..
if the physical
system were
every
of the CSA,
.C'n), O)
(Pi,P2-
-Pn)
to
that map
.Cn)
(Ci,C2,..
.Cn),
to be in
this
during
time period, and received input I* that maps to CSA input I, then it would
transition to a combination of substates (P\, P2..
that map respectively to
.P'n)
and would emit an output O* that maps to CSA output O.
(CpC^,..
.C7n),
we
Again
sense
treats
can
between
distinguish
and
inputs
outputs
narrow
and
narrow
The
CSAs.
criterion
CSA
(The
abstractly.
sense
broad
above
uses
the
symbols "I" and "O" in this abstract sense.) A broad sense CSA specifies what the
system's inputs and outputs should actually be. So showing that a physical system
realizes a broad sense CSA involves both showing a mapping between the physical
structure
and
the narrow
sense
and
CSA,
that
showing
the
inputs
and
substates.
1 can
be
Without
system with
means
complex"
a state
step,
that
dependent
contingency
4 Contingency
Fig.
the right input-output

the
on
system
the
is
history
in a unique
of
inputs
properties.
total
substates
different
tree,
as
in Fig.
As
before,
physical
state
"sufficiently
at every
section. We
first represent
4.
tree for a
CSA
time
received.
is similar to that in the previous
The argument
as
of
as
treat substate Q2
to a particular
We
state, 2, that region
might
mapping
in. But
also
the criterion
for realization
above
allows
other mappings.
given
some
a
extra
is
realized
CSA
constraint,
any
by
sufficiently
complex
physical
CSA
are
outputs
the right kind.

The natural-looking way to map physical features of the system to CSA
is to take different regions within
the physical
system to implement
Springer
the
against
functionalism
as we
saw
arguments
Triviality
It is possible,
to represent
If we
earlier,
as
properties
input-output
283
a tree
any
also.
the
superimpose
same
the
with
system
physical
trees
contingency
for
the physical system and the CSA, they will differ only in the internal states at each
node. On the physical tree we have total physical states. On the CSA tree we have
vectors
of
substates.
of the right kind, we represent each physical
To construct a mapping
system
as
grain"
the physical
of
conjunction
of
disjunctions
as
tree,
contingency
that a particular disjunction
of physical
tree
to the
that
exactly
correspond
tree. Vectors
CSA
total
physical
states can occupy
locations
We
occupied
by
state of the
"coarse
also
The
section.
in the previous
is to show
aim
locations
in the physical
particular
substate
to conjunctions
substates correspond
of CSA
states.
in the
of
of disjunctions
states.
physical
The disjunctions are constructed as follows. Consider the first state of the first
substate variable, Cn. We look at all the locations on the CSA tree where the
system is in a total state that includes Cn. Call this set of locations on the CSA tree
En. We then look at the physical tree, and note the total physical state of the system
at all locations in En. These physical states are collected and labeled with a
CSA
Qn.
disjunction,
we
Then
of
construction
begin
coarse-grained
where
tree,
physical
the
physical state of the system is specified at each location with a conjunction of

disjunctions. We make Qn the first member of the conjunctive specification of the
state
system's
the
in
vector.
The
substates.
are
tree
This
preserved.
assumes
initial states are handled

before, it is also possible
system,
as
physical
states which
Chalmers
obvious
of
but
solutions,
assume
framework
not
it does
however,
a
realizing
above,
to functionalism
So
So
II]].
will
Qn
with
CSA
indefinite
to
The
alternate
number
as
unique
in
operate
to give
to require
on
constraint
substates
combine
step,
of
in advance.
add,
CSA
which
system,
as a next
and
in the previous section. As

about the physical realizing
is
earlier.
natural,
other
to disjunctions
states
terminal
an
problem
the
the
states and inputs and outputs on
that
this
all
one-to-one
enter can be specified
mentioned
of
and
C2],
that
to the
rise
the physical
to CSA substates be states of independent components of the

In the final section I will argue that this is indeed the right next
contraint
similarity
was
parts
it seems
independent
system.
physical
CJ2,
be mapped
arising
ensure
to
response
the whole.
states which map
issues
which
that
feature
as
explanations
move,
can
the system might
initial
realization,
behavior
for
repeated
in the same way described

to make richer assumptions
does,
in
in En, and nowhere else. This is the

i appears as the first member of the CSA state
substates
that
argument
The
be
that CSA
shows
location
every
location
states such that all relations between
of physical
the
can
procedure
This
at
tree
coarse-grained
appear as the first conjunct at every

same as the set of locations where Q
the
solve
of
I will
system
derived
entire
one
discuss
with
from
more
particular
Putnam
It needs
problem.
the kind discussed
in the FSA
triviality
input-output
and Chalmers,
of the kind associated with Hinckfuss
to be
argument.
with
looking
at
combined
case. Before
This
one
properties.
a stronger
to make
It
does
not
uses
the
challenge
and Searle.
Springer
284
P. Godfrey-Smith
4 A new
An
argument
triviality
is as follows. Any
initial outline of the argument

can
system
to the "transducer
alone
system
properties.
not
alter
on
drawing
or
whether
earlier
not
sufficiently
an
of
duplicate
at
it has
we
arguments,
complex
physical
a
via
agent,
intelligent
change
the transducer layer of a physical
layer" of that system. Changing
ought
But
a behavioral
into
be made
know
some
least
basic
mental
if a complex
that
physical
system can be given the behavioral profile of an intelligent agent, it is thereby made
to realize the functional
profile of that agent, if realization is understood using a
simple
mapping
system
has
criterion.
mental
functional
features
transducer
change
we
whether
change
to transducer
change
every
have
which
mental
complex
mental
system
properties
alter
whether
has
already
the
properties. A change
to
not
to
we
properties
at all.
not
should
layer
physical
like you and I mental
that give agents
may
layer
As
properties,
but
have,
ought
functionalism
Consequently,
combined with a simple mapping account of realization collapses into triviality.

Every complex physical system has functional properties sufficient to give it some
mental properties of the kind found in paradigm human agents.
I now discuss the premises
in more detail. I begin with the idea that any
characterized
link it with
its
system whose
functionally
physical
operations
environment can be broken down into (what I will call) a transducer layer and a
control
The
system.
a form
that
human
transducer
is
layer
the
between
interface
the
rest
the
of
transducer
the
can
system
include
layer
use
the
in further
Parts
processing.
which
retina,
to
responds
and
system
On the input side, the transducer layer responds to physical
environment.
of
its
impacts in
the normal
electromagnetic
radiation with neural firings, and hair cells in the inner ear, which respond to
physical vibrations with neural firings. On the output side, muscle fibers are part of
the
transducer
as
layer,
they
neuron
to motor
respond
in a system
other
of
manipulation
When
the
transducer
layer.
learning,
representations,
This
planning,
contractions.
that is functionally
includes
important
of memory,
the basis
the
so on.
and
in abstract terms, the role of each part of the transducer layer is
described
one
mapping
than
with
firings
The term control system will be used for everything
physical
variable
to another.
Its role
could
be
represented
a look
with
up table. This simplicity is essential to transducer layers as I conceive them. If we

are looking at a peripheral part of a system, and we find a role for memory or
learning, then we have not looked peripherally enough to find the transducer layer.
The simplest cases to think about are those where the transducer layer remains fixed
in
its
input-output
properties
over
time.
Systems
like
ourselves,
however,
may
be
plastic with respect to these features. In that case the argument should be applied to
a system
over
an
in which
interval
the
transducer
layer
remains
fixed.
it may be unclear where the border between transducer layer and control
system is. But locating the transducer layer is a consequence of locating the divide
between system and environment. If there is a problem with the idea of a boundary
Often
between
functionally
for mainstream
are made
within
characterized
functionalism
itself.
system
The
and
assumptions
its environment,
Imake
about
that
is a problem
transducer
functionalism.
4y Springer
layers
arguments
Triviality
I took functional
Earlier
and
outputs.
any system with

input
in human
possibly
have
of
and
same
the
and
inputs
was
outputs
in
outputs
give
functional
as a human
profile
But we
to functionalism;
have to receive
bucket
agent,
as
form.
such
of inputs
specification
favorable
profile of a human would
the functional
form
to include a "concrete"
profiles
treatment
That
285
functionalism
against
of
water
it does
not
sensory
cannot
have
the
of taking a
look at the possibility
right input-output properties.
while
characterized
and
its
transducer
system
changing
functionally
layer,
keeping
the control system intact. This is done by changing the physical devices
that
interface with external objects. We might alter the hair cells in the ear so they are
not moved by vibrations, but by magnetic fields. We might have muscle fibers
a mouse
moving
a computer
on
now
screen.
transducer
Altering
has
layers
important
for people with sensory and motor disabilities.

therapeutic possibilities
When
the transducer layer of an intelligent system is altered, what are the
for its psychological
consequences
properties? Here I do not mean the changes that
will result to its history of experience. Rather, I imagine looking at a snapshot of the
or
system,
assuming
the
transducer
of
preservation
and asking which
the
some
mental
control
input
over
an
interval,
implied by a change
to
attitudes.
such
include
may
able
to reason
unaffected
But
hypothetically.
the
depend
of
how
also
they may
seems
to
as
having
of its
the
agent.
or
reinforcement,
by
as basic
the
The
layer.
the truth-conditions
learn
be
of
properties
transducer
the world
to
able
the
the
not include
include
being
on
only
properties
will
not
as
features
that
by
may
They
implied, but it is natural to think there
changes
agent
that are unaffected
properties
They
an
of
are
and
propositional
at
features
system,
mental
of
layer.
There may be many psychological

are
structure
formal
states are logically
to its mental
changes
being
states
mental
all.
a commitment
Such
to give
many
is implicit
in therapeutic
better
of
environmental
its mental
interfaces
least
properties?at
to
on
work
we try to equip a disabled person with novel
When
a control
transducer
that we
system
a mental
to have
human
layers.
transducer capacities,
the aim is
to have
take
of
life?independently
the
features of its transducer layer. There is not taken to be a risk that altering the
transducer layer will rob the system of all of its mental properties. A similar
commitment
is implicit in many science fiction stories, especially of the paranoid
style
of
"The
theoretical
Matrix."
This
can
argument
transducer
layer
does
transducer
layers
that
less
control
complex
differences
system
be
given.
is quite
simple.
are in some ways
than
systems
between
and
plants
ours,
humans
could
From
be
Plants
and
to
features
than
on
what
happens
point
These
and
however,
course.
of
single-celled
to ours.
similar
seem
false,
a functionalist
lie.
what
that
have
the
of
processes
the
have
organisms
organisms
it is here
The
a more
But
of view,
much
cognitive
reasoning,
are much more dependent
and learning studied by psychology
decision-making,
control
commitment
also
to be
transduced
when
the
on
system
interacts with the physical world.

This argument should not be pushed too far. It is false to think we could work our
way
into
system
from
its
simple input-output device,

wholesale changes to whether
periphery,
and
argue
that
whenever
we
encounter
that device cannot imply

removing or changing
system has mental states. Every part of an intelligent
4?
Springer
286
P. Godfrey-Smith
viewed
system,
input-output
is needed.
I only
boundary
some
mental
altered
their
to
changes
alters
layer
a more
magnitudes
if the mapping
And
one
new
as
operation,
takes
The
boundary
between
automatically
pushed
"outwards"
an
dissimilar
set of
output.
apparently
change
as well
are
as which
one
no
transducer layer. An
same
The
rest
the
a
with
set
physical
not
table.
some
physical
on
the
can
An
and
stimuli
use.
This
device,
output
to an
them
maps
not
is
layer
is grouping
a
So
equivalent.
to the formal properties
are
magnitudes
as
them
variable.
transducer
and
treating
can
input device might
system
variable,
system
in
layer
of
if we find that a transducer device

and
are
properties.8
constraints
of
the
of
look-up
internal
control
inputs
that
that takes some set of physical
that
an
values
of
range
the
is
at the output
end.
applies
not be a clustered,
natural
need
"many"
layer" can include changes
to transducer
mapping,
They
there
that
is
"non-marginal"
that a change
a transducer
of
parts
ideas
that
properties
functional
is
is just a device
magnitude
total
map
continuous
be
might
represented
some
set of values
of
again,
similarly,
some
the
premise
above.
to one.
to one,
is many
to
them
maps
premise
or many
transducer layer input device

and
noted
in a bonafide
to one,
such
collection,
looking
as
their
simple
needed
qualification
is so simple
The
premise.
devices,
The
of
a
these
layer
mental
system
proportion
contentious
character of these mappings,

map
large
control
application
of
the
The
layer.
whose
having
transducer
non-marginal
transducer
as
described
Further,
with
system
this
input-output
to another.
magnitudes
environment.
like worms
introduce
as
of
be
purposes,
only
to a very
"thin"
changes
and
can
up,
a weak
for my
system
features
transducer
seen
be
But
organisms
I now
close
sufficiently
consider
between
by
excludes
from
character.
"mere
of the
involved.
is certainly suspicious. If a transducer layer input device is mapping

many disparate frequencies of EM radiation to a single rate of neural firing, itmight
be argued that this must be such a complex device that it cannot be regarded as a
This claim
"mere"
can
properties
to
be
show
hence
transducer,
mental
used
that
of
strengthen
functionalism
an
Consider
the
actual
something
intact.
system
existing
can
if these
into
agent,
behaviorism,
A,
be
arguments
triviality
collapses
human
that
But
with
while
changed
assumptions
so that
but
they
something
mental
non-marginal
some
leaving
are
granted,
they
not merely
do
more
troubling.
properties.
If
functionalism
is true, this agent has its mental properties in virtue of its functional
This
functional organization will be labeled S, and I assume it is
organization.
in
the
form
of a CSA. Then we take a complex physical system, B, that has
specified
interactions
with
its environment.
It is "complex"
in the
sense
used
earlier;
at every
instant it is in a different maximal physical state. Following Hinckfuss'

example, B
a
sea
be
bucket
of
isolated
from
its
environment
water,
except in ways
might
large
an agent can control. There will be a possible transducer layer (described below)
that can be added to B that will give it the input-output profile associated with S, the
functional organization that makes A an agent with mental states. Call B with its
modified
transducer
layer BL. But
if BL would
have
the input-output
properties
We
should probably also stipulate, as Susanna Rinard pointed out, that when a transducer layer is
the general kind of interface it has with the control system is preserved. Some transducer layers
changed,
may interface lethally with some control systems.
4Q Springer
arguments
Triviality
associated
of
in this paper,
S, then, by earlier arguments
with
realization
287
functionalism
against
in
the
functional
sense,
an
assuming
simple
also be a
it would
account
mapping
of
realization.
The
actual
transducer
layer
a mere
properties,
either
not
has
not
should
layer
transducer
special
have,
Two
fact.
and transducer
mental
non-marginal
mental
non-marginal
alter
this
in physical make-up
Its
layer.
So
properties.
if
lacks only the right transducer layer to be a functional duplicate
it must
then
transducer
or both
have,
this
if a system
that differ only
systems
both
the bucket of water

of A,
its
have
But
surface.
to
change
not
does
B,
is the water/air
similar
functionally
layer must
of water,
bucket
some
have
already
mental
non-marginal
properties.
The part of this argument that needs to be outlined in detail is the claim that if B
is a complex physical system, there is a possible transducer layer that can be given
to it to yield a system with the input-output profile associated with S. The key to
showing this is to note (or require) that all B's physical outputs, as well as inner
states, are unique. This makes it possible to apply the same techniques used in the
previous
sections.
example,
and
To
how
show
discuss
to
how
the
procedure
a bucket
of
turn
The transducer layer that has to be given

coke
machine
an
includes
device
input
works,
I will
sea water
into
return
to the bucket of water

an output
and
earlier
machine.
it into a
to make
At
device.
to my
a coke
the
input
we
end,
need the device to accept 5c and 10c coins. This is no problem; they can be dropped
into the bucket. We do have to assume a stock of very physically similar 5c and 10c
coins, and a uniform method of dropping. Each coin sends the bucket of water into a
new
unique
effects
of
the
effects
of
ripples
the water
also
and
state,
physical
the
surface
generates
on
in the water
unique
at
air molecules
on
these
output.
the
surface.
are
air molecules
At
unique
are
the outputs
Here,
each
moment,
of
products
the
prior state of the water and the particular impact of a coin. (A distracting feature of
the combination of the Hinckfuss example and the coke machine is the possibility of
tracking the displacement of the water by each coin. For generality, assume this
easy
is not
option
available.)
The coke machine builder would next draw a new kind of contingency
tree for
the bucket with its added input device. In Fig. 5, outputs
are the unique physical
Of
effects
the water
of
coke
generate
machine
does nothing when

a coke
emits
device
with
on
surface
All that then has to be

three desired outputs of
coke), and 03 (emit coke
some others), should map
and physical outputs
0?
the
over
behavior
air molecules.
neighboring
in response
or
the
interval
is build
a coke when
O^ (etc.), emits
it detects
and change
no memory
of
layer
done is collect the Of's that should map onto each of the
the coke machine. Those are Oj (null output), 02 (emit
and 5c change). So physical outputs Op
O^, and O3 (and
to Ob physical outputs
should
map to 02,
Oj, O7 (etc.)
and Og should map to 03. All the designer has to do to
internal
to
O^
(etc.).
This,
It is as
processing.
a transducer
device
that
it detects O4 (etc.), and
is an input-output
again,
if a designer
had enormous
knowledge of the physical dispositions of the bucket of water, and very fine-grained
ways of building input-output devices, but no way of building the memory needed
for
coke
memory.
taken
by
machine.
So
the
designer
uses
the
water's
as
complexity
a natural
The designer builds a suitable input device, notes the exact physical paths
the water
surface
in response
to each
sequence
of
inputs,
and
the
system's
4u Springer
288
P. Godfrey-Smith
tree for a
5 Contingency
a coke machine
input
Fig.
pond with
device
Pis, or h
at appropriate
outputs
physical
which
out
gives
cokes
and
as is covered by the designer's

The
coke
machine
have mental
If a normal
CSA.
is, of
states. But
are
stages
This
change.
physical
this process
human's
can
device
transducer
for
whose
FSA
simple
could be applied
functional
another
continue
as
long
an
interval
knowledge.
a very
course,
to control
used
process
over
organization
some
do
realizations
in principle
interval
not
to any FSA or
is represented
by a CSA, then our designer could build a transducer device that perturbs the bucket
of water in specific ways in response to every possible sequence of inputs that a
human
receive,
might
to appropriate
human
and
another
behaviors.
transducer
So
device
a bucket
of
the water's
that maps
sea water
act
could
as
responses
the
control
that our designer was extraordinarily

system for a humanoid robot, provided
tree and skilled in the building of
about the object's contingency
knowledgeable
and
transducer
devices.
input
(especially) output
The argument of this section can then be summarized as follows.9 It is presented
as
a valid
be
denied.
argument
with
an
so at
conclusion,
unacceptable
least
one
premise
must
1.
For any sufficiently complex system B, there is a possible system that differs
internally from B only in its transducer layer, and that has the input-output
2.
Any sufficiently complex system with the input-output properties of a human

criterion for
agent is a functional duplicate of that agent. (Simple mapping
3.
4.
Functional duplicates share all their mental properties (functionalism).

Two systems that differ only in their transducer layers must either both have, or
properties
of
a human
agent
with
non-marginal
mental
properties.
realization.)
both
lack,
non-marginal
mental
properties.
Therefore
5.
Any
sufficiently
complex
I am indebted to an anonymous
adapted here.
system
referee
has
non-marginal
for suggesting
a simplified
mental
summary
properties.
of the argument,
have
4y Springer
which
arguments
Triviality
Even more briefly:

functional
if human agent A has mental

then
organization,
Agent A
But
layer.
B
and
if BL
differ
BL
any
only
some
mental
to an imagined BL, which

mental
have
would
in
in virtue of his or her
properties
B has
system
complex
identical
is functionally
transducer
because
289
functionalism
against
transducer
B must
states,
too.
properties
a modified
is B with
have
actually
them,
layers.
to be drawn is that the kind of

It might be thought that one conclusion
a position which
functionalism discussed in these sections leads to panpsychism,
an
not
be
But
would
the
be
"ordinary" form of
accepted.
position implied
might
the
functional
As
would
realize
profile of many
complex systems
panpsychism.
different intelligent agents, the position implied would be a doctrine of super
position of multiple divergent minds on the substrate provided by each complex
system.
physical
5 Consequences
for functionalism
in this paper are clearly too weak. Even if

The criteria for realization discussed
the near-complete
into triviality discussed
in Sect. 4 is avoided, by
collapse
some
rejecting
that
are
changes
transducer
concerning
to make
functionalism
premises
needed
the
layers,
viable.
earlier
Otherwise,
results
show
standard
many
functionalist
thought-experiments
involving systems with the behavioral proper
ties associated with mental processing,
but inappropriate internal organization,
to functionalism
too (Block 1981, Braddon-Mitchell
become counter-examples
and Jackson 1996). So in this section I discuss responses.
I emphasize
that
of familiar difficulties
that
although the problems may look like a manifestation
arise from the introduction of disjunctive predicates,
the "fixes" that suggest
themselves
have consequences
for functionalist
in philosophy
and
projects
science. One way to put the point is as follows. The criteria for
cognitive
realization discussed above look weak because of the existential quantifiers; all
that
is
that
required
or
contain
structure,
a
set
given
of
of
"autonomy"
of
high-level
formulation
In the
physical
of
case
but
weakness
of FSAs,
similarity
in relation
in
often
of
such
something
that
that
functionalism
the
alleged
systems.
complex
the criteria for FSA and CSA

look
given
occupy
they
and
realizability,
onto
that map
a way
different
when
realization.
sentence
Ramsey
is used.
what
some
is not
is
matters
of whether
states
physical
related
of multiple
of strengthening
functionalism
is for
is needed
physical
this
seems
states that are disjoined
requirement
What
that
descriptions
the question
revisit
are
states
the message
I first discuss ways

then
But
some
have
system
roles.
because
seeks,
some
substantive
some
measure
to the
description. Many
is an extra
needed
and mapped
causal
between
similarity
of
overall
properties
similarities will
constraint
to each formal
the
the
the members
sets
of
of
each
in a metaphysical
similarity,
of
on
system,
be irrelevant,
total
state. The obvious
under
set.
sense,
lower-level
those involving
4y
or
features
Springer
290
P. Godfrey-Smith
probably
rather
an
than
standard
absolute
be
constraint.
on
point
will
in the
role
causal
the
approach
realizations,
a system
will
be
that a non-arbitrary
that
accept
it is unlikely
tree,
This
system.10
better and worse
between
we
Once
a contingency
of
economy
a gradient distinction
yield
at every
states
no
that have
color)
(e.g.,
will
in unique
absolute
recovered.
Another response that some may have had to the problem in the FSA case is to
question not the status of the disjunctions of physical states themselves (the Qi's),
but the conditionals relating them. The simple mapping criterion did not require that
the physical states be linked by causal relations, as opposed to dependence relations
in a broader
but
sense,
terms.
Familiar
binary
distinction
we
situation
conditionals
states
will
collected
level
of
or
how
examples.
the
but
the
of
the
characterized
by
interpretation
across
low levels of similarity

account of FSA
(more)
unnatural
the
physical
states
The
device
relevant
but
one,
extreme
An
one
is
be
should
are not,
that
to a causal
and
in the mapping.
case-independent
of
description
there
those
below,
resistance
the original
in
similarity
use
for
categories
an overall
and
and
I see
conditions.
in less promising
the
categories.
natural
(more)
from
the
it appear
causal
to questionably
opt to supplement
derives
furnish
above
degrees
make
are
that
coarse-grained
between
naturalness
causation
argued
be responses
So I cautiously
grained
is not
as
Various
into
a distinction
about
connections
facing,
of degree.
of
strengthening
similar to the one above, but expressed

talking
between
are
distinctions
of
ways
is a natural
this
perhaps
this approach as essentially
works.
example
collected
notion
The
coke
where
similarity,
to
relation
once
machine
that
is
coarse
into
of
in
similarity
a machine
of
realization with
realizations,
again,
a
lower
can
again
very
unnatural
realizer of the coke machine FSA would be one which responds to an initial
insertion of 5c or 10c by activating one of two entirely different ensembles of
If the first coin is a 5c, the left half of the machine
is activated and the
machinery.
a
side
If
shuts
down.
the
first
coin
is
the
side
10c, only
right
operates. Then
right
there is nothing in common when "S3" in the FSA is reached by its two different
possible
input
state
one
through
invisible
differences,
microphysical
the
"S3"
reached
10c
contrasts
This
paths.11
through
with
to a macroscopic
two
insertion
of
case
where
causal
the
and
only
between
description,
coins
5c
are
there
state
reached
coin.
Here my treatment contrasts with Chalmers (1996), who accepts that an FSA is
realized by any complex physical system with the right input-output properties. He
treats
more
as
this
account
of
and
a conclusion
realization
less
we
with
natural
realizations
profiles). This gives us a way

the preceding
10
to. Rather
forced
I opt
standard,
(within
for
have
to a yes-or-no
holding
distinction
gradient
that
systems
to functionally
than
a
the
right
between
input-output
the two coke machines
differentiate
paragraph.
I am indebted
MA member
seem
a weak
to Nick
Shea
of an audience
for comments
at a conference
substantially
at Aarhus,
improving
2005,
this part of the argument.
suggested
this example.
4y Springer
in
arguments
Triviality
I now
turn
realizations
to
the
to
state
example,
should
system,
a component
the
the
realization
seems
initially
some
of
of
to
two
map
the
to
states
should
CSA
one
and
of
and
the
Chalmers
C22.
of
"good"
substate
and
Cn
state
of
maps
for
C]2,
the
of
component
independent
as C2i
states
such
feature
each
Substates
of
be
key
cases,
system.
different
the
Here
In good
whole
state
whose
a CSA.
of
to capture.
easier
part
that maps
system
291
functionalism
against
physical
the part
of
a
endorses
(1996)
constraint of this kind. I agree that this is the first step that should be taken, but after
fleshing out this idea I will argue that an appeal to lower-level similarity, as in the
FSA
as well.
is required
sense
case,
The
relevant
some
specifying
of
of other components.
this
pass
test. Given
is a
"independence"
of a state
components
vector
The Qy's used in the CSA

that
the
are
Q^'s
In a good
one.
logical
not
should
realization,
logically
above do not
triviality argument
of Pi's
disjunctions
are
that
state
the
constrain
exclusive
of
each
of Q^'s
other, the instantiation of some combinations
logically
implies the
instantiation of others. For example, if (P] v P2), (Pi v P3), and (Pi v P4) are all Qij's
a system,
for
to CSA
mapping
that occur
substates
in vectors
at different
then the fact that (Pj v P2) and (Pi v P3) are both instantiated
also
instantiated.
"distinct
contrasts
This
and
existences,"
the case
with
the
where
instantiation
of
all
the
of
values
i,
implies that (Pi v P4) is

substate
variables
substate
particular
to
map
value
no
has
logical implications about the instantiation of substate values in different regions.

Then CSA analysis will, as it is intended to, make it possible to capture the idea of a
causal
of distinct
economy
external
internal
as well
factors,
as
interaction
of
the
system
with
influences.
In expressing
the
needed
an
both
constraint,
absolute
standard
and
approxima
tions to it are available. The absolute standard is complete independence of the

states mapped
to each CSA substate. But this standard can be also
physical
as
the
of physical
of CSA
approximated,
logical "entanglement"
analogues
comes
substates
This will
each
to constrain
vector
states
which
arose
that
than wholes.
The
So
the
CSA
of
that part
FSA
are
collected
realization
can
problem
includes
to a distinct
maps
for
there is a CSA with
Suppose
states.
state
in the
problem
rather
solve part of the problem. Even when we have determined
only
variable
The
in degrees.
be
four
the
now
again,
illustrated
with
the
simplest
to each
of
level
possible
each of which
states.
possible
that
it is necessary
system,
to be mapped
at the
together
arises
two substate variables,

only
of
part
Qj.
parts
case.
has two possible

also
Suppose
there
is
physical device with a left half and a right half, which has behavioral properties that
match the CSA. States of the first substate variable are mapped to the left hand side
of
the
realization
system,
states
and
is supposed
of
to come
the
second
from
the
are
fact
mapped
to
that Cj
i, for
the
right.
Constraint
on
is a component
example,
in two state vectors, (Cn,C2i) and (Cn,C22). These two total states of the system
are supposed to have "something in common." The physical state mapped to Cn is
combination and another
supposed to play one role when it is part of a (Cn,C2i)
role when it is part of a (Cn, C22) combination. But if any physical states of the left
hand side of the system can be combined in a disjunction and mapped to a CSA
substate,
then
realization. Any
this
combinatorial
physical
feature
of
the
formalism
exerts
no
constraint
on
states of the left hand side can be regarded as different
4y Springer
292
P. Godfrey-Smith
lower-level ways of being
in a single coarse-grained
state which
ismapped
If something
like CSAs
physical
tod.12
The picture emerging

the kind
formalize
of
mapping
substate
variables
the parts'
Both
of
Once
used
requirement
in mainstream
the
requirements.
the
score
The
system.
be
come
second,
One
is that
other
is that
similar.
physically
in degrees.
arises
question
requirement
will
Q^s
the
functionalism,
to involve not just a
categories
obviously
is on the table,
"entangled"
other
realizing
coarse-grained
of the independent-parts
above,
the
but most
requirements,
similarity
of
parts
into
grouped
two
but
states,
to independent
map
could do the work

example
formal
and
physical
microstates
these
seen
characterization
of such a structure should be understood
then the realization

between
is as follows.
from these arguments
functional
whether
this
as well. At
least in the
on
lower-level
poorly
a test
for
similarity. I do not know whether this will be true of all cases. If it is, the appeal to
lower-level similarity may be able to do all the work.
The
consequences
philosophical
of
these
are
requirements
Both
significant.
are
at
tension with things that many functionalists have liked to say. Functionalists
like to
stress the "autonomy" of high-level
functional description, and this has included
rejecting the idea that the realizers of functional states should be distinct, localized
parts
of
the
It is common
system.
is carried
description
to
states to be localized
psychological
at a "higher
out
that
say
there
to physical
level
of
is no
regions
need
for
the
It is not merely
analysis."
realizers
of
of the brain. Functional

a coarse
but a special kind of analysis that can

grained description of physical machinery,
a
not
at
at
entities
visible
all
lower
level
(Fodor 1974).
posit
The response to CSA triviality arguments above pulls against those ideas. The
basic idea of "multiple realizability" of a given functional organization in different
sorts of materials is not threatened. And spatial localization of the simplest kind is
not necessary for the independence of parts discussed above. A single part of a
system
be
might
scattered?as
physically
the
human
immune
system
is scattered
through the body. But even a scattered object of this kind is visible to lower-level
description. So if an account of realization that requires that CSA substates map to
states of distinct physical parts of the system is the best option, this puts pressure on
familiar habits of functionalist thinking.
It is not
only
philosophers
who
often
assume
a relaxed
standard
for
realization.
issues here connect to a divide seen in psychology and cognitive science. One
style of work in these areas holds that an abstract characterization of psychological
The
processes
can
be
carried
out without
worrying
about
finding
simple
match
between
12
is stronger than I
Chalmers
has argued that the combinatorial
requirement
(personal communication)
to C\ | have to produce the right behavior
here. Instead, each of the physical states mapped
acknowledge
states mapped
to C2i and also C22- This
is a possible
when
with each of the physical
combined
of the conditionals
states, but it is too strong an
linking the coarse-grained
interpretation
physical
at the end of Sect. 2 is again relevant. In the
purposes. Here the discussion
interpretation for functionalist
case of a system that ages or undergoes other kinds of physical development,
this stronger combinatorial
state
requirement would require that the system behave appropriately when one part of it is in a physical
of early stages in life, and the other parts of the system are in physical states characteristic
characteristic
it is older
of late stages in life. This surely is not required for realization of a CSA. Often system when
to realize the same CSA
will be realizing a different CSA altogether, of course, but it is surely possible
while the physical parts of the system develop
through time.
Springer
arguments
Triviality
structure
psychological
293
functionalism
against
and
the physico-chemical
of
organization
the
brain.
Others
think that close attention to brain structure is needed to anchor the explanatory
accounts
posits of cognitive science (Churchland 1989). Some recent philosophical
of
in
explanation
treated
the
et al. 2000),
kind
"mechanistic"
localization
of
as
factors
cell
an
and
biology
neuroscience
important
desideratum
though it has not been made very clear what
localization
requires.
make
arguments
Triviality
I return to the question
Lastly,
like
sciences
causal
functionalism
uses
that
these
of how
a form of
these issues appear within

-
sentences
Ramsey
of the relevant
ones.
urgent
problems
the
have
(Machamer
"Ramsey-Lewis"
to
approach
functionalism (Lewis 1972; Braddon-Mitchell

and Jackson 1996).13 In the Ramsey
Lewis approach, the notion of mapping
does not appear explicitly. A set of
hypotheses, given by folk theory or science, is seen as specifying a set of causal
roles.
sets
Various
one
is not
relation
of
can
objects
of mapping
as
act
of
occupants
two
between
causal
these
but
structures,
the
interlocking descriptions by a collection of objects.

I claimed earlier that the change in formalism does not make
status
of
number
and
can
of
states
inner
is in a given
state,
as before.
raised
word
a system,
of
with
"occupant,"
but
concrete,
machine's
this
sort
What
other
of
its
thing
estate
real
so on,
as
count
then
we
still
of
to
have
to the
a
exist
given
input
problems
a role?
such
suggests
If we
express
framework.
key
a set of
there
triviality
the occupant
connotations,
the
that
receives
system
and
state,
can
actually
required
by
as a
sentence,
Ramsey
workings
the
The
of
a difference
says
merely
that when
is not
question of why a heterogeneous

the
sentence
such
some
it enters
be
The
If a Ramsey
arguments.
triviality
roles.
satisfaction
something
our coke
with
grapple
the
of states is not a legitimate occupant of
disjunction
role.
S3
This
raises
It can
however.
puzzle,
that
appear
the Ramsey
sentence
approach
does not have triviality problems of the kind discussed

in the computationally
oriented literature. (I have often heard this claim made in discussions.)
I offer three
comments.
diagnostic
The
usual
examples
The
examples.
features
special
of
these
of
constraints,
seen
the kind
in this
earlier
used
examples
in
this
mere
constrain
to be
tend
the
entities
specified
dependence.
with
In addition,
ones?have
just
"peripheral"
on functional
characterization.
13
Here
are
not
cases
in the
found
again
I include
cases where
that
can
"thick"
the Ramsey-Lewis
such
as
approach have special
Lewis'
murder
case
mystery
engine, have features built into them that

realize
the
as
verbs,
the
functionally
of
direct
connections
is folk-theoretic
In
roles.
causal
many
the theory
in the
to handle.
additional
section.
tradition,
( 1972) and the example of an automobile

strongly
the case of those
triviality problems?in
examples
(i) The usual examples given to motivate

features that defuse triviality problems
roles
the Ramsey-Lewis
of mind
that the Ramsey-Lewis
is supposed
analysis
the Ramsey-Lewis
can
be
approach
supplemented with
philosophy
However,
Standard
to motivate
given
approach have special features that defuse
these
opposed
examples,
to statements
characterized
to entities
that
and cases where
causal
about
entities?not
are
not
dependent
it is scientific.
4y Springer
294
P. Godfrey-Smith
In Lewis'
1972
murder
case,
mystery
in the
bomb
Here
car
In the
attic."
functional
are
roles
case,
engine
specified
causal
in
roles
and "planted the
is "mixes
role
concrete
of
the
in Uganda"
typical
terms
in
of
components
typical
are "met the victim
the theory given by the detective
causal
fuel
air."
with
are
that
relations
antecedently understood (planted the bomb, mixes fuel with air). And in many
cases, the functionally characterized entity has direct relations with entities that are
not functionally characterized
(met the victim in Uganda; mixes fuel with air.)
features
These
on
constraints
impose
in a way
realization
is
that
to
internal
the
functional specification, as opposed to being imposed by an additional commentary.

Similar results might be obtained by specifying the causal roles in thinner and more
abstract
but
terms,
then
on
restrictions
imposing
have to be distinct physical
occupants
occupants
(ii) The special features of these examples are not applicable

Ramsey-Lewis
analysis is supposed to help us with
in the cases that the
the Ramsey-Lewis
approach is applied to the mind, we
sorts
in
which
these
of thickly specified causal relations
system
then deal with a

are typically not
When
available
use
for
a
with
system
in
relations
structure.
links to entities not dependent

most
between
states
will
we
Instead,
specification.
a rich
internal
and
periphery
have few direct

the
functional
be
in "all
(as
exogenously
etc.).
objects,"
are
on functional
as
treated
abstract
(iii) The Ramsey-Lewis

approach can be supplemented
of the kind discussed earlier
to be
with
confronted
states
The
characterized
and
characterization,
relations.
dependence
with additional
constraints
should not expect the Ramsey-Lewis

approach to automatically avoid triviality
like
but
the
this
can,
problems,
approach
mapping approach, be supplemented with
We
extra
we
constraints.
can
functional
At
had
Both
exclude
the
constraints
gerrymandered
roles be occupied
some
least
extra
these
discussed
discussions
within
in mind.
require
the Ramsey-Lewis
be
may
some
used;
of
kinds
have
may
approach
in particular
Lewis,
section
that
states can vary independently.
by distinct parts whose
constraints
in this
earlier
and
collections,
1994),
(e.g.,
always
have
may
as bona fide parts of the

that the occupants of roles be recognizable
envisaged
system, by a criterion independent of the utility of a particular functional
most
troubled by the arguments
The versions of functionalism
description.
discussed here are those developed specifically to steer a path between the identity
theory
and
This
behavior.
enter
to avoid
behaviorism?seeking
the identity theory while

into
is what
causal
retaining
generates
explanations
the
the
reductionism
the idea that mental

of
idea
while
an
abstract
retaining
and
of
"chauvinism"
states are inner causes of

"functional
in
"autonomy"
state"
that
can
relation
to
the
(Fodor 1981). Steering such a path is harder than has

system's physical make-up
been supposed. This threat is that such views will either collapse into behaviorism,
or
start
to draw
them
is not
on
the
sorts
of physical
features
are
In conclusion,
as
arguments
triviality
as many
functionalists
easy
not
that were
fatal
assume.
to
to be kept
functionalism,
Dealing
with
at arm's
but
the
length.
avoiding
arguments
requires a treatment of localization and the relations between levels that is at odds
with much 20th century functionalist thinking. Further, if functional description is to
4y Springer
arguments
Triviality
be
against
stronger
logically
system
appropriate
295
functionalism
than behavioral
a
realizes
structure
functional
particular
a behaviorally
then whether
description,
should
be
seen
as
matter.
gradient
to David Chalmers, Alan Hayek,

I am grateful
Peter Koellner, William
Acknowledgments
Lycan,
Susanna Rinard, Nick Shea, and a referee for the journal for very helpful comments on earlier drafts. The
at the Australian National University
paper also benefited from audience comments during presentations
and Oxford University.
References
N. (1978). Troubles with functionalism.

In C. W. Savage (Ed.), Perception
and cognition:
Issues
in the foundations
of Minnesota
Press.
of psychology
Minneapolis:
University
(pp. 261-325).
and behaviorism.
Block, N. (1981). Psychologism
Review, 90, 5-43.
Philosophical
states are not. Philosophical
Block, N., & Fodor, J. A. (1972). What psychological
Review, 90, 159-181.
Braddon-Mitchell,
D., & Jackson, F. (1996). The philosophy
of mind and cognition. Oxford: Blackwell.
automaton? Synthese,
Chalmers, D. (1996). Does a rock implement every finite-state
108, 309-333.
Block,
P. S. (1989). Neurophilosophy.
Cambridge, MA: MIT Press.
12, 159-179.
(2002). On effective procedures. Minds and Machines,
J. (1996). What
is computation?
108, 335-359.
Copeland,
Synthese,
to minds,
mind: A philosophical
introduction
Crane, T. (1995). The mechanical
London: Routledge.
representation.
Churchland,
Cleland, C.
J. A.
Fodor,
(1974).
Special
sciences
(or the disunity
of science
as a working
machines
and mental
Synthese,
28,
Research,
58,
of Philosophy,
50,
hypothesis).
97-115.
Fodor, J. A. (1981). Representations.
'intrinsic. Philosophy
and Phenomenological
Langten, R., & Lewis, D. (1998). Defining
333-345.
Lewis,
D. (1972).
249-258.
Psychophysical
and theoretical
identifications.
Australasian
Journal
D. (1994). Reduction
of mind.
In S. Guttenplan
to the philosophy
(Ed.), A companion
of mind
Blackwell:
Oxford.
(pp. 413-431).
Lycan, W. (1981). Form, function, and feel. Journal of Philosophy,
78, 24-50.
Machamer,
P., Craver, C, & Darden, L. (2000). Thinking about mechanisms.
Philosophy
of Science, 67,
1-25.
Lewis,
Piccinini,
G.
(2004).
Functionalism,
computadonalism,
35, 811-833.
and mental
states.
Studies
in the History
and
Philosophy
of Science,
Putnam, H. (1960). Minds
and machines.
in H. Putnam, Mind,
and reality.
Reprinted
language,
Press.
papers
(Vol. 2, pp. 362-385).
Philosophical
Cambridge:
Cambridge University
.
Putnam, H. (1988). Reality and representation.
J. (1990).
Is the brain a digital
and Addresses
Searle,
computer?
Proceedings
of the American
Association,
64, 21-37.
Philosophical
InM. Scheutz (Ed.), Computationatism:
Smith, B. (2002). The foundations of computing.
New directions
(pp. 23-58). Cambridge, MA: MIT Press.
to cognitive
science: The case against belief. Cambridge, MA:
Stich, S. (1983). From folk psychology
MIT Press.
4y Springer

Godfrey-Smith. Triviality Arguments Against Functionalism (Article)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Godfrey-Smith. Triviality Arguments Against Functionalism (Article)

Uploaded by

Copyright:

Available Formats

Triviality Arguments against Functionalism

Author(s): Peter Godfrey-Smith

Stud (2009) 145:273-295

in the earlier arguments. Resisting

that systems with mental

I group in this category a family of arguments with different ambitions. An early

is often described by saying that a bucket of water

of interpreting his wall

sitting in the sun

we know that it can be interpreted as having

that a system has

that functional organization as well.

(1996) gave a formal treatment of Searle's

about the mind. Copeland

into logical behaviorism.

Some conclude that a functional or computational description

it clear that triviality

that it is uncertain how these problems

structures that make

systems and a set of idealizations associated with that project.

to say exactly what computationalism

(in Sect. 5) that while

the threat raised by these problems has been underestimated,

structures. The other, deriving

roles are described

(In the final Section

state at each time. A finite

single total functional

(1972), uses a Ramsey

but the arguments are supposed to work within

one way of formalizing the kind of functional characterization often envisaged in

account of the realization

of this kind of structure, which

as the outputs are associated with

inputs and outputs

and the symbols

"I" and "O" will

be used for inputs and outputs

rounds of input and output, in Fig. 1.

tree for the

inputs (if it can interact with them at all), and itmay

all the influences

(ormay not) emit outputs of the

receive from the rest of the

the system might

Suppose we have a physical object whose input-output dispositions over some

unique state, but also affect the system's physical

system down a path comprising

to later inputs, perhaps in

states that cannot be reached by any other

here is intended to apply to intrinsic properties

state that is specified with

of all the unique physical

states created by the method

tree pictured in Fig. 2.

Then for all i e {1, 2, 3}, M(S?) = Q?

need not suppose

states are identical. But in natural

used in the criterion for realization. The problem has two

simplest way of understanding

the Q states is not

location of physical states in the history of the system.7

tree of the kind above

on Iwill compare this handling of the problems