cusses Prolog's role as a major language for Al and expert ... beginning user needs to get a good grip on the language ... desktop microcomputers.Written by the.
Prolog Programming and
Applications
W.D. Burnham and A.R. Hall (Halsted Press; New York, N.Y., 114 pp., $16.95)
Over the last few years, widening interest in Prolog has resulted in the recent appearance of various microcomputer versions for the IBM PC-and lately for the Macintosh and Atari ST. This brief introduction to Prolog is, therefore, quite timely. For interested readers, in its premier issue (Spring 1985, pp. 100-109) IEEE Expert discusses Prolog's role as a major language for Al and expert system applications. Chapter 1 explains the fundamental difference between conventional procedural languages and declarative languages like Prolog, with simple examples effectively highlighting how in Prolog one has only to describe or specify problems and requirements to be met. By using suitable rules and variables, the system can extract answers without users having to concern themselves with procedural details. Language structure is explained more formally in Chapter 2, with examples illustrating basic principles of search and pattern matching on which Prolog works. Chapter 3 deals with arithmetic operations and the power of recursive programs characteristic of Prolog. Chapter 4 outlines the facility for processing lists and structures-another of Prolog's major strengths. Chapter 5 covers input and output. These five chapters present lucidly and concisely-in just 67 pages-all a beginning user needs to get a good grip on the language. Chapter 6 introduces more advanced concepts, such as file processing, giving examples of Prolog as a convenient tool for natural language processing. And Chapter 7 explains typical interactive debugging facilities available in most implementations. The book concludes with two case studies-one of a typical advice system that asks users relevant questions and then gives quotations for auto insurance rates, and the second a natural language processing program. I tested out the first, written in Expert Systems International Prolog-1, and found it quite interesting; incidentally, it contains a few typographical errors, but readers who have read the preceding chapters should detect them easily. The second is in Quintus Prolog and I could not test it, but the program is fully understandable in the context of Chapter 6's explanations. According to their preface, the authors intended to produce a reference book allowing readers to achieve Prolog 108
proficiency rapidly while helping them
to understand "the great power and
depth behind its apparent simplicity." They have succeeded admirably, focusing on practical examples, then directing
readers to Clocksin and Mellish's classic text (Programming in Prolog, SpringerVerlag; New York, N.Y., 1984, $17.95) for a deeper understanding of underly-
ing theory. Two appendices give the syntax of ESI Prolog-I and Quintus Prolog-the
used by the authors. Since both are close to the Clocksin and Mellish standard used by most implementations, readers should have no difficulty making the slight adjustments necessary for whatever version they use. Although highpriced for a paperback of 114 pages (including index), this excellent summary provides a clear guide to Prolog and I recommend it highly. However, trying examples out is the best way. To start, try the educational-version EdProlog interpreter available from Automated Design Associates. It works well on IBM PCs and is priced at about $30. ones
controversy"
a discussion of how much that passes today as AI is (in the minds of many) good ol' systems engineering. Parts of this discourse are indeed convincing, leaving this reader wondering why he should finish a book about conventional systems design and development. Of course, the book is well worth reading, but I question the early placement of such potentially mis-
leading paragraphs. Simons takes a decidedly European viewpoint. References to the American push in Al and expert systems, though generally sound, are shallow in spots. For example, reference to DARPA's strategic computing program mistakenly describes that program as part of a larger national policy "not keen on
coordinated national research efforts that do not have military connections" (p. 37). I would argue that DARPA's program is more than benevolent toward
nonmilitary "enabling technologies"
and systems, including especially the development of multiprocessing architectures and systems. Given the book's recent publication date, there is really no explanation for -Ramachandran Bharath Simons's lack of attention to one of the Computer Information Systems Dept. largest application areas (on all contiNorthern Michigan University nents). We already have clear evidence Marquette, MI 49855 that parts of industry with little or no military vested interests have benefited substantially from DARPA seed investments-investments that can pay huge nonmilitary dividends in years to come. And the stimulus in basic university research from DARPA funding will G.L. Simons (NCC Publications; Manalso continue to fuel all segments of chester, UK, 1985, 247 pp., $19.95) society, not just the defense industry. Also curious is the lack of illustraThis short book provides an excellent tions. Al concepts are often best overview of expert system migration described visually. In fact, this is the from dedicated workstations and double-duty, high-end minicomputers to first time I've ever read about semantic desktop microcomputers. Written by the nets and frames without pictures. While Simons writes excellent prose, a picture chief editor of NCC Publications, it would frequently have been worth a chronicles that migration-looking at the revolution in microcomputing gener- thousand well-crafted words. ally, at the tools and techniques of Completely ignored is the Al state of expert systems design and development, mind. It's possible, after all, to design and at some early microcomputer-based and develop intelligent systems without expert systems and systems development using Lisp or a Lisp machine (such as a tools. Symbolics, Xerox, or LMI dedicated Not a highly technical study, this workstation). Al and expert systems work seeks to communicate a perspecseek to recreate in software the processes tive on the applied potential of by which human cognitive processors microcomputer-based Al and expert sys- solve complicated analytical problems. As always, the real design challenge lies tems, a perspective not anchored in heaps of technical jargon or attempting in the engineering of a knowledge base to extend the state of the art of Al or capable of rendering verifiable expert-system technology. From the out- inferences. set, however, the reader is left wondering Contrary to what Simons believes, it isn't always necessary to use Lisp or just how viable Simons believes microcomputer-based expert systems Prolog to program expert systems. So
Expert Systems
and Micros
really are. He begins with a fascinating dis-
course on what he dubs the "Al
long as intelligence-supporting functions are maintained in software (such as recursion, the ability to treat data as
IEEE EXPERT
objects, and the like), it doesn't matter which high-level language we select. Pascal, Ada, and (especially) C support these and other important functions, while running much more efficiently on microcomputers such as the IBM PC XT/AT and compatibles. In fact, we frequently refer to many microcomputer Lisp and Prolog versions as toys since they seldom permit the development of expert systems with knowledge bases deep enough to solve large classes of problems. Much of the American Al community has already discovered the joys of nonLisp intelligent-systems programming. When the target application system is a microcomputer, it's easy to find designers calling for design in Lisp perhaps- and implementation in anything but. We have no accepted definition of "microcomputer," nor does Simons attempt one. Five years ago, before the development of the 80286 chip, the notion of a desktop PC AT was still high in the sky. Soon, we will have VAXs sitting on our desks. Simons's assertion that "mainframes will continue to be most effective for many [AI] development projects" and that "microbased systems will be important as educational tools, as demonstrations systems, and for running a growing range of decision-support packages" challenges the technologists who gave us the PC AT, the new RT, the DEC MicroVAX II, and the soon-to-be-released 32-bit desktops. As one originally skeptical about microcomputing power, I suspect we'll see bona fide Al systems on next-generation micros. While much of the above sounds critical, I seriously recommend Expert Systems and Micros to those interested in acquiring a quick overview-especially if the reader is not versed in Al jargon. The book has an easy style, a solid bibliography, and an excellent glossary. On the other hand, I don't recommend the book all by itself. Readers interested in a comprehensive study should seek other books in the field. Similarly, those interested in the next level of detail should consult several texts. We are in the midst of what seems a continual state of redirection and redefinition. Today, Al is many things to many people. Given such confusion, the best approach (as always) involves sifting through various competing perspectives. Nevertheless, Simons's book is a good place to start.
-Stephen J. Andriole International Information Systems 802 Woodward Rd. Marshall, VA 22115 FALL 1986
I Literature Books. Artificial Intelligence: The Search for the Perfect Machine by Lawrence Stevens ($14.95) addresses natural language programs, expert systems, image understanding systems, robotics, voice recognition systems, distributed AI systems, teaching systems, and the Japanese fifth-generation project. Hayden Book Co., 10 Mulholland Dr., Hasbrouck Heights, NJ 07604; (800) 631-0856 or (201) 393-6300.
Artificial Intelligence: Bibliographic Summaries of the Select Literature by Henry M. Rylko reviews a total of nearly 450 titles in two volumes, emphasizing current literature. Titles include books, technical and research reports, proceedings, reference works, and journals. Volume I is $145 ppd., Volume II $180 ppd., and the set costs $295 ppd. The Report Store, 910 Massachusetts St., Ste. 503, Lawrence, KS; (913) 842-7348.
Artificial Intelligence & Robotics: Five Overviews (Artificial Intelligence; Robotics; Expert Systems; Computer Vision; and Computer-Based Natural Language Processing and Computer Vision) by William D. Gevarter (hardcover $99.50) is a 643-page storehouse prepared under the auspices of NASA and the National Bureau of Standards. Individual overview volumes are offered for sale in softcover velobind format (from $19.50 to $24.50 each). Business/Technology Books, 14 Evergreen Dr., PO Box 574, Orinda, CA 94563; (415) 839-3370. SDL Index to Artificial Intelligence Research in four volumes ($495) lists virtually every available research paper from every major US Al research facility, providing in its approximately 3000 pages an abstract for each paper, author and title indices, and a cross-referenced subject guide. Annual updates are published at $85 each. Scientific DataLink, Att: Chia Reinhard, 850 Third Ave., New York, NY 10022; (212) 838-7200. A paperback published in June of 1986-A Comprehensive Guide to AI and Expert Systems by Robert Levine, Diane Drang, and Barry Edelson ($19.95)-offers sample programs illustrating (for computer users) the major procedures used in developing working systems. Byte Books/McGraw-Hill, 1221 Ave. of the Americas, New York, N.Y. 10020.
Other recent Al texts include Design of an Interactive Manipulator Programming Environment by Ron Goldman ($44.95), describing a system for programming robots. Based on research at Stanford's Al Laboratory, Goldman discusses force sensing, motion control, problem identification, troubleshooting, and also outlines an interactive programming system for facilitating AI language manipulator program development. Query Optimization by Semantic Reasoning by Jonathan J. King ($37.95) describes how Al planning and inference methods, combined with knowledge of database storage and access, can be applied to large database
retrieval problems. Understanding ThreeDimensional Images: Recognition of A bdominal Anatomy from CA T Scans by Uri Shani ($42.95) shows how the CAT scan can be enhanced by computer analysis of 3D images. All three texts are sold by UMI Research Press, 300 North Zeeb Rd., Ann Arbor, MI 48106; (800) 421-3042. From Canada, call (800) 343-5299; from Michigan, Alaska, and Hawai call (313) 761-4700 collect. Magazines. Wall Street Computer Review, a monthly magazine featuring what's new in software and computer use in investment analysis, is offering their Annual Buyer's Guide as a free bonus to new subscribers. The guide (a $25 value according to the publisher) provides complete listings on investment software, databases, futures and commodities products, and investment industry automation products and services. The annual subscription rate for US residents is $39-the foreign rate, including Mexico and Canada, $59. US. Wall Street Computer Review, Subscription Dept., Box 77327, Trenton, NJ 08628. AI Expert, a commercial publication focusing on Al topics, first appeared in June. Produced by the publishers of Computer Language magazine and not to be confused with IEEE Expert, Al Expert will appear monthly. A one-year charter subscription costs $27; on newstands, the cover price is $3.50. CL Publications, 2443 Fillmore St., Ste. 500, San Francisco, CA 94115; (415) 957-9353.
Artificial Intelligence, Elsevier/NorthHolland's international journal first published in 1970, is offering a reduced ($50 US) 1986 individual subscription rate to members of recognized Al societies. In the US and Canada, contact the Journal Information Center, Elsevier Science Publishers, 52 Vanderbilt Ave., New York, NY 10017; overseas readers should contact North-Holland, Att: Karin van Schouten, PO Box 1991, 1000 BZ Amsterdam, The Netherlands. Reports. Now used mostly by industry, especially in automotion and electronics, machine vision is expected to show marked growth as potential users discover what it is and what it can do for them. The market, $433 million in 1986, is forecast to reach $1.4 billion by 1990. A new report entitled Machine Vision (Vol. 1), The Status of the Machine Vision Industry in 1986 examines the current market, defining what machine vision is and what it is not. Of the 56 machine vision companies currently in the US, all are covered and 30 are discussed in detail. Volume 2 of the report, Technology Issues in Machine Vision, will soon be available. Market Intelligence Research Co., 4000 Middlefield Rd., Palo Alto, CA 94303. 109