Professional Documents
Culture Documents
intelligence
What business leaders need to
know about cognitive technologies
Contents
Overview|2
Endnotes|14
1
Demystifying artificial intelligence
Overview
2
What business leaders need to know about cognitive technologies
3
Demystifying artificial intelligence
more diverse problems. Amid disappointment Intel cofounder Gordon Moore, has benefited
with a lack of continued progress, AI fell out of all forms of computing, including the types AI
fashion by the mid-1970s. researchers use. Advanced system designs that
In the early 1980s, Japan launched a might have worked in principle were in prac-
program to develop an advanced computer tice off limits just a few years ago because they
architecture that could advance the field of AI. required computer power that was cost-pro-
Western anxiety about losing ground to Japan hibitive or just didnt exist. Today, the power
contributed to decisions to invest anew in AI. necessary to implement these designs is readily
The 1980s saw the launch of commercial ven- available. A dramatic illustration: The current
dors of AI technology products, some of which generation of microprocessors delivers 4 mil-
had initial public offerings, such as Intellicorp, lion times the performance of the first single-
Symbolics,17 and Teknowledge.18 By the end of chip microprocessor introduced in 1971.20
the 1980s, perhaps half of the Fortune 500 were Big data. Thanks in part to the Internet,
developing or maintaining expert systems, social media, mobile devices, and low-cost
an AI technology that models human exper- sensors, the volume of data in the world is
tise with a knowledge base of facts and rules.19 increasing rapidly.21 Growing understanding of
High hopes for the potential of expert systems the potential value of this data22 has led to the
were eventually tempered as their limitations, development of new techniques for manag-
including a glaring lack of common sense, the ing and analyzing very large data sets.23 Big
difficulty of capturing experts tacit knowledge, data has been a boon to the development of
and the cost and complexity of building and AI. The reason is that some AI techniques use
maintaining large systems, became widely statistical models for reasoning probabilisti-
recognized. AI ran out of steam again. cally about data such as images, text, or speech.
In the 1990s, technical work on AI con- These models can be improved, or trained, by
tinued with a lower profile. Techniques such exposing them to large sets of data, which are
as neural networks and genetic algorithms now more readily available than ever.24
received fresh attention, in part because they The Internet and the cloud. Closely related
avoided some of the limitations of expert to the big data phenomenon, the Internet
systems and partly because new algorithms and cloud computing can be credited with
made them more effective. The design of advances in AI for two reasons. First, they
neural networks is inspired by the structure of make available vast amounts of data and infor-
the brain. Genetic algorithms aim to evolve mation to any Internet-connected comput-
solutions to problems by iteratively generating ing device. This has helped propel work on
candidate solutions, culling the weakest, and AI approaches that require large data sets.25
introducing new solution variants by introduc- Second, they have provided a way for humans
ing random mutations. to collaboratesometimes explicitly and at
other times implicitlyin helping to train AI
Catalysts of progress systems. For example, some researchers have
used cloud-based crowdsourcing services
By the late 2000s, a number of factors
like Mechanical Turk to enlist thousands of
helped renew progress in AI, particularly in
humans to describe digital images, enabling
a few key technologies. We explain the fac-
image classification algorithms to learn from
tors most responsible for the recent progress
these descriptions.26 Googles language transla-
below and then describe those technologies in
tion project analyzes feedback and freely offers
more detail.
contributions from its users to improve the
Moores Law. The relentless increase in
quality of automated translation.27
computing power available at a given price and
New algorithms. An algorithm is a routine
size, sometimes known as Moores Law after
process for solving a program or performing a
4
What business leaders need to know about cognitive technologies
Figure 1. The field of artificial intelligence has produced a number of cognitive technologies
Natural
Computer Machine language
vision learning processing
Speech
recognition Optimization
task. In recent years, new algorithms have been leaders should focus their attention on. Below
developed that dramatically improve the per- we describe some of the most important cogni-
formance of machine learning, an important tive technologiesthose that are seeing wide
technology in its own right and an enabler of adoption, making rapid progress, or receiving
other technologies such as computer vision.28 significant investment.
(These technologies are described below.) The Computer vision refers to the ability of
fact that machine learning algorithms are now computers to identify objects, scenes, and
available on an open-source basis is likely to activities in images. Computer vision technol-
foster further improvements as developers con- ogy uses sequences of imaging-processing
tribute enhancements to each others work.29 operations and other techniques to decompose
the task of analyzing images into manageable
Cognitive technologies pieces. There are techniques for detecting the
We distinguish between the field of AI and edges and textures of objects in an image, for
the technologies that emanate from the field. instance. Classification techniques may be used
The popular press portrays AI as the advent to determine if the features identified in an
of computers as smart asor smarter than image are likely to represent a kind of object
humans. The individual technologies, by con- already known to the system.30
trast, are getting better at performing specific Computer vision has diverse applica-
tasks that only humans used to be able to do. tions, including analyzing medical imaging
We call these cognitive technologies (figure 1), to improve prediction, diagnosis, and treat-
and it is these that business and public sector ment of diseases;31 face recognition, used by
5
Demystifying artificial intelligence
6
What business leaders need to know about cognitive technologies
seems clear, until you encounter the sentence cleaners,48 and a slew of consumer products,
Fruit flies like a banana. Substituting fruit from toys to home helpers.49
for time and banana for arrow changes Speech recognition focuses on auto-
the meaning of the words flies and like.40 matically and accurately transcribing human
Natural language processing, like computer speech. The technology has to contend with
vision, comprises multiple techniques that may some of the same challenges as natural lan-
be used together to achieve its goals. Language guage processing, in addition to the difficulties
models are used to predict the probability of coping with diverse accents, background
distribution of language expressionsthe noise, distinguishing between homophones
likelihood that a given string of characters or (buy and by sound the same), and the
words is a valid part of a language, for instance. need to work at the speed of natural speech.
Feature selection may be used to identify the Speech recognition systems use some of the
elements of a piece of text that may distinguish same techniques as natural language pro-
one kind of text from anothersay a spam cessing systems, plus others such as acoustic
email versus a legitimate one. Classification, models that describe sounds and their prob-
powered by machine learning, would then ability of occurring in a given sequence in a
operate on the extracted features to classify a given language.50 Applications include medical
message as spam or not.41 dictation, hands-free writing, voice control of
Because context is so important for under- computer systems, and telephone customer
standing why time flies and fruit flies are service applications. Dominos Pizza recently
so different, practical applications of natural introduced a mobile app that allows customers
language processing often address relative to use natural speech to order, for instance.51
narrow domains such as analyzing customer As noted, the cognitive technologies above
feedback about a particular product or ser- are making rapid progress and attracting
vice,42 automating discovery in civil litigation significant investment. Other cognitive tech-
or government investigations (e-discovery),43 nologies are relatively mature and can still be
and automating writing of formulaic stories on important components of enterprise software
topics such as corporate earnings or sports.44 systems. These more mature cognitive tech-
Robotics, by integrating cognitive technol- nologies include optimization, which auto-
ogies such as computer vision and automated mates complex decisions and trade-offs about
planning with tiny, high-performance sensors, limited resources;52 planning and scheduling,
actuators, and cleverly designed hardware, which entails devising a sequence of actions
has given rise to a new generation of robots to meet goals and observe constraints;53 and
that can work alongside people and flexibly rules-based systems, the technology underly-
perform many different tasks in unpredictable ing expert systems, which use databases of
environments.45 Examples include unmanned knowledge and rules to automate the process
aerial vehicles,46 cobots that share jobs with of making inferences about information.54
humans on the factory floor,47 robotic vacuum
7
Demystifying artificial intelligence
Cognitive technologies
are already in wide use
8
What business leaders need to know about cognitive technologies
9
Demystifying artificial intelligence
10
What business leaders need to know about cognitive technologies
to apply Watson to a broad range of domains natural language processing and machine
outside of game-playing, from medical diag- learning. These tools use natural language
nostics to research to financial advice to call processing technology to help extract insights
center automation.75 from unstructured text or machine learning
Not all cognitive technologies are seeing to help analysts uncover insights from large
such rapid improvement. Machine transla- datasets. Examples in this category include
tion has progressed, but at a slower pace. One Context Relevant, Palantir Technologies,
benchmark found a 13 percent improvement in and Skytree.
the accuracy of Arabic
to English translations
between 2009 and
2012, for instance.76 Many companies are working to tailor and
Even if these technolo-
gies are imperfect, they package cognitive technologies for a range
can be good enough
to have a big impact of sectors and business functions, making
on the work organiza-
tions do. Professional them easier to buy and easier to deploy.
translators regularly
rely on machine trans-
lation, for instance, to
improve their efficiency, automating routine Cognitive technology components that
translation tasks so they can focus on the can be embedded into applications or business
challenging ones.77 processes to add features or improve effec-
tiveness. Wise.io, for instance, offers a set of
Major investments in modules that aim to improve processes such as
customer support, marketing, and sales with
commercialization machine-learning models that predict which
From 2011 through May 2014, over $2 customers are most likely to churn or which
billion dollars in venture capital funds have sales leads are most likely to convert to cus-
flowed to companies building products and tomers.80 Nuance provides speech recognition
services based on cognitive technologies.78 technology that developers can use to speech-
During this same period, over 100 companies enable mobile applications.81
merged or were acquired, some by technology Point solutions. A sign of the maturation
giants such as Amazon, Apple, IBM, Facebook, of some cognitive technologies is that they are
and Google.79 All of this investment has nur- increasingly embedded in solutions to spe-
tured a diverse landscape of companies that are cific business problems. These solutions are
commercializing cognitive technologies. designed to work better than solutions in their
This is not the place for providing a detailed existing categories and require little expertise
analysis of the vendor landscape. Rather, we in cognitive technologies. Popular application
want to illustrate the diversity of offerings, areas include advertising,82 marketing and sales
since this is an indicator of dynamism that automation,83 and forecasting and planning.84
may help propel and develop the market. The Platforms. Platforms are intended to pro-
following list of cognitive technology vendor vide a foundation for building highly custom-
categories, while neither exhaustive nor mutu- ized business solutions. They may offer a suite
ally exclusive, gives a sense of this. of capabilities including data management,
Data management and analytical tools tools for machine learning, natural language
that employ cognitive technologies such as processing, knowledge representation and
11
Demystifying artificial intelligence
reasoning, and a framework for integrating broaden and adoption to grow. The billions of
these pieces with custom software. Some of investment dollars that have flowed to hun-
the vendors mentioned above can serve as dreds of companies building products based
platforms of sorts. IBM is offering Watson as a on machine learning, natural language pro-
cloud-based platform.85 cessing, computer vision, or robotics suggests
that many new applications are on their way
Emerging applications to market. We also see ample opportunity for
organizations to take advantage of cognitive
If current trends in performance and
technologies to automate business processes
commercialization continue, we can expect
and enhance their products and services.86
the applications of cognitive technologies to
12
What business leaders need to know about cognitive technologies
Read more on cognitive technologies in Cognitive technologies: The real opportunities for
business, published in issue 16 of Deloitte Review in January 2015.
13
Demystifying artificial intelligence
Endnotes
1. CB Insights data, Deloitte analysis. The 10. For a high-level discussion of AI as an
$2 billion figure includes investments exponential technology, undergoing rapid
in companies selling AI technology or progress and with transformative implications,
products with the technology embedded. see Bill Briggs, Tech Trends 2014: Exponen-
2. IBM, IBM Watson, http://www-03. tials, Deloitte University Press, February 2014,
ibm.com/press/us/en/presskit/27297. http://dupress.com/articles/2014-tech-trends-
wss, accessed October 3, 2014. exponentials/, accessed October 9, 2014.
3. Amit Chowdhry, Google to acquire artificial 11. Nils Nilsson, The Quest for Artificial
intelligence company DeepMind, Forbes, Intelligence, (Cambridge: Cambridge
January 27, 2014, http://www.forbes.com/ University Press, 2010), p. 13.
sites/amitchowdhry/2014/01/27/google- 12. Stuart Russell and Peter Norvig, Artifi-
to-acquire-artificial-intelligence-company- cial Intelligence, third edition, (Saddle
deepmind/, accessed October 3, 2014. River: Prentice Hall, 2010), p. 1-5.
4. Josh Constine, NYU Deep Learning Profes- 13. Oxford Dictionaries, Definition of artificial
sor LeCun will head Facebooks new artificial intelligence, http://www.oxforddictionaries.
intelligence lab, TechCrunch, December 9, com/us/definition/american_english/artificial-
2013, http://techcrunch.com/2013/12/09/ intelligence, accessed October 3, 2014.
facebook-artificial-intelligence-lab- 14. A number of research projects with federal
lecun/, accessed October 3, 2014. and private funding are at work on this and
5. Carl Benedikt Frey and Michael A. Osborne, other questions about the brain. They are
The future of employment: How susceptible are preparing for a multi-year effort. See, for
jobs to computerisation?, Oxford Martin School, instance, The White House, BRAIN Initiative
University of Oxford, September 17, 2013. Challenges Researchers to Unlock Mysteries
6. Andrew McAfee and Erik Brynjolfsson, of Human Mind, http://www.whitehouse.
The Second Machine Age (New York: gov/blog/2013/04/02/brain-initiative-
Norton, 2014), http://books.wwnorton. challenges-researchers-unlock-mysteries-
com/books/The-Second-Machine-Age/. human-mind, accessed October 8, 2014.
7. Allen Wastler, Elon Musk, Stephen 15. Douglas R. Hofstadter, Godel, Escher, Bach:
Hawking and fearing the machine, CNBC, An Eternal Golden Braid, (Harmondsworth,
http://www.cnbc.com/id/101774267, Middlesex: Penguin, 1980), p. 597, accessed
accessed October 3, 2014. October 9, 2014 at http://www.physixfan.com/
wp-content/files/GEBen.pdf. Hofstadter called
8. Eliene Augenbraun, Elon Musk: Artificial this Teslers Theorem. Tesler says Hofstadter
intelligence may be more dangerous misquoted him and that what he really said
than nukes, CBS News, http://www. was, Intelligence is whatever machines
cbsnews.com/news/elon-musk-artificial- havent done yet. See Larry Tesler, Adages
intelligence-may-be-more-dangerous- and coinages, http://www.nomodes.com/
than-nukes/, accessed October 3, 2014. Larry_Tesler_Consulting/Adages_and_Coin-
9. Stephen Hawking, Stuart Russell, Max Teg- ages.html, accessed October 9, 2014.
mark, and Frank Wilczek, Stephen Hawking: 16. This historical summary is adapted
Transcendence looks at the implications of arti- from Nick Bostrom, Superintelligence:
ficial intelligencebut are we taking AI seri- Paths, Dangers, Strategies (Oxford:
ously enough?, The Independent, May 1, 2014, Oxford University Press, 2014).
http://www.independent.co.uk/news/science/
stephen-hawking-transcendence-looks-at-the- 17. David E. Sanger, Smart machines get
implications-of-artificial-intelligence--but- smarter, New York Times, December 15,
are-we-taking-ai-seriously-enough-9313474. 1985, http://www.nytimes.com/1985/12/15/
html, accessed October 3, 2014. business/smart-machines-get-smarter.
html, accessed October 5, 2014.
14
What business leaders need to know about cognitive technologies
18. Maura McEnaney, Teknowledge retools 500,000 scholarly papers on the topic of neural
expert systems for business market, networks, for example, published since 2006.
Computerworld, August 4, 1986, p. 76. Geoffrey Hinton is a widely published and
19. Beth Enslow, The payoff from expert cited researcher in this area credited with
systems, Across the Board, January/Febru- several important innovations. See: Geoffrey
ary, 1989, p. 56, citing the estimate of an Hinton, Home Page of Geoffrey Hinton,
analyst in High Technology Business. http://www.cs.toronto.edu/~hinton/, accessed
October 6, 2014. Other researchers who are
20. Andrew Danowitz et al., CPU DB: Record- widely recognized for contributions in this
ing microprocessor history, ACMQueue 10, area include Yann LeCun (See Yann LeCunn,
no. 4 (2014), http://queue.acm.org/detail. Yann LeCuns Home Page, http://yann.
cfm?id=2181798, accessed October 11, 2014. lecun.com/, accessed October 9, 2014), and
21. As Peter Lyman and Hal R. Varian noted in Yoshua Bengio (see Yoshua Bengio, Yoshua
2000, Not only is digital information produc- Bengios Research, http://www.iro.umontreal.
tion the largest in total, it is also the most ca/~bengioy/yoshua_en/research.html, ac-
rapidly growing. See Peter Lyman and Hal R. cessed October 9, 2014). Recently, Microsoft
Varian, How much information, 2000, http:// demonstrated a new machine learning
www2.sims.berkeley.edu/research/projects/ architecture that dramatically accelerates the
how-much-info/, accessed October 10, 2014. machine learning process, improving precision
and accuracy. See, Microsoft Research, On
22. Interest in big data was fueled by examples
Welsh Corgis, computer vision, and the power
of extracting valuable insights from large
of deep learning, http://research.microsoft.
amounts of unlabeled, noisy data. See Alon
com/en-us/news/features/dnnvision-071414.
Halevy, Peter Norvig, and Fernando Pereira,
aspx?0hp=002c, accessed October 6, 2014.
The unreasonable effectiveness of data, IEEE
Intelligent Systems, 24, no. 2 (2009): pp. 812. 29. The Apache Software Foundation sponsors
Apache Mahout, an open source machine
23. See, for instance, Jeff Bertolucci, Hadoop:
learning library. Startup PredictionIO is
From experiment to leading big data platform,
offering an open-source machine learning
InformationWeek, June 24, 2013, http://www.
server and recently received $2.5 million
informationweek.com/software/hadoop-from-
in venture funding. See Steve OHear,
experiment-to-leading-big-data-platform/d/d-
PredictionIO raises $2.5M for its open
id/1110491?, accessed October, 9, 2014.
source machine learning server, TechCrunch,
24. For a discussion of the challenges of extract- July 17, 2014, accessed October 6, 2014.
ing meaningful insight from big data, see
30. Russell and Norvig, Artificial Intelligence.
James Guszcza et al. Too big to ignore,
Deloitte University Press, January 31, 31. C.H. Chen, Computer Vision in Medi-
2013, http://dupress.com/articles/too-big- cal Imaging (Singapore: World Scientific
to-ignore/, accessed October 9, 2014. Publishing Company, 2014).
25. For discussion of cognitive analytics, 32. Justin Mitchell, Making photo tag-
including the role of cloud computing, see ging easier, Facebook, https://www.
Rajeev Ronanki and David Steier, Cogni- facebook.com/notes/facebook/making-
tive analytics, Deloitte University Press, photo-tagging-easier/467145887130, ac-
February 21, 2014, http://dupress.com/ cessed on October 18, 2014.
articles/2014-tech-trends-cognitive- 33. Charlie Savage, Facial scanning is making
analytics/, accessed October 9, 2014. gains in surveillance, New York Times, August
26. Catherine Wah, Crowdsourcing and its ap- 21, 2013, http://www.nytimes.com/2013/08/21/
plications in computer vision, U.C. San Diego, us/facial-scanning-is-making-gains-in-
May 26, 2011, http://vision.ucsd.edu/~cwah/ surveillance.html, accessed October 8, 2014.
files/re_cwah.pdf, accessed October 8, 2014. 34. Dawn Chmielewski, Amazons Fire smart-
27. Google Inc., Google Translate Community phone uses Firefly image recognition,
FAQ, https://docs.google.com/document/d/1 Recode, http://recode.net/2014/06/18/
dwS4CZzgZwmvoB9pAx4A6Yytmv7itk_XE- amazons-fire-smartphone-uses-firefly-image-
968RMiqpMY/pub, accessed October 8, 2014. recognition/, accessed October 3, 2014.
28. Multiple researchers have devised algorithms 35. Mark Graves and Bruce G. Batchelor,
that have improved the performance of Machine Vision for the Inspection of Natural
machine learning. Google Scholar finds some Products (London: Springer, 2003), p. 8.
15
Demystifying artificial intelligence
36. CB Insights data, Deloitte analysis. 49. Erico Guizzo, Friendly robot is a hands-
37. For instance, Microsoft recently announced free home helper, Discovery, July 17, 2014,
that it had developed a computer vision http://news.discovery.com/tech/robotics/
system able to identify dog breeds. It relies friendly-robot-is-a-hands-free-home-
in part on machine learning techniques and helper-140717.htm, accessed October 9, 2014.
was trained using a database of millions of 50. Russell and Norvig, Artificial In-
images. See Microsoft Research, On Welsh telligence, pp. 912919.
Corgis, computer vision, and the power of 51. Bruce Horovitz, Dominos app lets you
deep learning, http://research.microsoft. voice-order pizza, USA Today, June 17,
com/en-us/news/features/dnnvision-071414. 2014, http://www.usatoday.com/story/
aspx?0hp=002c, accessed October 6, 2014. money/business/2014/06/16/dominos-voice-
38. CB Insights data, Deloitte analysis. ordering-app-nuance-fast-food-restau-
39. Chowdhry, Google to acquire artificial rants/10626419/, accessed October 3, 2014.
intelligence company DeepMind. 52. IBM, Decision optimization, http://www-03.
40. For an early use of this example see Gilbert ibm.com/software/products/en/category/deci-
Burck, The Computer Age and its Potential sion-optimization, accessed October 13, 2014.
for Management, (New York: Harper & Row, 53. Russell and Norvig, Artificial In-
1965), p. 62, accessed at https://archive.org/ telligence, pp. 366431.
details/TheComputerAgeAndItsPotential- 54. Microsoft, Best practices for rule-based
ForManagement on October 9, 2014. application development, http://msdn.
41. Russell and Norvig, Artificial In- microsoft.com/en-us/library/aa480020.
telligence, pp. 860885. aspx, accessed October 13, 2014.
42. See, for instance, Clarabaridge, Inc., 55. Matthew Finnegan, Barclays to offer voice
http://clarabridge.com/, and Luminoso recognition for telephone banking, Computer-
Technologies, Inc., http://www.luminoso. worldUK, June 23, 2014, http://www.comput-
com/, accessed October 13, 2014. erworlduk.com/news/applications/3526401/
43. John Markoff, Armies of expensive law- barclays-offer-voice-recognition-for-telephone-
yers, replaced by cheaper software, New banking/, accessed October 12, 2014.
York Times, March 4, 2011, http://www. 56. HIMMS Analytics, Essentials of the U.S.
nytimes.com/2011/03/05/science/05legal. hospital IT market: Support services, 2013,
html, accessed on October 9, 2014. http://assets.fiercemarkets.com/public/
44. Jason Belzer, Automated insights poised newsletter/fiercehealthit/himssanalyt-
to revolutionize sports media, Forbes, ics10-17.pdf, accessed October 3, 2014.
http://www.forbes.com/sites/jasonbel- 57. VuComp, M-Vu CAD for mammography,
zer/2013/02/26/automated-insights-poised- http://www.vucomp.com/products/m-
to-revolutionize-sports-media/, February vu-system, accessed October 3, 2014.
26, 2013, accessed on October 9, 2014. 58. IBM, Memorial Sloan-Kettering Cancer
45. For a discussion of robotics as an exponential Center, IBM to collaborate in applying Watson
technology, see Briggs, Tech Trends 2014. technology to help oncologists, https://
46. Dominic Rushe, Google reveals home www-03.ibm.com/press/us/en/pressre-
delivery drone program Project Wing, lease/37235.wss, accessed October 3, 2014.
Guardian, August 29, 2014, http://www. 59. Leonardo Rodrigues et al., Berg Inter-
theguardian.com/technology/2014/aug/29/ rogative Biology Informatics Suite:
google-joins-amazon-in-testing-home- Data driven integration of multi-omic
delivery-drones accessed October 9, 2014. technologies using Bayesian AI, Cancer
47. See, for instance, the Baxter Robot by ReThink Research, 73, no. 8, supplement 1 (2013),
Robotics, http://www.rethinkrobotics.com/. DOI: 10.1158/1538-7445.AM2013-5230.
48. See, for instance, iRobots Roomba, 60. George E. Dahl, Navdeep Jaitly, and
http://www.irobot.com/For-the-Home/ Ruslan Salakhutdinov, Multi-task
Vacuum-Cleaning/Roomba. neural networks for QSAR predictions,
arXiv:1406.1231v1 [stat.ML] (2014).
16
What business leaders need to know about cognitive technologies
61. Jenni Whalen, Revolutionizing the 71. Facebook, DeepFace: Closing the gap
healthcare system, Boston Magazine, to human-level performance in face
July 31, 2013, http://www.boston- verification, https://www.facebook.
magazine.com/health/blog/2013/07/31/ com/publications/546316888800776/,
revolutionizing-the-healthcare-system- accessed October 3, 2014.
bergpharma/, accessed October 3, 2014. 72. David Ferrucci et al., The AI behind Wat-
62. Ravi Somaiya, The A.P. plans to automate sonthe technical article, AI Magazine,
quarterly earnings articles, New York 2010, http://www.aaai.org/Magazine/Watson/
Times, June 30, 2014, http://www.nytimes. watson.php, accessed October 3, 2014.
com/2014/07/01/business/media/the-ap-plans- 73. IBM, IBM Watson, http://www-03.
for-computers-to-write-corporate-earnings- ibm.com/press/us/en/presskit/27297.
news.html?_r=0, accessed October 3, 2014. wss, accessed October 3, 2014.
63. Csar E. Bravo et al., State of the art 74. Total US search query volume was 17.7 billion
of artificial intelligence and predictive in February 2014. 67.5 percent of search
analytics in the E&P industry: A technology queries are conducted on Google. See Search
survey, SPE Journal 19, no. 04 (2014): pp. Engine Land, Google search share stable,
547563, http://dx.doi.org/10.2118/150314- Bing growth continues at Yahoos expense,
PA, accessed October 12, 2014. http://searchengineland.com/google-search-
64. Karen Boman, Artificial intelligence software share-stable-bing-continues-cannibalize-
aids decision-making in onshore drilling, Rig- yahoo-187124, accessed October 3, 2014.
zone, July 10, 2014, http://www.rigzone.com/ The volume of mobile search is on track to
news/oil_gas/a/133973/Artificial_Intelligence_ surpass the volume of desktop search. See Greg
Software_Aids_DecisionMaking_in_Onshore_ Sterling, Matt Cutts: Google mobile queries
Drilling/?all=HG2, accessed October 12, 2014. may surpass PC search this year, Search Engine
65. Information Week, Georgia solves campaign Land, March 17, 2014, http://searchengineland.
finance data challenge via OCR, http:// com/matt-cutts-mobile-queries-may-surpass-
www.informationweek.com/government/ pc-year-186816, accessed October 13, 2014.
cloud-computing/georgia-solves-campaign- The share of mobile searches performed by
finance-data-challenge-via-ocr/d/d- voice ranges from 1 to 4 percent. See
id/1204471, accessed October 3, 2014. Google and Nielsen, Mobile search mo-
ments: Understanding how mobile drives
66. See, for instance, RichRelevance, conversions, http://ssl.gstatic.com/think/
Product recommendations and per- docs/creating-moments-that-matter_research-
sonalizationRichRecs, http://www. studies.pdf, accessed October 13, 2014.
richrelevance.com/solutions/merchandis-
ing/, accessed October 12, 2014. 75. Joab Jackson, IBM expands Watsons
presence across multiple industries,
67. iRobot, iRobot Roomba vacuum clean- PCWorld, October 7, 2014, http://www.
ing robot, http://www.irobot.com/ pcworld.com/article/2712392/ibm-expands-
For-the-Home/Vacuum-Cleaning/ watsons-presence-across-multiple-industries.
Roomba, accessed October 12, 2014. html, accessed October 8, 2014.
68. Nest Labs, Nest Labs introduces the worlds 76. NIST Information Technology Laboratory,
first learning thermostat, https://nest.com/ OpenMT12 Evaluation Results, August 28,
press/nest-labs-introduces-worlds-first-learn- 2012, http://www.nist.gov/itl/iad/mig/open-
ing-thermostat/, accessed October 12, 2014. mt12results.cfm, accessed October 8, 2014.
69. Apple Insider, Tests find Apples Siri BBNs system, BBN_ara2eng_primary_cn,
improving, but Google Now voice search performed better than all competitors in
slightly better, http://appleinsider.com/ both years but improved just 13 percent.
articles/14/07/22/tests-find-apples-siri- 77. Martin Williams, Tech is removing language
improving-but-google-now-voice-search- barriersbut will jobs be lost in translation?,
slightly-better, accessed October 3, 2014. Guardian, September 19, 2014, http://www.
70. Olga Russakovsky et al., ImageNet Large theguardian.com/education/2014/sep/19/
Scale Visual Recognition Challenge, tech-removing-language-barriers-jobs-lost-
arXiv:1409.0575v1 [cs.CV] (September 1, translation, accessed October 3, 2014.
2014), http://arxiv.org/pdf/1409.0575v1. 78. CB Insights, Deloitte analysis.
pdf, accessed October 3, 2014.
79. Ibid.
17
Demystifying artificial intelligence
82. Rocket Fuel, Inc., About Rocket Fuel, 85. IBM, IBM Watson: Build with Watson,
http://rocketfuel.com/about-rocket- http://www.ibm.com/smarterplanet/
fuel#what_we_do, accessed October 10, 2014. us/en/ibmwatson/build-with-watson.
html, accessed October 10, 2014.
83. Conversica LLC, How Conversica Works,
https://www.conversica.com/persona- 86. We will explore this subject in greater detail
how-it-works-email-inbound-marketing- in a forthcoming issue of Deloitte Review.
ai/, accessed October 10, 2014.
18
What business leaders need to know about cognitive technologies
Craig Muraskin
Craig Muraskin is managing director of the Innovation group in Deloitte LLP. He works with lead-
ership to set the groups agenda and overall innovation strategy, and counsels Deloittes businesses
on their innovation efforts.
Ragu Gurumurthy
Ragu Gurumurthy is national managing principal of the Innovation group in Deloitte LLP, guiding
overall innovation efforts across all Deloittes business units. He advises clients in the technology
and telecommunications sectors on a wide range of topics including innovation, growth, and new
business models.
19
Demystifying artificial intelligence
Acknowledgements
The authors would like to acknowledge the contributions of Mark Cotteleer of Deloitte Services
LP; Plamen Petrov, Rajeev Ronanki, and David Steier of Deloitte Consulting LLP; and Shankar
Lakshman, Laveen Jethani, and Divya Ravichandran of Deloitte Support Services India Pvt Ltd.
Contacts
David Schatsky
Senior Manager
Deloitte LLP
+1 646 582 5209
dschatsky@deloitte.com
Craig Muraskin
Director
Deloitte LLP
+1 212 492 3848
cmuraskin@deloitte.com
20
Follow @DU_Press
Sign up for Deloitte University Press updates at DUPress.com.
About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee (DTTL), its network of
member firms, and their related entities. DTTL and each of its member firms are legally separate and independent entities. DTTL (also
referred to as Deloitte Global) does not provide services to clients. Please see www.deloitte.com/about for a more detailed description
of DTTL and its member firms.
Deloitte provides audit, tax, consulting, and financial advisory services to public and private clients spanning multiple industries. With
a globally connected network of member firms in more than 150 countries and territories, Deloitte brings world-class capabilities and
high-quality service to clients, delivering the insights they need to address their most complex business challenges. Deloittes more than
200,000 professionals are committed to becoming the standard of excellence.
2014. For information, contact Deloitte Touche Tohmatsu Limited.