Introduction This report summarizes our experience in building PIES, a knowledge-based system that diagnoses problems in semiconductor fabrication processes by analyzing parametric test data. Parametric measurement, which is performed on test circuits at the end of a complicated semiconductor fabrication process, provides semiconductor engineers with early information to monitor the "health' ' of the overall fabrication process. Typically, hundreds of measurements are made on each wafer. The problem is to reduce the resulting ream of data to a concise summary of the process status: whether the process is functioning correctly and, if not, what the nature and cause of the abnormality is. Currently, this interpretation taskis performed by a group of semiconductor specialists known as failure-analysis or yield-enhancement engineers and routinely consumes a large portion of their time. It is critical that problems be identified quickly to avoid a major operational loss.

The Workshop on Term Subsumption Languages in Knowledge Representation was held 18-20 October 1989 at the Inn at Thorn Hill, located in the White Mountain region of New Hampshire. The workshop was organized by Peter F. Patel-Schneider of AT&T Bell Laboratories, Murray Hill, New Jersey; Marc Vilain of MITRE, Bedford, Massachusetts; Ramesh S. Patil of the Massachusetts Institute of Technology (MIT); and Bill Mark of the Lockheed AI Center, Menlo Park, California. Support was provided by the American Association for Artificial Intelligence and AT&T Bell Laboratories. This workshop was the latest in a series in this area. Previous workshops have had a slightly narrower focus, being explicitly concerned with KL-One, the first knowledge representation system based on a term subsumption language (TSL), or its successor, NIKL.

Department of Computer Science Rutgers Universaty New Brunswick, New Jersey 08903 Abstract In this article we discuss a method for learning useful conditions on the application of operators during heuristic search Since learning is not attempted until a complete solution path has been found for a problem, credit for correct moves and blame for incorrect moves is easily assigned We review four learning systems that have incorporated similar techniques to learn in the domains of algebra, symbolic integration, and puzzle-solving We conclude that the basic approach of learning from solution paths can be applied t,o any situation in which problems can be solved by sequential search Finally, we examine some potential difficulties that may arise in more complex domains, and suggest some possible extensions for dealing with them. PEOPLE LEARN FROM EXPERIENCE, and for the past 25 years, Artificial Intelligence researchers have been attempting to replicate this process. In t,his article we focus on learning in domains where search is involved. Furthermore, we will restrict our attention t,o cases in which the legal operators for a task are known, and the learning task is to determine the conditions under which those operators can be usefully applied. Once such a set of heuristically useful conditions has been discovered, search will be directed down profitable We would like to thank Jaime Carbonell and Hans Berliner for helpful comments on an earlier version of this article.

Editor's Note: AI workers have claimed for some time A partial evaluator is an interpreter that, with only partial information about a program's inputs, produces a specialized version of the program which exploits the partial information. A similar example is described in more detail in Kahn (1982b). Programming methodology in AI shares much with general programming methodology but differs in significant ways. An AI researcher does not typically understand the problem being programmed very well. An essential aspect of a very common style of doing AI research is to write programs in order to understand something better.

We report on the spring 1992 symposium on diagrammatic representations in reasoning and problem solving sponsored by the American Association for Artificial Intelligence. The symposium brought together psychologists, computer scientists, and philosophers to discuss a range of issues covering both externally represented diagrams and mental images and both psychologyand AIrelated issues. In this article, we develop a framework for thinking about the issues that were the focus of the symposium as well as report on the discussions that took place. We anticipate that traditional symbolic representations will increasingly be combined with iconic representations in future AI research and technology and that this symposium is simply the first of many that will be devoted to this topic. The emphasis of this symposium was diagrammatic (or pictorial) representations in problem solving and reasoning.

The object is to bring the situation, or problem state, forward from its initial configuration to one satisfying a goal condition. For example, an initial situation might be the placement of chessmen on the board at the beginning of the game; the desired goal, any board configuration that is a checkmate; and the operators, rules for the legal moves in chess. This difference is then used to index the (forward) operator most relevant to reducing the difference. If this especially relevant operator cannot be immediately applied to the present problem state, subgoals are set up to change the problem state so that the relevant operator can be applied. After these subgoals are solved, the relevant operator is applied and the resulting, modified situation becomes a new starting point from which to solve for the original goal.

ARTIFICIAL INTELLIGENCE is that part of Computer Science that concerns itself with the concepts and methods of symbolic inference and symbolic representation of knowledge. But within the last fifteen years, it has concerned itself also with signals-with the interpretation or understanding of signal data. AI researchers have discussed "signal-tosymbol transformations," and their programs have shown how appropriate use of symbolic manipulations can be of great use in making signal processing more effective and efficient. Indeed, the programs for signal understanding have been fruitful, powerful, and among the most widely recog-Many different people helped in building HASP/SIAP in many different capacities. The people acknowledged below are project leaders (*), consult.ants;

Editor: We are currently working on a project that attempts to integrate artificial intelligence and legal reasoning for the purpose of simulating judicial decision making. The project has defined legal reasoning and legal analysis-the former taking place before the latter begins. Using a historical approach with our legal system's basis founded in English common law, we attempted to examine the role of stare decisis in decision making. More extensively we examined the role of reasoning in legal analysis, relying on Wittgenstein and to some extend Hofstadter, for an explanation of the foundation of the thought behind man's reasoning process. Legal reasoning is a specialized thought process, but reasoning is generic to all processes that attempt to incorporate artificial intelligence.

AIpanel discussion at AAAI-84, in which I was one of the panelists, appeared in the Fall, 1985 issue of AI Magazine, since due to some communication gap I wasn't aware that the panel discussion was going to be published and I hadn't had a chance to proofread my section of the transcript. I was rather unhappy when I read the section that contained my remarks: Perhaps because of an accent that would not vanish after 20 years in this country, my remarks were, in significant places, embarrassingly garbled by the transcriber ("the most performed paradigmatic change"?, "AI has been the whole expectation of the problem"?, "Knowledge use invalidities has been the cause of misunderstanding"?), and in other places, the crucial "not" had been omitted or added, completely changing my intended meaning, "not" being generally very unforgiving in this regard (where I had said, "The problem is underestimation of the problems of multiplicity of generic knowledge structures," "is" appears as "isn't;" I am pretty sure I didn't say, "I also believe that faster architectures could do the trick," since at that stage in my talk, I was criticizing the belief that what it takes is faster architectures, while crucial epistemic problems remained unsolved). Perhaps it is best to outline the main points of my panel presentation to make clear what I really said (this time without an accent and slowly): 1. AI has already made significant paradigmatic contributions by fostering the idea of cognition as computation. This notion is bound to have far-reaching consequences to philosophy and psychology. This is a weak theory of mind (or mental architecture) in the sense that it says something about organization, but doesn't make any strong commitment about content.

MIT Artificial Intelligence Laboratory The MIT AI Laboratory has a long tradition of research in most aspects of Artificial Intelligence. Currently, the major foci include computer vision, manipulation, learning, Englishlanguage understanding, VLSI design, expert engineering problem solving, commonsense reasoning, computer architecture, distributed problem solving, models of human memory, programmer apprentices, and human education. Understanding Visual Images Professor Berthold K. P. Horn and his students have studied intensively the image irradiance equation and its applications. The reflectance and albedo map representations have been introduced to make surface orientation, illumination geometry, and surface reflectivity explicit. Recent work has centered on modelling the effects of the atmosphere which distort intensity values and make classification of terrain and related computations using the albedo map inaccurate.