There are some small errors and disappointments in the text (inevitable in a book of such ambitious scope). Claude Shanon's work on switching circuits is placed in 1948 instead of ten years earlier (p. 10) and HAL is said to be flying to Saturn instead of Jupiter in the discussion of Arthur Clark's 2001 (p. 179). Japan's obvious lead in specialized robotics is ignored in the discussion of that subject, leaving the uninformed reader with the impression that the Untied States, as always, is the technological bastion of work in that field. Finally, even though the non-fiction and fiction bibliographies are extensive, the fiction bibliography does not contain any references other than author, title, date: no sources! If the work is footnoted, the note contains a source, but the majority of the books in the bibliography are not footnoted. This is a very odd situation considering that the author goes to great pains to explain the efforts she made locating some of the stories (including raiding private collections). This book is a good place for AI researchers to find out how their work is portrayed to the public. 3. MICS - Mixed-initiative conversational system or in other words, computer programs making it possible to converse with humans in fairly normal speech. Each of these models are explained in good detail with good diagrams and illustrations, Some of the important concepts emerging from these models include, goals trees, goal-based counter-planning, structures of plan conflicts and goal trees representing personality traits. T h e book is clearly written with good tables and specific illustrations of the point being made. It has a number of clear and well-written flow charts describing the processes. The author is aware of some of the "problems" in the programs though it would appear that these are not really problems as much as indications of future work that needs to be done. Some of the "problems" lead to interesting and amusing examples, one is given on page 183 on the Triad System where a discussion occurs as follows: Interpretation of U.S. position on politics with a liberal viewpoint. The input is "Russia massed troops on the Czech border." The question is "What should the U.S. do about it?" The computer responded with the "The Unitesd State should congratulate Brezhnev." James E. Tomayko Dept. of Computer Science Wichita State University Wichita, KS 67208 At first blush this is really quite humorous but the author goes on to explain how this could easily come from the model. In this case, the goal conflict model suggests that we might want to improve diplomatic relations and one way to improve diplomatic relations is to support the actions of the other country. Examples like this show keen insight by the author into some significant implications of his system. SUBJECTIVE UNDERSTANDING, COMPUTER MODELS OF BELIEF SYSTEMS Jaime G. Carbonell UMI Research Press Ann Arbor, Michigan Copyright 1981 This is a very interesting and very important book. It is based on Carbonell's 1979 Yale Ph.D. thesis working under Schank and Abelson and others. The book is only 285 pages long but it has a lot of outstanding material in it. It has a very useful discussion of the whole question of subjective understanding. It appears that there are operating programs running on the DEC 10 or DEC 20 though no program listing are given. It would have been helpful if there had been a little more technical information on the program languages used and the amount of memory and machine capacity used. In the book there are three main computer models as follows: 1. POLITICS - A model of ideological reasoning. 2. TRIAD - Techniques for analyzing social and political conflicts. There are a couple of unsettling things as one studies the book, one problem is that it is not at all clear how the author develops the goal trees. It would be nice if he could give the reader some help on how to go about developing new goal trees in a particular situation. It would be nice if these programs could be tied to the "BELIEVER" of Schmidt and Sridhran. It is not quite clear either how the programs are tied together in some manageable form. It appears that the programs are complete in the sense of handling the processing of English input and output and the internal logical processing. These must be large programs and there is some inference that these are not modular in terms of not being able to debug it in smaller pieces. I suspect that there are some problems there. Overall, this is an outstanding book. l believe that it should be required reading for all AI enthusiasts with interests in language and computer/human interaction. It is a job well done and I am sure we will be hearing additional good things from the author. Dr. Gary Carlson Computer Translation, Inc. 1455 South State Street Orem, Utah 84057

/lp/association-for-computing-machinery/review-of-subjective-understanding-computer-models-of-belief-systems-oGDNteauat