To start with contents, the claim that "the time is ripe..." to analyze multiple cues aroused my first doubts; while theoretically it makes logical sense, the authors do not attempt to substantiate it by any evidence from the natural world, such as anatomy, physiology, or psychophysics. I would like to see some discussion of results that would support such an approach, if such can be found. In the lack of such evidence, justifying the approach would become quite difficult. As a matter of fact, it seems like the jury is still out on whether separate modules interact with each other or are completely autonomous in their contributions to the process of solving visual problems. (Compare, e.g., M. Livingstone and D. Hubel, Segregation of Form, Color, Movement, and Depth: Anatomy, Physiology, and Perception, Science,Vol. 240, May 6, 1988, pp. 740.-748 and E.A. DeYoe and D. C. Van Essen, Trendsin Neurosciences, Vol. 11, No. 5, May 1988, pp. 219-226.) More generally, Mart's book is a study in vision (not computer vision). True to this, it is extremely well researched and documented and presents a rich reference list in all aspects of vision. The current book features a reference list that is very strong in computer vision but very weak in other aspects of vision. I find it very difficult to accept theories on computer vision without suffi- cient support from the real thing. In addition, though the authors promised us in the beginning what appeared to be a balanced account of both the bottom-up and top-down approaches, I found the top--down discussion (Chapter 8) much less organized and, perhaps, much less well thought through than the bottom-up approach discussion, which is broken into five chapters (Chapters 3-7); it made reading of the bottom-up part much easier, and made me speculate that perhaps the authors did not spend as much time and effort on the top-.down part. From the stylistic point of view, the authors adopt some of Marr's philosophical style of presentation (which one may or may not like). However, when I first flipped through the pages, I was quite happy to find more mathematical expressions, which I hoped would describe matters in a clearer way than Marr's long verbose style. Unfortunately, I was later disappointed to find out that a lot of the mathematics seems to be "hanging on its own" with the connection to the text around it not clearly stated. The same goes for the figures; the captions are very brief, and do not help in the understanding of the figures and their contribution to the text. In particular, one cannot obtain any information from looking at the figures and reading the captions without going to the text. Knowledge Representation: An Approach to Artificial Intelligence By T.J.M. B e n c h - C a p o n Academic Press, London, 1990 H a r d cover, pp. 220, I S B N 0-12-086440-1, $39.00 Reviewed by: Marc L a u n t s e n H a r v a r d L a w School Cambridge, Massachusetts 02138 laurit@hulawl .bitnet Who needs yet another introductory text on artificial intelligence? If it's as lucid and comprehensive as T.J.M. Bench-Capon's Chapter 8 is devoted to PROLOG,which is discussed not only as a concrete example of logic programming, but as an illustration of the compromises necessary in a practical language. PROLOG's extra-logical features in the areas of input-output, database update, and program control recceive particular attention. The treatment of expert systems (Chapter 9) is a rudimentary one, but contains extensive discussions of both early and contemporary expert systems, available shells, and trends in the field. Chapter 10 closes the book with a comparison of the paradigms and a discussion of such cross-cutting knowledge representation issues as treatment of negation, non-monotonicity (plausible and default reasoning, truth maintenance), inexact reasoning (uncertainty, probability, fuzzy logic), representation of control knowledge (e.g., through meta-predicates), temporal reasoning, and model-based representations. Arising from the author's own extensive background in logic programming at London's Imperial College, the book places somewhat more emphasis on logical and rule-based approaches than it does on frame-based and other structured object paradigms of knowledge representation. But Bench-Capon avoids all traces of dogmatics and delivers an admirably well-balanced treatment of the issues that leaves ultimate conclusions up to the reader. Knowledge Representation, many of us do. This text introduces AI through a study of knowledge representation issues; it assumes that the reader has a basic knowledge of computing and some familiarity with logic. While aimed primarily at students, the book serves as a very satisfying survey for those moderately well-educated in AI (like this reviewer) and should offer a useful brush-up to experts as well. References are provided at the end of all chapters, and exercises are included with most chapters. Following an opening chapter on the history and goals of AI, the author discusses the purposes of knowledge representation, the alternative paradigms that have emerged, and the criteria by which their adequacy and expressiveness can be judged. The deductive, inductive, and abductive forms of reasoning by which these representations are manipulated are noted. Chapters 3 and 4 introduce logic and search respectively, preparing the reader for the ensuing review of the major knowledge representation paradigms: production rules (chapter 5), structured objects (chapter 6), and logic and predicate calculus (chapter 7). The essential concepts of each of these approaches are presented, along with some of their advantages and disadvantages. I found the exposition refreshingly clear, except for the sections dealing with the difficult topics of resolution, unification, and skolemization, which were too abrupt for ready comprehension. Knowledge Representation is written in a clear, conversational style. While peppered with occasional subtle humor, it constitutes a meticulous introduction to the essential basics of its subject. It appears to be an excellent choice for an introductory textbook, as well as a refresher for the already initiated. S I G A R T Bulletin Vol. 2, No. 5

/lp/association-for-computing-machinery/book-review-knowledge-representation-an-approach-to-artificial-RgeJ7MYwJR