Browse Books

Reviewer: Dario A. Giuse

Once in a while, a book comes along on a topic that has been discussed ad infinitum and yet manages to convey a sense of freshness and new interest. This is one such book. Although the book does not contain any revolutionary new view of artificial intelligence, it presents the material under a new, interesting light and gives a unified view of the field. Whereas other books on AI read like laundry lists of different and unrelated techniques, the authors manage to keep the reader's attention through 700 pages, and provide a rich web of connections between different areas and topics. One reason for the success of the book is the clear effort to find and analyze the true innovations in the different areas of AI. Rather than relying on what everybody knows, Luger and Stubblefield have actually tried to go back to the real beginnings, and give descriptions of the pioneering works in each field that are both detailed and compelling. All the chapters are well written and balanced; the only one I did not like is chapter 7, the introduction to LISP, which takes an old-fashioned approach to the language and ignores its more modern features. These problems are corrected in later chapters about LISP. The presentation of Prolog, by contrast, is extremely lucid and quite interesting, and avoids unnecessary details that would obscure the central issues. The book covers many of the topics one would expect from a work on AI: formal logic (chapter 2); representation and search (chapters 3 through 5); AI languages (chapters 6, 7, and 13 through 15); and knowledge-based systems, knowledge representation, natural language processing, and machine learning (chapters 8 through 12). Several topics are omitted that are typically discussed in similar works, such as machine vision or speech understanding. Such omissions, however, are not a disadvantage; in fact, they may help avoid the typical laundry list impression. In addition to clear pseudocode descriptions, LISP and Prolog implementations are given for the main algorithms. The emphases on graph theory as a tool to represent search spaces, on logic as a tool to represent knowledge, and on embedded languages as a problem-solving technique act as unifying themes through the book. The primary emphasis of the book is on the symbol processing view of AI; subsymbolic theories are also explained clearly, however, and the differences between the two schools of thought are explained well. The philosophical roots of AI are usually covered in a perfunctory way. In this work, however, the authors go beyond the usual and show the gradual evolution of ideas from metaphysics and logic into modern-day AI research; in addition, they keep bringing up the central themes (the connections to logic, and AI as a model of human intelligence as well as a useful applied technology) throughout the book, giving the overall impression of unity I have mentioned. The book can be used as a classroom textbook for one- and two-semester courses on AI. An instructor's manual and a disk with all the programming examples are available separately from the publisher. Each chapter comes with a rich assortment of exercises, some of which could be full-scale projects (the exercises are not graded for difficulty). The bibliography is extensive; even better, most of the references are discussed briefly in the text, and the pioneering contributions are pointed out clearly and effectively. The typesetting is excellent, with virtually no typos. The index is also good. I enjoyed reading this book, which is something I cannot say of all the AI books I have read recently. It will give readers an excellent understanding of how the main building blocks of artificial intelligence (logic, graph theory, and so on) relate to the discipline, and how the main issues in the field (for example, symbolic versus subsymbolic representation) are playing out. I definitely recommend the book to anybody who wants to learn more about artificial intelligence.