AI Magazine Volume 19 Number 1 (1998) (© AAAI) Book Review

perspectives on research. Logic, rules, Mind: Introduction to and concepts (meaning frames, semantic networks, and so on) are widely used subdivisions. It is ar- guable whether the section is at the same level of generality as the A Review previous chapters, but there is a suffi- cient quantity of work in this area to constitute a separate chapter. The Bonnie Holte Bennett, Dwight Nelson, Russell Pannier, images chapter discusses visual images and processing. This discussion is nec- Thomas Sullivan, and Gregory Robinson-Riegler essarily at a high level and from the expected cognitive science perspec- tive. Thus, it would be different from what an image-processing researcher might expect. However, the discussion nderstanding the mind is one literature and theories put forth to is appropriate given the goal of the of the great “holy grails” of date. Thagard summarizes results of book and the perspective of the Utwentieth-century research. research to date: “the central hypoth- intended (general) audience. The con- Regardless of training, most people esis of cognitive science: Thinking can nections chapter introduces connec- who come in contact with the field of best be understood in terms of repre- tionist and parallel distributed pro- AI are at least partially motivated by sentational structures in the mind and cessing research in a general way. the glimmer of hope that they will get computational procedures that oper- Reading this text as a teacher of a better understanding of the mind. ate on those structures” (p. 10). He cognitive psychology, you can spot This quest, of course, is a rich and uses a shorthand notation for this one of the book’s strengths. Our cog- complex one. It is easy to get mired in approach: the computational-repre- nitive psychologist writes, “Previous- minutiae along the way, be they the sentational understanding of mind ly, I had been hesitant about covering optimization of an algorithm, the (CRUM). One wonders at this point cognitive science at any length in my details of a mental model, or the intri- just how pessimistic his view will be of cognitive psychology class, much less cacies of a logical argument. Thagard’s research to date. Is CRUM just a nice, offering an entire course in cognitive book attempts to call us back to the pronounceable acronym, or does he science. I had yet to see an overview larger picture and to draw in new intend to include all the overtones that was truly accessible to novices in devotees—and, in general, he suc- the field.” Cognitive science textbooks ceeds. tend to lose the reader in a morass of This book begins, “Cognitive sci- , Mind: labyrinthine detail. Thagard’s style is ence is the interdisciplinary study of clear and readable but still conveys mind and intelligence...” (p. ix); so, Introduction to Cogni- the complexity and breadth of the we assembled a cross-disciplinary tive Science, Cambridge, issues involved. The organization is review team that included researchers appealing; it is arranged around the from the fields of AI, cognitive sci- Mass.: The MIT Press, type of (for ence, neuroscience, and . 199,. 212 pp., $22.50, example, logic, rules, concepts, im- This multidisciplinary approach ages) instead of disciplinary line. This seemed appropriate because this book ISBN 0-262-20106-2. organization helps to make the text attempts to be a bridge book, written more accessible to everyone, regard- for a wide audience covering these less of his/her particular area of inter- areas and more. est. associated with the homonymous The book is divided into two major One aspect of the literature review crumb (that is, crummy, crumbly, sections: (1) Approaches to Cognitive that Thagard adds is an analysis of all Science and (2) Challenges to Cogni- crumbling)? Section 2 shows that it is the approaches (logic, rules, concepts, tive Science. The first section is a sur- somewhere in between. analogies, images, and connections) vey of major trends in the research. The chapters in this section are in terms of their representational The second section delineates and “Representation and Computation,” method, problem-solving capabilities, analyzes open problems and research “Logic,” “Rules,” “Concepts,” “Analo- learning approaches, and the types of issues. gies,” “Images,” “Connections,” and language that have developed to facil- “Review and Evaluation.” From an AI itate these tasks. Thus, you have a neat Approaches to perspective, these chapters adequately and useful framework to understand a Cognitive Science cover the research. Representation variety of aspects of these approaches. and computation are generally recog- In summary, this discussion covers This section is a broad review of the nized as the two major divisions or the field and does so in a way that,

Copyright © 1998, American Association for Artificial Intelligence. All rights reserved. 0738-4602-1998 / $2.00 SPRING 1998 133 Book Review

although unremarkable to a seasoned orous in its repeated analysis of alter- that no Turing machine can do is a AI researcher, is accessible to a general natives: deny, expand, supplement, task that no computer can do. audience. In short, this section of the and abandon. Although, occasionally, Second, for any Turing machine, we book could be useful in a class, semi- the distinction between expand and can devise a task that it cannot do, nar, or discussion group with a diverse supplement is lost on the reader. namely, define a computation such audience. From a cognitive psychology per- that the Turing machine will not be spective, Thagard does a nice job of able to tell whether it stops. incorporating cognitive psychology Third, human mathematicians, if Challenges to Cognitive theory and research into his general they know that the computation is Science overview of cognitive science. His dis- sound (consistent, error free), can tell cussion manages to include both tra- that the Turing machine does not In this second section, which consti- ditional research and more recent stop. tutes the last third of the book, Tha- developments that might be placed Fourth, humans can do something gard presents chapters entitled “Emo- under the rubric of ecological ap- that no Turing machine and, hence, tions and Consciousness,” “Physical no computer can do, namely, recog- and Social Environments,” “Dynamic proaches to the study of . nize mathematical truth. Systems and Mathematical Knowl- Ecological approaches emphasize the One of the difficulties with the for- edge,” and “The Future of Cognitive study of cognition within its everyday mulation is that the second and fourth Science.” This area is where the real context. Relevant topics discussed by technical “meat” lies for the seasoned Thagard include the relationship steps themselves express inferences. reader. Each chapter presents several between cognition and emotion and Of course, this problem is not serious, interesting challenges to CRUM and the important role that social context but it does force the careful reader to explores solutions that include deny, plays in cognition. break the steps apart. expand CRUM, supplement CRUM, From a philosophy perspective, Another difficulty is that the steps and abandon CRUM. The chapter on Thagard’s attempt to deal with some are not formulated as precisely as one emotions and consciousness describes challenges to CRUM has the virtue of might want. For example, what exact- the mind-body problem and explores bravely accepting the difficult facts. ly is the quantificational structure of how emotions and consciousness Defenders of CRUM sometimes deny step 2: “For any x, there exists a y…,” challenge CRUM. The chapter on the existence of consciousness and or “there exists a y such that for any physical and social environments dis- intentionality, but Thagard wants no x…”? Of course, these are not equiva- cusses the effects of the world and part of the “eliminativist’s” way out. lent forms. Literally read, step 2 “being in it,” with implications for He instead makes broad proposals appears to use the first form, but it’s robotics, situated action, the body about expanding and supplementing not clear that the first is strong and direct , and intention- CRUM. Perhaps a richer account of enough for the intended conclusion. ality. Thagard also relates these con- representation and computation will Another ambiguity in step 2 con- cerns to the social context of knowl- help with consciousness, and he the- cerns the phrase “a task that it cannot edge. In the chapter on dynamic orizes, biological considerations do.” What exactly is this task—defin- systems and mathematical knowl- might have to be invoked to help ing a computation such that an arbi- edge, Thagard brings in chaos theory explain intentionality if we are to trarily selected Turing machine will and the complexities of large dynamic avoid dualism or any other anticom- not be able to tell whether it stops or systems. Mathematical knowledge putational view. having an arbitrarily selected Turing turns out to be a discussion of Gödel’s Gödel’s incompleteness theorem: machine determine whether it will incompleteness theorem and Pen- One of the challenges to CRUM taken stop? rose’s (1994, 1989) version of Gödel’s up is based on Gödel’s incompleteness A more serious problem with the argument against computational theorem. The discussion focuses on a formulation is that the argument does views of the mind. In the final chap- particular version of the challenge for- not appear to be formally valid as it ter, on the future of cognitive science, mulated by Roger Penrose. In general, stands. Just one indication is the fact Thagard reviews the vast array of although Thagard has some interest- that the conclusion contains a predi- open questions and extends an invita- ing things to say, his discussion is cate (“recognize mathematical truth”) tion to interested parties to join the somewhat difficult to follow. This is that does not appear in any of the quest. unfortunate given that the text pur- premises. From an AI perspective, these chap- ports to be an introduction and the In any case, it’s unclear exactly how ters seem to do a good job of delineat- likelihood that at least some readers Thagard’s response to Penrose’s argu- ing the major objections and con- will presumably come to the book ment relates to its deductive structure. cerns about the field. Many AI texts lacking a background in the theory of His apparent rejection of the conclu- touch on one or several of these computable functions. sion (“Penrose has not shown some- objections, but Thagard clearly cate- Here is Thagard’s formulation of thing that a human mathematician gorizes them and provides instructive Penrose’s argument: can do that no computer could…” [p. delineations. His analysis of the possi- First, anything a computer can do, 178]) logically commits him to chal- ble responses to the objections is rig- a Turing machine can do; so, any task lenging either the argument’s validity,

134 AI MAGAZINE Book Review

CALL FOR PARTICIPATION Sixth International Conference on Principles of Knowledge Representation and Reasoning (KR’98) Trento, Italy June 2–5, 1998

Explicit representations of knowledge manipulated by inference algorithms provide an important foundation for much work in Artificial Intelligence, from natural language dialogue systems to expert systems. We intend KR’98 to be a place for the exchange of news, issues, and results among the community of researchers in the principles and practices of knowledge representation and reasoning (KR&R) systems.

Conference Chair: Stuart C. Shapiro, SUNY at Buffalo, USA ([email protected]) Program Co-Chairs: Anthony G. Cohn, University of Leeds, UK ([email protected]); Lenhart Schubert, University of Rochester, USA ([email protected]) Local Arrangements Chair: Fausto Giunchiglia, IRST, Trento, Italy ([email protected])

http://www.kr.org/kr/kr98/ Sponsored by KR, Inc. Organization supported by IRST-ITC [email protected] In cooperation with the AAAI

Workshops which will be held in conjunction with KR’98: Seventh International Workshop on Nonmonotonic Reasoning (NMR’98)–May 30 - June 1, 1998 IFIP Workshop on Knowledge Representation for Natural Language Processing – May 31 - June 1, 1998 Workshop on Validation & Verification of Knowledge-Based Systems (V&V’98)–June 1, 1998 International Workshop on Knowledge Representation for Interactive Multimedia Systems (KRIMS II)–June 1, 1998 International Description Logics Workshop (DL’98)–June 6 - 8, 1998 International Conference on Formal Ontology in Information Systems (FOIS’98)– June 6 - 8, 1998 its soundness, or both, which is where that anything Thagard says explicitly the Laws of Physics. Oxford, U.K.: Oxford things get a bit murky. He says that denies any of the steps he specifies as University Press. even if a full cognitive model (comput- premises (1, 2, and 3). er-aided machine [CAM]) of a human mathematician could be constructed Conclusions Bonnie Holte Bennett is director of the (an unlikely accomplishment accord- Artificial Intelligence Laboratory and an ing to the author), it is improbable that Thagard presents a broad introduction associate professor in the Graduate Pro- any human mathematician would ever aimed at a general audience and gen- grams in Software at the University of St. Thomas. Her e-mail address is bhbennett@ erally succeeds very well. His organiza- be able to construct the Turing stthomas.edu. machine equivalent of a CAM or prove tion is appealing and accessible to a that a CAM is sound. He closed the dis- variety of disciplines. Both experts and novices will find it a rewarding read, Dwight Nelson is a professor in the Biolo- cussion with, “Since CAM is not using gy/Neuroscience Department at the Uni- with much fodder for discussion. The a knowably sound algorithm, it is not versity of St. Thomas. obviously different from a human small section on Gödel’s incomplete- mathematician, characterized in Pen- ness theorem raised more questions for us than it answered, but this, too, Russell Pannier is a professor at the rose’s G as also not using a knowably William Mitchell College of Law. sound algorithm” (p. 178). How pre- can be a virtue. cisely does this relate to the deductive References Thomas Sullivan is a professor in the structure of Penrose’s argument? Does Penrose, R. 1994. Shadows of the Mind: A Department of Philosophy at the Universi- he maintain that it is formally invalid? Search for the Missing Science of Con- ty of St. Thomas. If so, what does he think are the miss- sciousness. Oxford, U.K.: Oxford University ing premises? However, does he main- Press. Gregory Robinson-Riegler is a professor in tain that a premise is false? If so, which Penrose, R. 1989. The Emperor’s New the Department of Psychology at the Uni- one? In this regard, it is not obvious Mind: Concerning Computers, Minds, and versity of St. Thomas.

SPRING 1998 135