Browse Books

Reviewer: John M. Artz

Anyone who walks into a technical bookstore today may easily be overwhelmed by the number of volumes available, the variety of subjects addressed, and the depth of titles in any given area. Even if it were economically feasible, temporal constraints would make it impractical to read every volume. Frequently, a choice is made based on cover design and number of illustrations. Sometimes it is made on the strength of the author's name. An astute reader may even browse through the topics or look for the author's objectives. Nonetheless, choosing between the available titles is a challenge. The purpose of this review is to aid in that process. This is a comparative review of five introductory books on expert systems. It is intended to provide a critical overview of each of the five books, including material covered, target audience, and major strengths or weaknesses. These five books represent, as well as possible, a cross section of the types of books available at the introductory level. Tables 1 and 2 below provide a quick overview of the quanti- tative and qualitative points of comparison. Expert systems is a relatively new field and a classic introductory text has yet to emerge. The field has not yet pulled away sufficiently from artificial intelligence and still relies a little too heavily on successfully implemented research projects, such as MYCIN, for validity. Only one chapter out of eight in the Hayes-Roth et al. book presents material which is independent of existing implementations. Harmon and King and Waterman improve that ratio to one-half of the chapters, but both still rely heavily on references to existing applications. The Nagy, Gault, and Nagy book does not rely on this at all, yet neither does it present any conceptual framework. The Hayes-Roth et al. and Waterman books are the first and second volumes in the Teknowledge Series in Knowledge Engineering, and one of the authors of the Harmon and King book is connected to Teknowledge, Inc. While they are producing some of the highest quality books available, the effects of the gene-pool show through in the same basic approach in each volume. This approach, which does vary somewhat from author to author, is to document existing applications and system building tools, while abstracting some general principles as a conceptual framework. This works for now but will weaken as the field progresses. Nagy, Gault, and Nagy offer a different approach entirely, but require access to a microcomputer, some technical background, and perseverance on the part of the reader. Sell again offers a slightly different perspective, but expects an unrealistic level of prior knowledge on the part of the reader, at least for an introductory level book. Following is a short description of each of the five books: Hayes-Roth, Waterman, and Lenat This book, originally published in 1983, was one of the first, if not the first, devoted entirely to expert systems as an endeavor apart from artificial intelligence. It was a bold undertaking because the field of expert systems was just emerging and the conceptual foundations had not yet emerged sufficiently to define a standard and proper development of the topic. The book documents many of the existing expert systems as well as some of the software available for building them. From this, they abstract some general principles and approaches for the development of expert systems. The format is that of a collection of papers, but the editing is adequate so the presentation is fairly consistent. The editors recognize that this effort is ambitious. Yet remarkable progress was made and the roots of many future books may be seen in this volume. It is a little difficult to follow for a first introduction, but a good book to refer back to for a more in-depth look at some of the current thinking in expert systems. It will surely be valuable for some time to come. Although the ideas had not yet crystalized, it provides a wealth of background information and some useful historical perspectives. Harmon and King This is probably the best book, overall, on expert systems available today. It spends three chapters developing some of the conceptual foundations in a clear, readable style with lots of good graphics and examples. It addresses human problem solving, knowledge representation, and drawing inferences with no mention of the existing implementations, and thus serves the future as well as the past. Yet it is not a purely conceptual treatment. The authors explore existing implementations and systems building tools with similar clarity and they pragmatically address the actual development process. The reader begins to get the impression that building expert systems has a lot in common with building information systems. The strengths of this book are its readability and its development of the topic. The authors are, among other things, technical trainers and have a talent for presenting technical material in an understandable fashion. Its major weakness is that, as with many others, it spends a little too much effort on existing implementations. Waterman This is really two books in one and needs to be reviewed as such. The first 22 chapters are a crisp, well-organized introduction to expert systems. The information is well thought out and well presented. This part would serve as an excellent text for an introductory course. The remaining eight chapters (over 150 pages) provide an abundance of supplementary information, such as an annotated catalog of current applications organized by application area and cross-indexed by development tool. The strength of this book lies in its crisp, concise presentation of the issues surrounding the development of expert systems. The major points jump out at the reader, just waiting to be highlighted or underlined. This makes the book particularly useful for a technical manager or a technically oriented business student. The major weakness lies in the divergent focus. Even though it is two books in one, it is not the case that the two books address the same target audience. Nagy, Gault, and Nagy This book represents a different approach to expert systems development. It presents no conceptual framework, and makes no mention of existing applications. Instead, it provides an expert system shell called Micro-PS (a scaled down version of KES) and steps the reader through the development of a prototype expert system. The software is provided with the book and the reader must have access to an IBM-PC or a compatible microcomputer in order to use the software. The reader can build one of the example systems or one of his or her own choosing. With a 20 rule maximum, there is a severe limit on the sophistication of the prototype system. However, it is sufficient to give the reader a feel for a prototype development process and the workings of an expert system shell. The strength of this book is that it helps the reader to overcome some of the conceptual barriers associated with expert systems through hands-on experience. The evolving prototype approach may also be useful for a technical manager attempting to assess the feasibility of expert systems technology for a particular application area. The major weaknesses of the book are that it is a little hard to read and follow, and that it is almost impossible to use without access to a microcomputer. Sell This spartan volume is a highly condensed overview of expert systems development. It is less than 100 pages, with few wasted words and very little in the way of illustration. The style of the book is that of “straight talk” about expert systems from a veteran practitioner to a less seasoned one. All the bases are covered but little time is spent on any one. The first chapter, for example, addresses scientific revolutions, domain-independent versus domain-dependent problem solving, and the paradigm shift in artificial intelligence—all in one paragraph. Whew]] The author states that “It is hoped that the glimpse is sufficient to show that the techniques used are neither too arcane to comprehend nor too intricate to implement.” This objective was met, and not a single word was wasted in the process. The main strength of this book is in its brevity. The reader who has already read many of the introductory books and papers will find it refreshingly concise. Unfortunately, there is a fine line between being terse and being obscure. The reader with less background may become frustrated as large strides are taken in short passages. Conclusion @ The book by Harmon and King is the best, overall, for the person on the street who might walk into a technical bookstore looking for an introductory book on expert systems. Waterman is the best introductory textbook, although Harmon and King would not be a bad choice. Waterman offers the additional advantage that it is written in a style more appropriate for an introductory textbook. The Nagy, Gault, and Nagy book would be a good choice for a hands-on course, but needs either the Harmon and King or the Waterman book for the conceptual framework. Nagy, Gault, and Nagy is also a good choice for a technically oriented individual who wishes to learn by doing and leave the theory for later. Hayes-Roth et al. and Sell are a little less refined and should be attempted after some introductory groundwork has been laid. Both are a bit more obscure and may be a little too much for a first introduction.

Reviewer: John M. Artz

For those readers who only read the first few lines of a review, it must be said without further delay that this is a very good book. It is well worth reading by anyone with a practical interest in using or developing expert systems, or tracking the progress of expert systems technology. The book demystifies this application of artificial intelligence with a down-to-earth presentation of theory, application, and history of computer systems which endeavor to embody the kind of intelligence which has been, up until the last few years, left to human experts. The book is free of intellectual hype and anthropomorphic speculation. It addresses its topic with the calm and deliberate rationality that one would expect from a book on systems development, which has thus far been hard to find in this new area of business systems development endeavors. The book is divided into four sections: (1) Basic Concepts and Techniques, (2) Languages, Tools, and Systems, (3) Developing Expert Systems, and (4) The Expert Systems Market. The first section introduces the theoretical concepts underlying expert systems. It begins by introducing MYCIN (a medical expert system) and then presents problem solving, knowledge representation, and inference. The section concludes with a revisitation of MYCIN which goes into more detail in the expertise and inference process. The middle chapters (3, 4, and 5) are the high points of this section and may be the high points in the book. Chapter 3 addresses human problem solving, which must be understood, at least at a cursory level, before one can understand how machines solve problems. The chapter addresses types of problems (well-formed vs. ill-formed), the role and categories of knowledge, and what is meant by expertise. Chapter 4 moves into a more abstract realm and discusses knowledge representation independent of how humans actually store knowledge. It presents five different approaches to representing knowledge, including the most common: semantic networks, frames, and if-then rules. Once the knowledge is stored, the expert system can then use this knowledge to draw inferences. This process is the topic of the next chapter. Chapter 5 discusses the inference engine and search strategies which are at the heart of the expert system. Chapters 3, 4, and 5 take the mysticism out of expert systems in a clear, well-written presentation of the fundamental concepts with plenty of well chosen examples and lots of useful diagrams and pictures. Unfortunately, however, the section as a whole suffers just a bit from the use of MYCIN as the example of an expert system. The reasons for choosing it are fairly clear in that it is one of the few examples of a successful expert system. However, the tone of this section is that of a general introduction to expert systems fundamentals which have been abstracted and simplified for the purpose of conveying concepts. The MYCIN examples are difficult owing to the syntax and the fact that most readers are not familiar with medical diagnosis. Contrived examples from a more widely shared field of expertise may have proved more useful. The authors do provide such examples later in the book, but something is lost by that time. The second section examines the technology of expert systems through a presentation of LISP, PROLOG, and their derivatives, as well as a genealogy of some of the high-level expert system shells and commercially available tools. It also, somewhat tediously, explores more existing expert systems from the earliest accomplishments to some systems which are being used today. This book, unfortunately, sags a bit in the middle as the authors pay homage to founding applications within expert systems. It becomes tiring, after a while, to pick up another book on expert systems and read yet another recapitulation of MYCIN and friends. This information is of historical interest, and it is certainly of interest to researchers. But this parade of acronyms and accomplishments, which mean little to those who have never tried to simulate intelligence, is out of place (or at least excessive in 100 pages) for the practitioners or managers, who are the target audience, and who are simply trying to gain some familiarity with the topic. Further, it seems that eventually the field must take on a life of its own and not depend on the successes of the past for legitimacy. The third section will rejuvenate the reader's interest as it explores the expert systems development process. The authors begin with a small systems development effort. They show that it is based upon some fairly concrete steps that bear some relationship to a traditional systems development effort. The process is then expanded and formalized into a development effort for large systems. The clarity of this presentation helps to overcome the conceptual barriers in the reader's mind with regard to the feasibility of developing this type of system for a business application. Finally, the fourth section discusses the present and future uses for expert systems. At this point the authors could have exploited anthropomorphic fears of machines replacing human experts or attempted to minimize the extent to which computers will ever be intelligent. Instead, they adhered to the pragmatic approach, established earlier in the book, and addressed productivity and the need to leverage both expertise and knowledge. These issues strike a familiar chord with managers and bring the exploitation of expert systems technology closer to the heart of the business enterprise. This book was written for “executives, middle managers, computer systems personnel, and corporate trainers.” The book is even segmented in such a way that different readers can read selectively or from cover to cover depending on their specific interests. The first two target audiences will begin to see expert systems from a pragmatic perspective. They will see the strengths and usefulness of this technology and will be able to identify potential applications. Computer systems personnel will read with more of an eye for detail and will be able to begin bridging the gap from traditional information systems technology to expert systems technology. Finally, the corporate trainer can develop a short introductory course centered around the topics in this book. As an academic text this book would need to be supplemented by another text with perhaps less readability but more technical depth. With all due respect to the authors, the classical introductory text for expert systems has yet to be written. When it is, however, it will probably bear some strong similarity to some of the more impressive chapters in this book. In general, this book is a refreshing breeze in the dust storm of recent books promoting this new application of artificial intelligence called “expert systems.” Indeed, at its weakest points it is only guilty of being too much like all the others. Although it is not the last word on the topic, it is well worth the price for anyone whose curiosity has been aroused by the recent surge of interest in business applications of expert systems technology.