See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3282235

Induction: Processes of Inference, Learning, and Discovery

Article  in  IEEE Expert · October 1987

DOI: 10.1109/MEX.1987.4307100 · Source: IEEE Xplore

CITATIONS READS
1,254 1,688

5 authors, including:

Keith J Holyoak Richard Nisbett

University of California, Los Angeles University of Michigan
280 PUBLICATIONS   26,933 CITATIONS    176 PUBLICATIONS   46,303 CITATIONS   

SEE PROFILE SEE PROFILE

Paul Thagard
University of Waterloo
169 PUBLICATIONS   10,114 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

philosophy of science View project

Exploring representation of relations through relational similarity View project

All content following this page was uploaded by Keith J Holyoak on 18 August 2014.

The user has requested enhancement of the downloaded file.

Randy H. Katz, Computer Sciences Division,
UC Berkeley, Evans Hall, Berkeley, CA
94720.
4 IEEE International Conference on
Computer-Aided Design, November
9-12, Santa Clara, California. Contact
Book Reviews
ICCAD-87 Secretary, Mentor Graphics
4$ Workshop on Computer Architecture Corp., 1940 Zanker Rd., San Jose, CA 95112;
for Pattern and Machine Intelligence,
October 5-8, Seattle, Washington. Contact
(408) 436-1500. Induction: Processes of
Steve Tanimoto, Dept. of Computer Science,
FR-35, University of Washington, Seattle, WA Inference, Learning,
98195; (206) 543-1695. December 1987
4 Eighth Real-Time Systems Symposium,
and Discovery
Compsac 87, October 5-9, Tokyo, December 1-3, San Francisco, Califor-
Japan. Contact Tosiyasu L. Kunii, Busi- nia. Contact Kang G. Shin, Dept. of EE and John H. Holland, Keith J. Holyoak,
ness Center for Academic Societies Japan,
Yamazaki Bldg. 4F, 2-40-14, Hongo, Bunkyo- Computer Science, University of Michigan, Richard E. Nisbett, and Paul R.
ku, Tokyo 113, Japan; phone 81 (3) 817-5831-or Ann Arbor, MI 48109-1109. Thagard (MIT Press; Cambridge, Mass.,
Stephen S. Yau, Northwestern Univer-sity, 1986, 416 pp., $24.95, hardcover)
Dept. of Electrical Engineering and Com-
puter Science, Evanston, IL 60201; (312)
491-3641.
January 1988 In "The Current State of Al: One
Man's Opinion" (AI Magazine win-
4$ Second International Conference on ter/spring 1983), Roger Schank offered
AAAIC 87-Aerospace Applications of Arti- Computer Workstations, January "a definition of Al that will disqualify
ficial Intelligence Conference, October 5-9, 31-February 3, Santa Clara, California. Con-
Dayton. Ohio. Contact Michael Johnston, most of its practitioners. . . Al is the
tact Patrick Mantey, 335A Applied Science science of endowing programs with the
BDM Corp., 1900 Founders Dr., Kettering, Bldg., Dept. of Computer Science, University
OH 45420; (513) 259-4434. of California, Santa Cruz, CA 95064. ability to change themselves for the bet-
ter as a result of their own
IEEE International Workshop on Al experiences. "
4$ Applications to CAD Systems for Elec- People exhibiting such behavior are
tronics, October 8-10, Munich, West Ger-
many. Contact Helmuth Benesch, Siemens
February 1988 said to be engaged in induction,
manifested by activities such as induc-
AG, Otto-Hahn-Ring 6, 8000 Munich 83, 4$ IEEE Compcon Spring 88, February tive inference, learning, and discovery.
West Germany; telephone (89) 636-46666. 7 29-March 3, San Francisco, California. How people exhibit such behavior has
Contact Compcon Spring 88, 1730 Mas- greatly concerned both psychologists
Second Knowledge Acquisition for sachusetts Ave. NW, Washington, DC
Knowledge-Based Systems Workshop, 20036-1903; (202) 371-0101.
and philosophers. When addressing
October 19-23, Banff, Canada. Contact John their concerns, more questions than
Boose, Advanced Technology Center, Boeing answers arise. Neat and plain solutions
Computer Services, PO Box 24346, Seattle, (to paraphrase H.L. Mencken) have
WA 98124-or Brian Gaines, Dept. of Com-
puter Science, University of Calgary, 3500
March 1988 quickly revealed their inadequacies.
Schank's manifesto can be viewed as a
University Dr. NW, Calgary, Alta., Canada 4 Fourth Artificial Intelligence Applica- call to arms, rallying Al forces to face
T2N IN4. tions Conference, March 14-18, San this imposing problem.
Diego, California. Contact Al Conference, It's not as if Al has ignored this
4 Third Annual Expert Systems in Computer Society, 1730 Massachusetts Ave.
Government Conference (AIAA), NW, Washington, DC 20036-1903; (202) topic. Marvin Minsky's bibliography
October 19-23, Washington, DC. Contact 371-0101. for Computers and Thought devotes a
Peter Bonasso, Mitre Washington Al Center, whole category to "inductive inference
7725 Colshire Blvd., MS W952, McLean, VA machines," replete with several dozen
22102; (703) 883-6908. citations. More recently, induction has
4$ FJCC 87, Fall Joint Computer Confer-
April 1988 become a major concern in machine
ence, October 25-29, Dallas, Texas.
learning. Unfortunately, these concerns
4$ Compeuro 88, April 11-15, Brussels, have not yielded impressive fruit. In the
Contact Computer Society of the IEEE, Belgium. Contact Jacques Tiberghien,
FJCC 87, 1730 Massachusetts Ave. NW, VRIJE Universiteit Brussels, Pleinlaan 2,
following excerpt, Holland et al.
Washington, DC 20036-1903; (202) 371-0101. 1050 Brussels, Belgium. explain why insights into induction have
been so disappointing:
. . . most treatments of the topic
have looked at purely syntactic
May 1988 aspects of induction, considering
November 1987 4 NCC 88-National Computer Confer- only the formal structure of the
ence, May 31-June 3, Los Angeles, knowledge to be expanded and
11th Symposium on Computer Applications California. Contact AFIPS, 1899 Preston leaving the pragmatic aspects,
in Medical Care, November 1-4, Washington, White Dr., Reston, VA 22091; (703) 620-8900. those concerned with goals and
DC. Contact SCAMC Secretariat, George problem-solving contexts, to look
Washington University Medical Center, out for themselves. In our view,
Office of Continuing Medical Education, this stance has produced little
2300 K St. NW, Washington, DC 20037. June 1988 insight into the way humans do,
Third Annual Conference on Artificial Intel- Third Conference on Man-Machine Systems or efficient machines might,
ligence for Space Applications, November (IFAC, IFIP, IEA, IFORS), June 14-16, make just the inferences that are
2-3, Huntsville, Alabama. Contact Thomas S. 1988, Oulu, Finland. Contact Third MMS, most useful. This is not to say
Dollman, NASA/EB4, MSFC, AL 35812; The Finnish Society of Automatic Control, that syntactic considerations are
(205) 544-3823. PO Box 165, SF-00101 Helsinki, Finland. irrelevant; indeed, at some level

92 IEEE EXPERT
 they are inescapable in any com- environment combined with the sion of how the framework may accom-
putational system. Our claim is generation of new rules. New modate history recording the
simply that pragmatic considera- rules are generated via triggering development of the wave theory of
tions are equally inescapable. conditions, most of which are sound.
best understood as responses to One could enumerate at great length
This observation serves as overture to the success or failure of current the many contributions of this book.
their own study of induction. model-based predictions. Most important, however, Holland et
As the authors carefully emphasize, al. directly and honestly confront a fun-
this book describes a computational The authors represent three depart- damental question-how experience
framework for implementing induction. ments at the University of Michigan- contributes to intelligence. Of almost
We should not confuse such a frame- computer science, psychology, and equal importance, this book presents an
work with an actual implementation. philosophy-each of which has helped object lesson of how computer science's
Rather, we should consider it a feasibil- forge their framework. Each contribu- "mechanistic" materials can profitably
ity study indicating issues that must ulti- tion was subject to review and interpre- be brought to bear on psychological and
mately be addressed and suggesting tation by practitioners from the other philosophical issues concerning great
potentially valuable exploration paths. disciplines, resulting in a gestalt whole mysteries of the mind. Such a mechanis-
We should not dismiss these suggestions that surpasses the sum of its parts. Let's tic approach is of particular interest in
as idle speculation, however; each is consider the influence of each dis- the attempt this book makes to harness
reinforced with behavioral evidence cipline. intuitively appealing notions regarding
recorded by experimental psychologists. Computer science contributed two mental models for the ultimate imple-
Thus, they can be considered desiderata technologies, one well established and mentation of intelligent machines.
of the following form: When humans one emerging-the former, the produc- Holland et al. provide some excellent
exhibit induction, they also exhibit the tion system; the latter, the body of con- counters to points raised by Hubert and
following behavior; therefore, nectionist models. The production rule Stuart Dreyfus in Mind Over Machine
implementations of induction should is fundamental to the framework. The (reviewed for IEEE Expert by Lotfi
exhibit the same behavior. major criticism against rule-based Zadeh, summer 1987, p.1 I0- 11 1), a sec-
Induction is not a traditional architectures resides in the attention tion of which appeared in our summer
scientific-method exercise culminating such architectures devote to deciding 1986 issue. A fundamental element of
in the validation of theory; the authors which rule to fire, as opposed to using the Dreyfus argument, excerpted below,
will not even go so far as to formulate a rules as knowledge representations. For concerns thought's holistic nature:
scientific theory. Their modesty and this reason, the framework introduces
humility is revealed in the final chap- limited parallelism with competition Experimental psychologists have
ter's epilogue- "Toward a Theory of and support among rules. Such compe- shown that people actually use
Induction." They gather essentials from tition and support exists in connec- images, not descriptions as com-
the preceding 11 chapters, suggest tionist models, but this framework puters do, to understand and
where the evidence is heading, and hope eschews the more massive parallelism of respond to some situations.
that further interdisciplinary research connectionism-a decision the authors Humans often think by forming
(clearly emphasizing "interdiscipli- base on evidence from experimental images and comparing them
nary") will lead to more concrete psychology. holistically. This process is quite
results. Results of such experiments consti- different from the logical, step-
Such modesty should not turn tute psychology's major contribution- by-step operations that logic
prospective readers away from this experiments covering many instances of machines perform.
book. Ultimately, more food for human behavior, as well as some of
thought exists in the desiderata inves- lower animals: conditioning, category
tigated than in the "syntactic" accounts formation, modeling the physical and While this book cites no Dreyfus
of induction decried. Furthermore, social worlds, understanding of varia- literature, Holland et al. are aware of
principles behind the authors' frame- bility, inferencing based on statistics these arguments (they base their discus-
work are not particularly outrageous, and deduction, and the use of analogy. sion on the writings of Fodor and
they simply reflect basic "pragmatic In addition, the concept of a "mental Quine). The preceding points are
considerations" the authors feel have model" (recently attracting considera- addressed and demonstrated as being
been neglected: ble attention in cognitive science) is either inaccurate or irrelevant criticisms
basic to the framework. Indeed, the of the proposed framework.
We view cognitive systems as framework can be regarded as propos- Simply put, this book effectively
constantly modeling their envi- ing an architecture for mental models. demonstrates what happens when sepa-
ronments, with emphasis on local Philosophy's contribution is the con- rate disciplines support rather than
aspects that represent obstacles to cern for basic epistemological issues, attack each other.
the achievement of current goals. particularly addressing the nature of
Models are best understood as knowledge and explanation. The
assemblages of synchronic and emphasis is on the philosophy of
diachronic rules organized into science, reviewing the framework in
default hierarchies and clustered light of historical evidence detailing the
into categories. The rules com- formation of scientific theories-
prising the model act in accord particularly fascinating since it demon-
with a principle of limited paral- strates that philosophy, like psychol-
lelism, both competing with and ogy, provides experimental evidence -Stephen W. Smoliar
supporting one another. Goal vital to the resulting framework's USC Information Sciences Institute
attainment often depends in part strength. In fact, this examination of 4676Admiralty Way, Ste. 1001
on flexible recategorization of the scientific theory culminates in a discus- Marina del Rey, CA 90292

FALL 1987 93

View publication stats