Johnson-Laird In a field choked with seemingly impenetrable jargon, Quick and thorough. Philip Johnson-Laird has done the impossible: written a By mixing forward and backward chaining, goal search book about how the mind works that requires no advance time can be shortenedramatically And, using GURU's knowledge of artificial intelligence, neurophysiology, or multiple rule firing capabilityou can refire rules psychology, providing the single best introduction to cognitive as values change GURU also comes equipped with science available. "Philip Johnson-Laird has that rare gift of being a cognitive seamlessly integrated 4th generation decision support scientist of the first order, yet he addresses himself to capabilitiesuch as data base, spreadsheet, and the deep classical issues in psychology, in the philosophy report generator

The Robotics Institute, Carnegie Mellon University Pittsburgh, PA 15213 Abstract In non-cooperative multi-agent planning, resolution of multiple conflicting goals is the result of finding compromise solutions. Previous research has dealt with such multi-agent problems where planning goals are well-specified, subgoals can be enumerated, and the utilities associated with subgoals known. Our research extends the domain of problems to include non-cooperative multi-agent interactions where planning goals are ill-specified, subgoals cannot be enumerated, and the associated utilities are not precisely known. Negotiation is performed through proposal and modification of goal relaxations. Case-Based Reasoning is integrated with the use of multi-attribute utilities to portray tradeoffs and propose novel goal relaxations and compromises. Persuasive arguments are generated and used as a mechanism to dynamically change the agents' utilities so that convergence to an acceptable compromise can be achieved.

This paper describes a Bayesian technique for unsupervised classification of data and its computer implementation, AutoClass. Given real valued or discrete data, AutoClass determines the most probable number of classes present in the data, the most probable descriptions of those classes, and each object's probability of membership in each class. The program performs as well as or better than other automatic classification systems when run on the same data and contains no ad hoc similarity measures or stopping criteria. AutoClass has been applied to several databases in which it has discovered classes representing previously unsuspected phenomena.

This article introduces a class of incremental learning procedures specializedfor prediction that is, for using past experience with an incompletely knownsystem to predict its future behavior. Whereas conventional prediction-learningmethods assign credit by means of the difference between predicted and actual outcomes,tile new methods assign credit by means of the difference between temporallysuccessive predictions. Although such temporal-difference method~ have been used inSamuel's checker player, Holland's bucket brigade, and the author's Adaptive HeuristicCritic, they have remained poorly understood. Here we prove their convergenceand optimality for special cases and relate them to supervised-learning methods. Formost real-world prediction problems, telnporal-differenee methods require less memoryand less peak computation than conventional methods and they produce moreaccurate predictions. We argue that most problems to which supervised learningis currently applied are really prediction problemsMachine Learning 3: 9-44, erratum p. 377

As Soloway attracted over 400 pnrticipants from all of Pittsburgh, Pittsburgh, Penn., described the changes in what he felt over the world who gathered to present 8-10 May 1987. The conference the construction of mechanisms and concerning AI and education This article cochairmen, Stellan Ohlsson and explanations last year to the design of presents a synopsis of the major Jeff Bonar, also gave brief welcomes to artifacts today, he was clearly giving presentations and an overview the participants. With so about transference, leading Soloway many attendees from abroad (The to conclude that transference is not Netherlands, Japan, Canada, West the ultimate goal for teaching and Germany, England, Sweden, France, tutoring programming. Instead, the and Hong Kong were all represented concern should be for the development by speakers), the international flavor of synthesis skills and "highorder of the conference was well established. The obvious disappointment This model does not vary significantly of the audience could be from standard software engineering felt. However, instead of giving the opening address, "Programming requiring these steps be followed in a as Artifact Design." This change strict order, Soloway contends that worked out well because Soloway the way real programmers work best acted like a cheerleader, getting the is to bounce from one stage to another crowd fired up about the subject of AI as the need arises. WINTER 1987 97 Andy di Sessa, in his talk "Social much rigidity has recently been the differences between beginner and Niches for Future Software," focused imposed on programmers by the engi-expert. Finally, Wender suggested that on the need to provide a medium neering approach. He demonstrated him, one could easily mistake him for teacher. Some of the kinds of software he felt should ... He considers current applications to be "the He also suggested that "current programming is to synthesis as a hammer is to a thumb. Each is as likely to challenge to the computer science der was echoed by Ben du Boulay in cause pain as [it is] to get the job community to develop higher-level "What Should a Programming Environment done." The Like?" Bonar's comment in his opening Beyond the usual categories supplied emphasis should be on synthesis welcome that we are "on the verge by the conference structure, several skills for designing, generating, and of a breakthrough" in developing themes linked many of the papers evaluating alternative artifacts that tutoring systems concerned du and presentations.

In the restaurant script, notated as $RESTAURANT, the roles might directly to the United Press International Yale researchers explored intentionality include customer, waitress, and cook; news wire and could skim news One of the earliest programs to the props could be a menu, table, and stories in dozens of different domains, embody goals and plans within the silverware; the locations could be the and produce summaries in several languages. CD paradigm was Jim Meehan's bar, dining area, and kitchen; and the On the DEC-20 (which by TALESPIN, which made up stories events would include arriving, seating, 1978 had replaced the PDP-101, similar to the fables of Aesop.

This report contains a historical perspective on previous Distributed Artificial Intelligence Workshops, highlights of the roundtable discussions, and a collection of research abstracts submitted by the participants.

One observer's report on the Artificial Intelligence and Human Mind Conference, held 1-3 March at Yale University. The conference was organized and sponsored by Truth ( a journal of modern thought) and The International Institute for Mankind. The conference included Sir John Eccles, the nobel laureate neurobiologist, physicists Henry Margenau and Eugene Wigner, and AI researchers Marvin Minsky, Michael Arbib, Hans Moravec and Doug Lenat.

The First International Workshop on User Modeling in Natural Language Dialogue Systems was held 30-31 August 1986 in Maria Laach, West Germany. Issues addressed by the participants included the appropriate contents of a user model, techniques for constructing user models in both understanding and generating natural language dialogue, and the development of general user-modeling systems. This article includes an overview of the presentations made at the workshop. It is a compilation of the author's impressions and observations and is, therefore, undoubtedly incomplete; and at times might fail to accurately represent the views of the researcher presenting the work.

This article is a written version of an invited talk on artificial intelligence (AI) and the history of science that was presented at the Fifth National Conference on Artificial Intelligence (AAAI-86) in Philadelphia on 13 August 1986. Included is an expanded section on the concept of an abstraction in AI; this section responds to issues that were raised in the discussion which followed the oral presentation. The main point here is that the history of science can be used as a source for constructing abstract theory types to aid in solving recurring problem types. Two theory types that aid in forming hypotheses to solve adaptation problems are discussed: selection theories and instructive theories. Providing cases from which to construct theory types is one way in which to view the history of science as "complied hindsight" and might prove useful to those in AI concerned with scientific knowledge and reasoning.