… The centrality and the ubiquity of analogy in creative thought have been noted again and again by scientists, artists, and writers, and understanding and modeling analogical thought have emerged as two of the most important challenges for cognitive science…” (from the book jacket).

Mitchell’s research as reported here grows out of Douglas Hofstadter’s continuing research program in cognitive science. It starts from the premise that analogy making is a high-level perceptual process in which the interactions of concept structures with current perceptions give rise to “conceptual slippages” that allow analogies to be made. Exploration of these ideas, themselves analogies, is carried out by implementing them in a computer model of analogy making and experimenting with its behavior under various types of input. The model, embedded in a program called “Copycat,” implements concepts and high-level perception as emergent phenomena, arising from large numbers of low-level, parallel, nondeterministic activities. Thus, it occupies a position between symbolic and connectionist models.

Copycat is designed only to deal with analogy making in a restricted cognitive domain, but this process is seen as an integral part of cognition in general. Thus, the book discusses such issues as the nature of concepts and perception and the emergence of highly flexible concepts from a lower-level sub-cognitive stratum. The long-term goal is to understand the mechanisms underlying the fluidity of concepts: their overlapping and associative nature, their blurry boundaries, their dynamic and graded relevance to a given situation, and their flexibility as a function of context.

In a real sense, this book contains elements of a theory of one aspect of cognition. The computer program implementing them can be said to be the theory, expressed in a form that can be executed rather than merely as a static diagram or descriptive paragraph. My own conjecture is that most social phenomena are sufficiently complex that only this kind of theorizing will provide sufficient precision for the development of useful models. One of the book’s contributions is that, by example, it provides a tutorial in computer-supported theorizing.

The book is designed as a research report, not as a text. Chapters 1 and 2 define the conceptual scheme used in the project. Chapter 3 relates these to the Copycat program architecture. Chapters 4 and 5 describe the behavior of the model on a series of five target problems and on variants of them. Chapters 6, 7, and 8 summarize the findings, discuss the shortcomings of the model, and compare the results with related work. Chapter 10 and an “Afterword” by Hofstadter discuss the work and its place in the larger study of intelligence. Three technical appendices document further types of problems and the details of the Copycat model. A bibliography and a good index conclude the book. Appropriate use of type styles, decimal numbering of paragraphs and sections, and the use of diagrams help the reader. The writing is exceptionally lucid and readable. I noticed no typographical errors.

The intended audience is professionals and graduate students working in artificial intelligence. It could and should also be read by anyone interested in social theory, however.