Browse Books

Reviewer: Pavol Navrat

The author's approach to AI is based on the fact that intelligence involves knowing things, and therefore artificial intelligence involves representing knowledge. The book is an attempt to cover the whole field of artificial intelligence by discussing all issues concerning knowledge representation. It consists of ten chapters and includes a well-balanced bibliography and index; additional references are appended to each chapter. The first chapter provides an introduction to AI and knowledge-based systems. Its content is limited to fundamental motivations of the field as a whole and of knowledge-based systems and expert systems in particular. To concentrate on motivations, ideas, and principles rather than on formal issues and development of appropriate formalisms and methods does not mean saying something less informative or valuable. On the contrary: understanding and explaining the basic motivations of problems, methods, and solutions of some field can be invaluable. Indeed, I find the author's formulation of the issues with clearly stated motivations invaluable. In the second chapter, Bench-Capon introduces knowledge representation. He defines it and presents criteria for adequacy and expressiveness as well as methods of reasoning. Basic notions of mathematical logic are introduced in the third chapter. Inevitably, any introduction to AI must discuss search, search methods, and their limitations; chapter 4 covers this material. The next three chapters discuss three typical approaches to knowledge representation: production rules, structured object representations, and predicate calculus. Chapter 8 examines Prolog as an instance of the predicate calculus paradigm for knowledge representation. Again, this chapter is not just another introduction to Prolog, but a clear discussion of what lies behind it. The ninth chapter discusses expert systems as systems applying the ideas described earlier in the book. The last chapter, somewhat inaccurately named “Some Issues in Knowledge Representation,” deals with important current topics of research. The author motivates even this: he starts by discussing similarities and differences between various paradigms, shows the partial inadequacy of their expressive power, identifies reasons for this inadequacy, and comments on proposed approaches to solving the problems. The book is intended for undergraduates taking their first course in AI, so no specific background is needed. It can be useful as a classroom textbook (chapters 2, 4, 5, 8, and 10 include exercises) in a first course, assuming that additional courses will follow. Since it provides almost no information on concrete methods and formalisms, the whole methodological apparatus of AI will have to be built later. It is a pleasure to read a book where everything fits together nicely, because we clearly understand each problem and appreciate how it can be solved. Many professionals will enjoy the book as a reintroduction to the field.