AAAI- Tenth National Conference on Artificial Intelligence IAAI- Fourth Innovative Applications of Artificial Intelligence Conference

Registration Brochure July  ‒ , 

San Jose Convention Center San Jose, California 2 AAAI‒ Please join us Opening Address  at AAAI– The Gardens of Learning— The National Conference on Artificial Intelli- The Vision of AI gence has been and remains the forum at which by Oliver G. Selfridge, Senior Staff Scientist, the highest quality new research in artificial in- GTE Laboratories telligence is presented. The highly competitive Tuesday Morning, July ,  review and selection process helps assure this. For the  conference, only  of  submit- With studies of other kinds of learning in ted papers were accepted. This year, the pro- symbolic and numeric domains, and with the gram committee was encouraged especially to resurgence of interest in neural networks, seek out research results that help to bridge the Oliver Selfridge hopes that learning, the sin- gaps growing between maturing sub-disciplines gle most important component of intelli- of AI. We have, as usual, also sought out the gence, may be eventually understood well new, innovative research ideas that may be to- Oliver G. Selfridge, enough even to be useful. He insists that morrow’s popular research areas. Senior Staff Scientist, GTE Laboratories learning itself is no single or simple thing; The content of the program reflects those ar- that why you learn is as important as what; eas where much current research is focused. A that learning is not only the key to AI, but al- series of nine sessions will explore exciting re- so its essence; and that in this way AI will in sults in learning, approached from diverse view- the long run become the cornerstone of com- points. Another series of ten sessions covers new puter science, and will help to make software results in planning and coordination, and prob- AAAI- Program technology effective and powerful. lem-solving topics including search and con- Selfridge has consulted very extensively for straint satisfaction. One session looks at the em- July -,  the United States Government, and has served on panels for the Department of De- pirical revolution in natural language process- The AAAI- program is multi-faceted, and fense, the Intelligence Community, and the ing, and another explores issues that arise in holds wide appeal for the varying interests of National Institutes of Health. He is a Fellow scaling up AI systems to realistic, real-world the members of the AI community. High- of the American Association for the Ad- sizes. Although applications are relatively rare lights include: vancement of Science and the American As- in the program, the conference is collocated with • Three days of technical paper presenta- the Innovative Applications of AI Conference, sociation for Artificial Intelligence. tions by top scientists in the field which has an excellent sampling of such topics. Oliver Selfridge studied mathematics as an • A series of invited speakers and panels, in- Session chairpeople will strive valiantly to undergraduate at the Massachusetts Institute make sure that time remains for questions after cluding the opening keynote address by of Technology, where he met Walter Pitts and every paper. We hope that you will take advan- Oliver Selfridge Warren McCulloch, who had just written the tage of the opportunity to ask incisive questions. • Twenty four-hour tutorials that explore first paper that proposed that neural nets were The resulting dialog can make for a more vi- evolving techniques taught by experienced powerful computational devices. He contin- brant meeting and can help advance the field. scientists and practitioners in AI (separate ued his study of mathematics under Norbert In addition to the refereed papers, we have registration fee) Wiener, and extended it into cybernetics, Wiener’s main concentration at the time. also invited a set of speakers to introduce or sur- • AAAI– / IAAI– Joint Exhibition, fea- In , Selfridge joined MIT’s Lincoln vey exciting areas of AI research. Oliver Self- turing approximately seventy exhibits and Laboratory, where his immediate early AI in- ridge, our keynote speaker, will present his vi- demonstrations sion of research in learning. Other topics include terests were in both Pattern Recognition and • AAAI Robot Exhibition and Competition, learning visual behaviors, a control-theory per- Learning: that is, not merely in how a net- combining a live competition of mobile spective on learning to act, case-based reason- work recognizes a pattern, but also in how it  ing, and distributed AI. In addition, speakers robots from research labs around the gets to be able to do so. In , he participat- will illuminate applications of AI in legal rea- world with video presentations from sev- ed in the Dartmouth Summer Study, often soning, molecular biology, machine translation, eral US robot manufacturers recognized as the beginning of AI. and multimedia. There will also be a survey of • AAAI Arts Exhibition, presenting several In the middle sixties, Selfridge initiated the newly-emerging discipline of artificial life. pieces judged to have both artistic merit Project MAC, the beginning of time-sharing This year we are introducing two new exhi- and a substantial AI component and personal computation, later serving as its bitions in addition to the traditional exhibit • A series of nineteen small workshops with Associate Director for two years. His interest program: a mobile robot exhibition and an AI selected focus. (Attendance is limited and de- and research in learning were continued  art show. The mobile robot exhibition marks termined prior to conference.) strongly in the s at Bolt Beranek and the resurgence of interest in AI in robotics and For more detailed descriptions of the exhi- Newman.  focuses on the interaction of various parts of AI bitions, please turn to page six. The AAAI– In the early s, Selfridge moved to GTE to achieve effective behavior in the real world. Preliminary Program begins on page sixteen. Laboratories, where with Bud Frawley he The AI art exhibition will feature the use of AI Tutorial descriptions begin on page seven. helped to start a considerable effort in ma- in the production of serious works of art in a chine learning, probably the largest in the in- number of media. dustrial world. He is continuing his work as a Paul Rosenbloom & Peter Szolovits Senior Staff Scientist in the Computer and Cochairs, AAAI-92 Program Committee Intelligent Systems Laboratory. IAAI‒ 3 “The biggest and most complete AI event IAAI- Program Special IAAI Plenary Session: of the year.... [Putting AAAI and IAAI to- Japan Watch  Panel—Expert gether] is a clever and natural teaming System Applications & Advanced because people interested in one are — or Paper Presentations should be — interested in the other .... Twenty-one deployed applications will be Knowledge-Based Systems Speakers at this conference offered a whole featured this year at IAAI. Sessions have been Research organized around the following topics: array of applications. And they were not : PM, July ,  • Finance pie-in-the-sky. They are working systems • Customer Service Panelists: Edward Feigenbaum (Chair), Stan- that are increasing productivity, increas- • Industrial ford University; Robert S. Engelmore, Stan- ing service, and saving money right now.” • Data Analysis ford University; Peter Friedland, NASA Ames Don Felice • Software Development Research Center; Bruce Johnson, Andersen The Computer Conference • Regulatory Consulting; Penny Nii, Stanford University; ,  Analysis Newsletter, July . • Routing. Herbert Schorr, USC/Information Sciences Please see page fourteen for the preliminary Institute; and Howard E. Shrobe, Symbolics, program. Inc./Massachusetts Institute of Technology This exciting panel will report the findings Please join us at AI-on-Line of a study, including many site visits in   IAAI– Now entering its third year, AI-on-Line is a March , of Japanese work in • Industrial and commercial applications of The Fourth Annual Conference on Innova- series of vendor-organized issue-oriented expert systems tive Applications of Artificial Intelligence will panels. Short presentations by users of de- showcase the most innovative deployed ap- ployed applications are followed by intensive, • Advanced knowledge based systems re- plications of AI. These applications are win- interactive discussions. This year, the AI-on- search in university and industrial labs. ners of a worldwide competition for the best Line series has been incorporated into the Work of national labs in fuzzy systems, fifth use of AI technology to solve real-world schedule of IAAI, and will provide a balance generation systems, and large knowledge problems. As in earlier years, applications each day to the presentation of the twenty- bases will also be covered. will be described in talks that are accompa- one award-winning IAAI papers. Topics for The study was conducted under the aus-  nied by audiovisual presentations and live include: pices of the Japan Technology Evaluation demonstrations. Meet-the-authors discus- • Technology Transfer/Corporate Adoption Center (JTEC), primarily sponsored by the sions at the end of each session will encour- • Case-Based Reasoning Systems National Science Foundation. age close interaction between the presenters • Neural Networks Due to the widespread appeal of this pan-  and other conference participants. • LISP Applications el, it will also be open to AAAI– attendees. This year the conference will also include • Hidden Experts The panel will be held in the Civic Auditori- AI-on-Line panel discussions that address is- Representatives from the following Ameri- um on Wednesday evening. sues of particular interest to the developers can and international companies and institu-  of AI applications. tions participated in : Accu- IAAI– sessions have been scheduled to rate Automation, American Air- allow participants to attend the plenary ses- lines, Ameritech, Andersen Con- sion of the concurrent National Conference sulting, Bellcore, Boeing, Boston on Artificial Intelligence (AAAI–) and the Children’s Hospital, The Boston joint exhibition. The conference schedule al- Company, CAMPSA, Carnegie so permits attendees to enroll in tutorials or Mellon University, Chase Man- other portions of the National Conference. hattan, Continental Insurance, Please join us for a stimulating and re- Covia Technology, Digital Equip- warding conference! ment, Duke Power, Dun & Brad- Carli Scott street, DuPont, Ford, Frito Lay, IAAI– Program Chair Hughes Aircraft, IBM, Inrets (France), LEGENT, Lockheed, London Life, Manufacturer’s “[The combination of IAAI and the national Hanover Bank, Metropolitan conference makes this] a gathering place not Life, NASA Johnson Space Cen- only for AI researchers, but also for people in ter, NCR, Promised Land Tech- industry who are trying to put AI to work. nologies, Sears, Texaco, Inc., Uni- My impression last year that AI was making versity of California at Berkeley, steady progress as a technology … was and US West. strengthened by this year’s papers at IAAI The AI-on-Line panels are  and AI-on-Line.” open to both AAAI– and  B. Chandrasekaran IAAI– attendees. IEEE Expert, October  The San Jose Convention Center: Site of AAAI–92 and IAAI–92 6 The Exhibit Program

AAAI- & IAAI- Robot Exhibition AAAI Arts Exhibition Joint Exhibition and Competition Intense recent development of multimedia technologies is a sign of the growing use of July –,  This exciting new addition to the exhibition computing in service of art and entertain- will feature a competition among mobile ment. AI will play an essential role in this robot entries from research laboratories and process, as artists and scientists work to de- universities throughout the world. The com- velop machines that have a substantive un- petition will involve three stages, including: derstanding of aesthetics. The AAAI– Arts • House-breaking (roaming in a natural en- Exhibition will be the first forum exclusively vironment) for the display of such systems. • Spatial search (exploration in a natural This first Arts Exhibition will be small, environment) presenting pieces from a range of areas to • Object detection / classification (directed convey the variety of work we wish to en- search in a natural environment). courage and hope to show in future years. At the time of print, the following partici- Each piece will demonstrate the use of con- pants have entered into the competition: crete AI technologies, in the acquisition, rep- Odysseus, Carnegie Mellon University; resentation, or processing of artistic knowl- Dewey, Brown University; Archi, University edge. of Chicago; George descendant, Georgia In- We expect to show approximately six stitute of Technology/Denning Mobile works from among several areas of artistic Robotics; CARMEL, University of Michigan; performance, including still images, anima- The exhibit program will once again include Flakey, SRI International; Soda-Pup, NASA tion, music, drama, expressive speech and approximately seventy exhibits and demon- Johnson Space Center; William, JPL/MIT; natural language generation, aesthetic robot strations by the leading suppliers of AI hard- Flash, MITRE Corp; Beamer, IBM TJ Watson motion, and realtime interactive environ- ware and software, as well as consulting or- Research Center; Flash Zorton, Ecole Poly- ments. ganizations and publishers displaying their technique of Montreal; and SPI- new products or services. Graphic presenta- DER, NIST. Demonstrations of tions by selected exhibitors will be featured Bert and Ernie, MIT and The in the Applications Pavilion. At the time of Flying Fusco Brothers, Georgia publication,  exhibitors include AAAI Institute of Technology will also Press, Ablex Publishing Corporation, Aca- be presented. demic Press, Addison Wesley Publishing In addition to the competi- Company, AICorp, AI Expert, AI Technolo- tion, the exhibition will include gies, AION Corporation, Alcatel-Austria, a video and graphic display of Andersen Consulting, Benjamin/Cummings robots by US manufacturers. Publishing Company, BIM SA/NV, Black- Anticipated participants include board Technology Group, Cambridge Uni- General Motors and Chrysler versity Press, Chestnut Software, Cognitive Corporation. Systems, Covia Technologies, Digital Equip- ment Corporation, Elsevier Science Publish- ers, EXSYS, Inc., Franz, Inc., French Pavilion (a consortium of French companies), Gen- sym Corporation, Harlequin Ltd., IBM, Ibu- ki, IEEE Expert, ImageSoft, Inference Corpo- ration, IntelliCorp, John Wiley & Sons, Kluwer Academic Publishers, Lawrence Erl- baum & Associates, Lockheed Missiles & Space, Inc., Lucid Inc., The MIT Press, Mor- gan Kaufmann Publishers Inc., Neural Ware, Inc., Neuron Data Inc., PC AI, Pergamon Press, Production Systems Technologies, Inc., Quintus Computer Systems Inc., Reduct Sys- tems, Rosh Intelligent Systems, Schwartz As- sociates, Springer-Verlag New York Inc., Symbolics Inc., Talarian Corporation, US Air Force, Venue, and W. H. Freeman and Com- San Jose's Tech Museum of Innovation features an exhibit pany. hall where interactive exhibits put today's technology in your hands. AAAI–92 and IAAI–92 attendees are eligible for a special ticket offer. See page twenty-four for details The  Tutorial Program 7

1992 AAAI Tutorials SA (Sunday,  AM ‒  PM, July ) SA (Sunday,  AM ‒  PM, July ) Knowledge Acquisition Techniques User Modeling & The AAAI tutorial program for 1992 features twenty four-hour tutorials that explore Jan Clayton, Vice President, Expert Support Inc. and User-Adapted Interaction Carli Scott, Principal Engineer, Expert Support Inc. evolving techniques. Each tutorial is taught Sandra Carberry, Associate Professor of Computer Science, by experienced scientists and practitioners in This tutorial focuses on the relationship between University of Delaware and Alfred Kobsa, Associate Profes- sor of Information Science, University of Konstanz AI. A separate registration fee applies to all knowledge acquisition and the development of ex- tutorials (see page twenty-one for fees). pert systems. The most common knowledge-ac- This tutorial will familiarize participants with the quisition activities will be discussed, along with key concepts, techniques, and issues involved in how those activities change as a project progresses. user modeling and user-adapted interaction. We In addition, the presenters will describe guidelines, will first consider the beliefs about the user that • Building Expert Systems in the Real techniques, and procedures that have proven suc- might be captured in a user model, including the World (MP) cessful in developing applications. user’s domain knowledge, goals, and partially de- Instead of discussing the theory behind knowl- veloped plans for accomplishing these goals, and • CASE, Knowledge-Based Systems and edge acquisition, the tutorial presenters will focus attitudes and emotions. Then we will present tech- Object-Oriented Programming (MA) on the practice of knowledge acquisition. The pre- niques for acquiring, representing, and revising senters will spend very little time discussing how to user models and for exploiting them to tailor the • Case-Based Reasoning: Theory and ask questions. Instead, they will focus on what  system’s behavior to the individual user. We will Practice (SP ) questions to ask. In addition, the tutorial will cover critically analyze each approach, discuss its advan- • Computers in Context: Tailoring Expert problems that knowledge engineers are likely to tages and disadvantages, and describe prototypical encounter, how to avoid these problems, and how Systems to Real Workplaces (SP) implementations. We will also examine practical to correct them when they do occur. and ethical considerations related to user modeling, • Constraint-Directed Reasoning (MA) Prerequisite Knowledge: This is an introductory describe existing shell systems for constructing user to intermediate level tutorial. Basic knowledge of models, and provide the participants with practical • Distributed Tools AI and expert systems is required. Development of tips and guidelines for systems development. (MP) a prototype of a small expert system is helpful. Prerequisite Knowledge: This is an intermediate- Knowledge of a particular expert-system shell is level tutorial. Participants should be familiar with • Experimental Methods for Evaluating not needed.  basic concepts in knowledge representation and AI Systems (MP ) Jan Clayton has eleven years of experience in reasoning. Some familiarity with user modeling • Genetic Algorithms & Genetics-based the field of artificial intelligence and expert-sys- will be helpful but is not required. tems technology. Her research and commercial ex- (SA) Sandra Carberry is an associate professor of perience includes: computer science at the University of Delaware • Graphical Interface Design for AI • Application development of small- and large- and has been working in the field of user modeling Applications (SP) scale expert systems since . Her work focuses on incrementally rec- • Documentation and training development ognizing the user’s plans and goals during the • KADS: An Overview of a Structured • Expert-systems methodology development course of an ongoing dialogue and on exploiting Methodology for KBS Development Jan was a member of Stanford’s Heuristic Pro- the acquired knowledge to produce more robust    (SP) gramming Project from to . In she and cooperative interaction with the user. Dr Car- joined Teknowledge Inc. as a knowledge engineer. berry is a member of the User Modeling and User- • Knowledge Acquisition Techniques During her seven years at Teknowledge, Jan partic- Adapted Interaction editorial board. She is also the (SA) ipated heavily in both application and training de- author of Plan Recognition in Natural Language velopment. As such, she worked on several success- Dialogue, and has coauthored several computer • Knowledge-Based Production fully fielded expert systems. In , Jan became a science textbooks. Dr. Carberry received an excel- Management (SP) founder of Expert Support Inc. lence in teaching award from the student ACM Carli Scott has seventeen years of experience in chapter at the University of Delaware. • Machine Learning for Classification the field of artificial intelligence and expert-sys-  Alfred Kobsa is an associate professor of infor- Tasks (MA ) tems technology. Her research and commercial ex- mation science at the University of Konstanz, Ger- • Machine Learning for Planning, perience span a wide range of activities including: many, and has been working in the field of user • Building expert-system applications modeling since . His research has been con- Problem Solving, and Natural Language • Designing, implementing, documenting, test-  cerned both with theoretical issues in user model- (MP ) ing, maintaining, and extending expert-system ing, such as representing and reasoning with user • Natural Language Generation (MP) shells beliefs, and with practical applications, including • Developing and delivering courses about ex- user modeling components for natural-language • Planning and Real-Time Reasoning pert-systems shells and knowledge engineering interfaces to expert systems and the development (SA) methodology of user modeling shell systems. Technical papers by Carli worked for Stanford’s Heuristic Program-  him have appeared at international AI conferences • Robot Architectures (SA ) ming Project from  to  on the MYCIN and and in AI-related journals. • Text Interpretation (MA) ONCOCCIN projects as a knowledge engineer and Dr. Kobsa was the cochairman of the First In- lead programmer, respectively. In , Carli be- ternational Workshop on User Modeling and is the • User Modeling & User-Adapted came founder of Teknowledge Inc., and in , coeditor of the first comprehensive survey of the Interaction (SA) she went to work for Teknowledge on a full-time user modeling field, which appeared in a special is- basis as a senior knowledge engineer, where she sue of Computational Linguistics and a volume in • Verification and Validation of worked on both application and product develop- the Symbolic Computation Series of Springer-Ver- Knowledge-Based Systems (MA) ment. In , she became a founder of Expert lag. He is also the editor of the User Modeling and Support Inc. User-Adapted Interactions journal. 8 Tutorials

SA (Sunday,  AM ‒  PM, July ) SA (Sunday,  AM ‒  PM, July ) SA (Sunday,  AM ‒  PM, July ) Genetic Algorithms & Planning and Real-Time Reasoning Robot Architectures

Genetics-based Machine Learning James Hendler, University of Maryland, and Michael Reid G. Simmons, Research Computer Scientist, Carnegie Georgeff, Director, Australian AI Institute Mellon University and R. James Firby, Assistant Professor, David E. Goldberg, Associate Professor of General Engineer- University of Chicago ing and the Beckman Institute, University of Illinois at During the last few years, AI researchers have be- Urbana-Champaign and John R. Koza, Consulting Associ- come increasingly concerned with the problem of This tutorial will introduce concepts in robot ar- ate Professor of Computer Science, Stanford University designing architectures for intelligent systems that chitectures, concentrating on architectures for au- This tutorial will introduce participants to the are required to operate in complex and dynamic do- tonomous mobile robots. The tutorial will begin ideas and applications of genetic algorithms mains. The potential applications are both diverse by presenting factors that influence modern robot- (GAs)—computer search procedures based on the and economically important—- for example, air ic architectures: actuator and sensor control, com- mechanics of natural genetics and natural selec- traffic control, intelligent networking, monitoring plex changing environments, and the desire to per- tion—and genetics-based machine learning industrial processes, on-line fault diagnosis and form a wide variety of tasks. The tutorial will con- (GBML)—machine learning techniques that use malfunction handling, and emergency health care. tinue with case studies of deliberative (hierarchi- genetic algorithms and their derivatives. GAs and This tutorial aims to provide participants with cal), reactive (behavior-based), hybrid and GBML are receiving increased attention in practi- an understanding of the issues involved in designing learning architectures. The strengths and weak- cal yet difficult search and machine learning prob- and building such intelligent agents. It will briefly nesses of each methodology will be analyzed in lems across a spectrum of disciplines. We review review past work in the design of AI planning sys- terms of the types of tasks and environments it the mechanics of a simple genetic algorithm and tems and discuss why these are inadequate to the readily supports. Particular attention will be paid consider the implicit parallelism that underlies its sorts of dynamic domains described above. Descrip- to the problems of dealing with uncertainty, com- power. A parade of current search applications is tion of new approaches and case studies of both toy- plexity, and resource constraints. Finally, the tuto- reviewed as are more advanced GA techniques world and real-world systems will be presented. rial will discuss specific architectural designs for such as niching and messy GAs. The two most Prerequisite knowledge: This intermediate-level the AAAI– robot competition. prominent techniques of GBML, classifier systems tutorial is aimed at both the industrial AI practi- Prerequisite Knowledge: This intermediate-level and genetic programming, are also surveyed. tioner interested in the development of planning- tutorial is intended for people who need to develop Prerequisite knowledge: This tutorial level is in- related systems and the AI researcher interested in or evaluate complex robotic systems. No previous troductory. Knowledge of genetic algorithms or bi- learning about current research in the planning experience with robotics is required but a basic un- ological concepts is not assumed. A general famil- area. The presenters will assume a background in derstanding of AI planning concepts is essential. iarity with computers and programming is re- AI, academic or industrial, but only a basic famil- Reid G. Simmons is a research scientist in the quired. iarity with planning research. School of Computer Science and Robotics Institute David E. Goldberg is an associate professor of James Hendler is an assistant professor in the at Carnegie Mellon University. Since coming to general engineering and the Beckman Institute at Department of Computer Science at the University CMU in , Dr. Simmons’s research has focused the University of Illinois at Urbana-Champaign. He of Maryland. He is the author of Integrating Mark- on developing self-reliant robots that can au- holds a Ph.D. from the University of Michigan and er-passing and Problem-Solving and is coeditor of tonomously operate over extended periods of time has written papers on the application and founda- Readings in Planning. Dr. Hendler has also done a in unknown, unstructured environments. This tions of genetic algorithms. His book, Genetic Algo- large amount of nonuniversity teaching in the area work involves issues of robot architectures that rithms in Search, Optimization, and Machine Learn- of AI as well as teaching management courses on combine deliberative and reactive control, selective ing (Addison-Wesley, ) is widely used and his the development of expert systems in the US and perception, and robust error detection and recov- recent studies have considered the theory of decep- abroad. His main area of research is in the design ery. The ideas are currently being applied to a six- tion, the role of noise in GA convergence, and the of situated planning agents. legged planetary rover and an indoor mobile ma- theory and development of messy GAs. Michael Georgeff is the director of the Aus- nipulator. John R. Koza is a consulting associate professor tralian Artificial Intelligence Institute, Australia’s Dr. Simmons received his Ph.D. from MIT in of computer science at Stanford University. He re- leading research and development organization for AI in . His dissertation focused on the combi- ceived his Ph.D in computer science from the Uni- advanced information technology. He is also a nation of associational and causal reasoning for versity of Michigan in the field of machine learn- member of the Artificial Intelligence Center at SRI planning and interpretation tasks. The research in- ing and induction in . He is currently investi- International, Menlo Park, California, one of the volved the use of multiple representations of the gating the artificial breeding of computer pro- world’s foremost research centers in artificial intel- physical world and developed a domain-indepen- grams and has recently completed Genetic ligence. He is coeditor of Reasoning about Actions dent theory of debugging plans. Programming (MIT Press) that surveys these ef- and Plans and was a senior member of the SRI R. James Firby is an assistant professor of com- forts. Between  and  he was chief executive team that developed the research and development puter science at the University of Chicago. His officer of Scientific Games Incorporated in At- plans for NASA’s proposed space station. main interest is the construction of systems that lanta, and he is currently a principal in Third Mil- interact with complex, changing environments. Dr. lennium Venture Capital Limited in California. Firby is currently working on an architecture to in- tegrate symbolic planning with low-level robot control. That architecture is being implemented on an indoor mobile robot. Dr. Firby received his Ph.D. in AI from Yale University in . His dissertation introduced the RAP system for reactive plan execution using situ- ation specific task methods. That research ad- dressed problems in error recovery, task-directed perception, and opportunism. From  to  Dr. Firby worked as a research scientist at the Jet Propulsion Lab designing and building local navi- gation software for the Pathfinder Planetary Rover project. Tutorials 9

SP (Sunday,  ‒  PM, July ) SP (Sunday,  ‒  PM, July ) SP (Sunday,  ‒  PM, July ) KADS: An Overview of a Structured Graphical Interface Design for Computers in Context: Tailoring Expert Methodology for KBS Development AI Applications Systems to Real Workplaces

R. A. Martil, AI Coordinator, Lloyd’s Register, UK and Michael D. Williams, Chief Scientist, IntelliCorp William J. Clancey, Senior Research Scientist, Institute for B. J. Wielinga, Professor, Social Sciences, University of Research on Learning, and John McDermott, Technical Amsterdam This tutorial surveys graphical interface design for Director, Artificial Intelligence Technology Center, Digital AI applications. A number of paradigmatic exam- Equipment Corporation This tutorial will present a practical overview of ples are examined and analyzed in order to high- the KADS methodology, the result of an ongoing This tutorial introduces software developers and light successful as well as unsuccessful techniques. managers to business analysis methods that are series of European Community-funded ESPRIT The first half of the tutorial presents a history of technology projects. The tutorial will concentrate sensitive to how people think and learn from each key interface developments and summarizes impli- other. How can we design computer systems that fit on two key areas in KAD, knowledge modeling and cations of selected current research. The second project management. The presenters will then the social realities of the workplace? How can we half of the tutorial focuses on practical guidelines design tools that deal not only with routine, well- bring attendees up to date on developments within for managing the development of AI systems with the latest ESPRIT project, KADS-II. The material will understood situations, but help people innovate substantial interface components. Based on experi- and change how their work is done? Knowledge en- be presented through a series of exercises support- ence with a variety of applications, potential pitfalls ed by lectures, intended to provide a more practi- gineering will come of age only when it comes to are identified and useful techniques are described. grips with its potential intrusiveness. New methods cal understanding of the methodology than the ex- Detailed examples are used to help characterize the tensive published literature can provide. The mate- of observation, user collaboration, and evolution- process and craft of effective interface design. ary design—based on ethnography and video inter- rial for the tutorial has been developed from an in- Prerequisite Knowledge: This beginning/inter- dustrial two-day course presented by Professor action analysis—are changing expert systems devel- mediate-level tutorial presupposes a basic under- opment. Building our lectures around case studies, Wielinga and Mr. Martil, who are both involved in standing of knowledge representation and object the KADS-II project. we illustrate how to usefully integrate and learn oriented programming. from these new perspectives on software design, Prerequisite Knowledge: This intermediate-level Mike Williams is chief scientist at IntelliCorp. course is aimed toward those wishing to develop, business practices, and human cognition. He has had over fifteen years experience building Prerequisite Knowledge: This is an intermediate- or manage the development of, knowledge-based AI and object-oriented systems and tools. He was a systems in a disciplined way. Anyone with some level tutorial. Familiarity with knowledge-based principal in the development of STEAMER, an early systems is helpful; knowledge of social sciences is experience of developing expert systems, or who AI-based computer training system to teach steam has been involved in knowledge acquisition with not required. plant operation; RABBIT, an intelligent database in- William J. Clancey is a senior research scientist an expert will certainly understand the issues and terface; and a number of knowledge engineering the material presented. at the Institute for Research on Learning (IRL). At tools, including KEE, KEEconnection, SimKit, Ac- Bob Wielinga has lectured at the University of IRL, Dr. Clancey works with social scientists to un- tiveImages, and IntelliScope. Most recently he was derstand their perspective on software design and Amsterdam on knowledge-based systems for more one of the principals in the design and develop- than ten years, and prior to that acted as a Senior relate it to qualitative modeling techniques. His ap- ment of ProKappa, an object-oriented develop- proach includes relating AI programming to tradi- Research Officer in AI at the University of Essex ment environment for commercial software appli-   tional engineering modeling, facilitating commu- (UK) from to . He became a full professor cations. Prior to joining IntelliCorp he was a mem- at the Social Science Informatics Department in nication instead of merely automating what people ber of the technical staff at Xerox PARC. . Besides teaching, his duties include research do, and reconceiving how models of knowledge re- in AI and cognitive science. Currently he is project late to human reasoning. Before joining the IRL in leader of externally funded research projects on ac- , Dr. Clancey was a member of the Stanford quisition of problem solving skills, models of Knowledge Systems Lab for thirteen years, where learning processes, and intelligent coach for teach- he developed NEOMYCIN, one of the earliest sec- ing physics. Since , he has performed research ond-generation expert systems. on the methodology of knowledge-based system Dr. Clancey received his Ph.D. from Stanford acquisition, analysis and design. He was team lead- University in Computer Science in . He is the er for Uv A of the KADS project, and now leads author of three books and over thirty-five journal ar- work in KADS-II and ACKnowledge. As such he is a ticles and book chapters, a senior editor of Cognitive major contributor to the development of the KADS- Science, and editor-in-chief of the AAAI Press. II methodology. He is action coordinator of the ES- John McDermott is a member of the technical PRIT REFLECT basic research project. staff at Digital Equipment Corporation and is also a Robert Martil is technical R & D manager in AI faculty member of the Computer Science Depart- at Lloyd’s Register, UK. He is project manager for ment at Carnegie-Mellon University. Dr. McDer- the KADS-II project and Coordinator of all LR’s AI mott was a cofounder of Carnegie Group, Inc. He work. His technical work on the KADS-II has been received a Ph.D. in philosophy from the University primarily concerned with the development of the of Notre Dame in . His AI research interests are KADS-II life cycle model and project management in the application of AI techniques to industrial cycle, and with the development of the complete problems. This has included addressing the prob-  KADS-II model set, a focus for KADS-II work. He ini- lems involved in developing expert systems. In tiated and organizes an ongoing series of industrial he completed work on R1, a program used by DEC  courses on KADS copresented with Professor to configure computer systems. Since , he has Wielinga, and is involved in the organization of the built on the earlier knowledge acquisition work, KBS methodologies and KADS user groups in the UK but focused it on creating software tools to make and Europe. the development of effective application programs Mr. Martil gained a degree in computer science easier. His experiences have made him acutely from Warwick University, England, and an M.Sc. aware of the importance of workplace analysis for in intelligent systems at Brunel University. developing successful expert systems. 10 Tutorials

SP (Sunday,  ‒  PM, July ) SP (Sunday,  ‒  PM, July ) MA (Monday,  AM ‒  PM, July ) Knowledge-Based Case-Based Reasoning: Machine Learning for Production Management Theory and Practice Classification Tasks

Stephen F. Smith, The Robotics Institute, Carnegie Mellon Kevin D. Ashley, University of Pittsburgh and Evangelos Haym Hirsh, Rutgers University and Jude Shavlik, University and Norman M. Sadeh, The Robotics Institute, Simoudis, Lockheed AI Center University of Wisconsin Carnegie Mellon University Case-based reasoning (CBR) is a new AI method- After an initial survey of the field of machine This tutorial will introduce participants to the ology that applies past experience, as represented learning, the primary focus of this tutorial will be concepts, techniques, and methodologies that have by prior cases, to current decision making. CBR on symbolic and neural network approaches to in- emerged from work in knowledge-based produc- systems have been designed to handle such diverse ductive learning from examples. This problem can tion management. We will first consider the short- tasks as planning, design, diagnosis, argumenta- be briefly defined as follows: given descriptions of comings of traditional approaches to production tion, negotiation and running a help desk. CBR re- a set of examples each labeled as belonging to a management (e.g. MRP / MRP II) and identify op- quires a memory where past cases are organized. particular class, determine a procedure for correct- portunities provided by knowledge-based tech- Successful cases are stored so that they can be re- ly assigning new examples to these classes. Both nologies both in overcoming these limitations, and trieved and used in similar situations. Failures are symbolic and neural-network approaches to this in contributing to effective implementation of also stored so that they can warn the problem problem will be presented. Quinlan’s ID, Michals- modern manufacturing philosophies (e.g. Just in solver of potential difficulties and provide repairs. ki’s AQ, and Rumelhart’s Backpropagation algo- Time). We will then review in more detail the es- This tutorial will introduce the basics of CBR, il- rithms will be described and illustrated with sim- sential concepts and techniques underlying domi- lustrate and compare five paradigmatic CBR ap- ple examples. These algorithms have been applied nant approaches to knowledge based production proaches, and examine design issues and tradeoffs to real-world tasks; the approaches taken and the management, including object-centered modeling in building CBR systems. As a major new compo- results obtained will be covered. The tutorial will frameworks, simulation and rule-based tech- nent, the tutorial will focus on recent real-world also describe some recent comparisons between niques, temporal constraint management, black- applications of CBR technology and lessons ID and Backpropagation using real-world data board and multi-perspective techniques, con- learned in transferring this technology. sets. Finally, it will discuss the strengths and weak- strained heuristic search, uncertainty manage- Prerequisite Knowledge: No special prerequisites, nesses of this form of machine learning, as well as ment, iterative improvement techniques, distribut- but familiarity with basic AI concepts helpful. This current research problems in the area. ed production management and Leitstand intermediate-level tutorial is intended for AI pro- Prerequisite Knowledge: This intermediate-level frameworks. Finally, we will examine a few suc- fessionals interested in current CBR research is- tutorial is addressed to computer scientists with cessful applications, and assess the current state of sues, managers interested in alternative method- some experience in AI, roughly at the level of an theory and practice. ologies for expert system development, and knowl- introductory textbook. A basic understanding of Prerequisite Knowledge: This intermediate-level edge engineers considering CBR applications. rule-based expert systems, particularly the con- tutorial is aimed at anyone interested in applying Kevin Ashley is an assistant professor of law and cepts of pattern matching and backward/forward knowledge-based techniques to practical produc- intelligent systems at the University of Pittsburgh chaining, is necessary. tion management problems. The tutorial assumes and a research scientist at the Learning Research Haym Hirsh is an assistant professor of com- knowledge of AI at the level of an introductory and Development Center. His has written Model- puter science at Rutgers University. He received his course as well as some familiarity with basic pro- ing Legal Argument: Reasoning with Cases and Hy- Ph.D. () degree in computer science from duction management concepts. potheticals, (The MIT Press/Bradford Books). His Stanford University, and spent a year as a visiting Stephen F. Smith is a senior research scientist in research interests include case-based and analogi- scholar at Carnegie Mellon University and the Uni- the Robotics Institute at Carnegie Mellon Univer- cal reasoning, argumentation and explanation and versity of Pittsburgh. He is the author of Incremen- sity, and Director of the Production Scheduling designing computer systems for assisting attorneys tal Version-Space Merging: A General Framework and Control Laboratory within the Robotics Insti- in law teaching and practice. He received a J.D. for Concept Learning and of a number of book tute’s Center for Integrated Manufacturing Deci- from Harvard Law School in 1976. chapters and papers in machine learning. His cur- sion Systems. He received his Ph.D. degree in Evangelos Simoudis is a research scientist in the rent machine-learning research interests include Computer Science from the University of Pitts- AI center of Lockheed corporation where he heads applications of machine learning in both molecu- burgh in 1980. His current research focuses on the Learning Systems Group. He directs research lar biology and ship design, as well as computa- mechanisms for reactive schedule/plan adaptation on case-based reasoning and inductive learning in tional issues for inductive learning. under changing circumstances, integration of manufacturing settings. He is also conducting re- Jude Shavlik is an assistant professor of comput- planning and scheduling processes, distributed co- search in the area of distributed AI, applying it to er science at the University of Wisconsin. He re- ordination in resource-constrained multi-agent help desks. Dr. Simoudis received a Ph.D. in Com- ceived B.S. degrees from MIT in electrical engi- domains, and automated acquisition of scheduling puter Science from Brandeis University in 1991. neering and biology in , a S.M. degree in and planning strategies from experience. His Ph.D. research resulted in the development of molecular biophysics and biochemistry in  Norman M. Sadeh is a research scientist at CASCADE and COAST, two case-based diagnostic sys- from Yale, and, after working for the Mitre Corpo- Carnegie Mellon’s Robotics Institute, and a mem- tems that can be used in help desk settings where ration for several years, was awarded a Ph.D in ber of the Center for Integrated Manufacturing they reason from past cases captured by help desk computer science from the University of Illinois. Decision Systems, where he directs a large project engineers. He is the author of Extending Explanation-Based in decentralized production scheduling and con- Learning by Generalizing the Structure of Explana- trol. Dr. Sadeh is also the developer of the MICRO- tions, coeditor of Readings in Machine Learning, BOSS factory scheduling system, a system that has and has published over two-dozen papers on ma- yielded important reductions in scheduling costs chine learning. His current research interests in- (both tardiness and inventory costs) over more clude comparing and combining symbolic and traditional bottleneck-centered approaches to fac- neural network approaches to machine learning, as tory scheduling. well as the application of machine learning tech- Dr. Sadeh’s research interests include the appli- niques to problems in the Human Genome Pro- cations of AI and operations research techniques ject. He is an invited speaker at the  Interna- to both engineering and management problems. tional Machine Learning Conference. Tutorials 11

MA (Monday,  AM ‒  PM, July ) MA (Monday,  AM ‒  PM, July ) MA (Monday,  AM ‒  PM, July ) CASE, Knowledge-Based Systems and Text Interpretation Constraint-Directed Reasoning

Object-Oriented Programming Jerry R. Hobbs, SRI International and Lisa Rau, General Bernard A. Nadel, Computer Science Dept., Wayne State Electric Research Center University and D. Navinchandra, Robotics Institute, School Jan Aikins, Vice President of Technology, AION Corporation of Computer Science, Carnegie Mellon University and Paul Harmon, Editor, Object-Oriented Strategies Text interpretation is that branch of natural lan- newsletter guage processing that concerns itself with analyz- Constraint satisfaction techniques are central to This tutorial will introduce participants to the ing natural language texts to extract some of the many areas of AI. This tutorial will survey these concepts underlying computer-aided software en- information they convey. Applications of this tech- techniques and their applications. The constraint satisfaction problem (CSP) and several of its basic gineering (CASE), knowledge-based systems, and nology include message routing and prioritizing, algorithms will be introduced, such as backtrack- object-oriented programming (OOP). We will con- automatic database entry, and text retrieval. This sider the advantages of each approach, and pro- field has made significant advances recently. We ing, backjumping, forward checking and network pose possible ways of combining them for system will cover the principal techniques that have been consistency techniques. Heuristics will be given for ordering the search performed by these algo- development. We will review several popular CASE developed. Among the topics covered will be pat- tools, object-oriented tools, and knowledge-based tern identification, statistical methods, syntax and rithms. Applications of CSP will be used to illus- system tools, and discuss the experiences of some parsing methods in relation to the particular prob- trate advanced concepts. These will include areas companies that have used both conventional and lems of naturally-occurring texts, methods for such as temporal reasoning, layout problems, engi- neering design and concurrent engineering. These intelligent CASE products. We will also provide par- making processing faster and more robust, reason- applications will be used to motivate some impor- ticipants with an overview of the current CASE ing about the content of texts for disambiguation, market and an understanding of the evolving role linking different parts of the text together and is- tant extensions of the standard version of CSP that object-oriented and knowledge-based systems sues in evaluating the performance of systems. The such as the integration of continuous-valued vari- techniques are playing within the software devel- discussions will be grounded in the experience of ables and hierarchically-structured domains of val- opment market. the instructors with large-scale text-interpretation ues as well as versions of CSP that require all solu- Prerequisite Knowledge: This is an introductory- systems. tions, any single solution, or the optimum solu- level tutorial. Familiarity with knowledge-based Prerequisite knowledge: This intermediate-level tion. We will also treat the use of constrained-in- systems is helpful; knowledge of object-oriented tutorial presumes familiarity with the basic con- fluence graphs to guide problem solving processes and the management of large constraint networks programming or CASE is not required. cepts and terminology involved in natural lan- Paul Harmon. is the editor of two newsletters, guage processing. in organizations with multi-agent interaction and Object-Oriented Strategies and Intelligent Software Jerry R. Hobbs received his Ph.D. degree from constraint negotiation. The CSP formulations of Strategies. The former reports on commercial de- New York University and has taught at Yale Uni- many of our application examples will be made velopments in object-oriented technologies while versity and the City University of New York. He concrete by the use of explicit, compact Prolog the latter reports on the commercial developments has been at SRI International since , where he programs for their solution. in the field of AI and expert systems, neural net- led the development of the Tacitus system for text Prerequisite Knowledge: Some familiarity with the use of constraints in AI systems would be help- works and intelligent CASE. interpretation. He has done extensive research in He is also the coauthor of Expert Systems: Arti- discourse analysis, knowledge representation, lexi- ful. This is an intermediate-level tutorial. ficial Intelligence in Business and two other books cal semantics, syntax, and parsing. Bernard Nadel is an assistant professor of com- on expert systems. His latest book, ObjectCraft, Lisa Rau, since joining GE in , has initiated puter science at Wayne State University, Detroit. His interests include the design and analysis of al- (Addison-Wesley) provides an overview of OOP and led GE’s research in text interpretation. Ms. and is a manual for the graphical programming Rau is the principal designer of the SCISOR proto- gorithms in general, with constraint satisfaction al- tool of the same name. Mr. Harmon is the CEO of type and the GE NLToolset, deployed in applica- gorithms being his main focus. He is also interest- ObjectCraft, Inc. tions across the company. Prior to joining GE, she ed in the application of constraint satisfaction Jan Aikins is a founder and vice president of received her BS and MS in computer science from techniques to engineering, manufacturing and technology for Aion, a knowledge-based system’s the University of California at Berkeley. She has physics problems. Recent applications of his in- tool vendor for IBM PC and mainframe tools. At authored over thirty publications in the areas of clude the automated design of automobile power transmissions, with Ford Motor Company, and the Aion she is responsible for assessing technological natural language, text and conceptual information developments in AI and related fields. She directs retrieval. manufacture of synthetic diamond, with the advanced development projects and provides tech- In addition to leading research, Ms. Rau has di- Wayne State University Institute for Manufactur- nological and competitive information to Aion’s rected the transfer of technology in text interpreta- ing Research. marketing and sales departments. tion from the laboratory to GE businesses. Her D. Navinchandra is an assistant professor at the Prior to joining Aion, Dr. Aikins held research work has been featured in Popular Science and In- Robotics Institute, Carnegie Mellon University. He positions at both Hewlett-Packard and IBM where formation Week and was cited recently in More Fu- is also an adjunct faculty member of the Depart- she gained extensive experience in creating expert ture Stuff as “an invention that will change your life ment of Civil Engineering at CMU. His research systems for both scientific and commercial appli- by the year .” interests include knowledge representation, design cations. theories, project management, and the application Dr. Aikins received her Ph.D. from Stanford of AI to engineering problems. He has published a University in AI in . Her dissertation intro- book and several papers in these areas. Dr. Navin- duced one of the earliest hybrid knowledge repre- chandra is originator of the concept of Green Engi- sentations and explored the impact of an explicit neering: the study of product design for environ- control representation for expert systems. mental friendliness without compromising prod- uct quality. Dr. Navinchandra cofounded the Intel- ligent Engineering Systems Laboratory at MIT in . He heads up the Concurrent Engineering and KAD laboratory at the Robotics Institute. 12 Tutorials

MA (Monday,  AM ‒  PM, July ) MP (Monday,  ‒  PM, July ) MP (Monday,  ‒  PM, July ) Verification and Validation of Machine Learning for Planning, Prob- Building Expert Systems in the Real Knowledge-Based Systems lem Solving, and Natural Language World

Daniel E. O’Leary, Associate Professor, University of Pat Langley, Senior Scientist, NASA Ames Research Center Tod Hayes Loofbourrow, President and CEO, Foundation Southern California and Kirstie Bellman, The Aerospace and Raymond Mooney, Assistant Professor, University of Technologies, Inc. and Ed Mahler, Program Manager for Ar- Corporation Texas at Austin tificial Intelligence, DuPont This tutorial will survey the concepts underlying This tutorial will present an overview of machine This tutorial is designed to provide participants the verification and validation of knowledge-based learning methods for problem solving, planning, with an understanding of how companies which systems. We will present a hierarchical approach to control, and natural language. Historically, re- have been most successful in applying knowledge- expert systems quality that includes verification search in machine learning has focused on based systems technology have organized, per- and validation. Techniques of verification of rules, acquiring knowledge for classification. Although formed, and managed their activities. The tutorial frames and other forms of knowledge representa- classification has many applications, such as diag- will give participants a look behind the technology tion are investigated. Considerable attention is nosis, it is not directly relevant to many domains. at the organizational steps taken by corporate and then aimed at structuring, designing, and imple- Consequently, there has been a recent growth of divisional managers, project managers, knowledge menting validation tests. Statistical approaches to interest in learning for alternative tasks, although engineers, functional specialists, and data process- verification and validation are summarized in a much of this research builds on the earlier work in ing professionals to successfully build integrated case-study setting. Recent developments in verifi- classification. We will review basic learning meth- knowledge-based systems, and to successfully cation and validation tools are also presented. The ods for the acquisition of search heuristics, primi- manage knowledge-based systems projects. tutorial concludes with a review of what is actually tive operators, and macrooperators drawing on Participants should leave the tutorial with an done in practice. empirical, explanation-based, genetic, analogical, understanding of: the key factors which have led Prerequisite Knowledge: This is an introductory and connectionist approaches to these tasks. We organizations to success in developing integrated to intermediate-level tutorial, requiring some fa- will also discuss the application of such methods to knowledge-based systems programs; the strategic miliarity with knowledge-based systems and dif- various problems, such as robot planning, natural choices facing individuals and organizations ferent forms of knowledge representations, such as language parsing, and plan recognition. charged with building knowledge-based systems, rules and frames. The tutorial assumes an interest Prerequisite Knowledge: Participants in this in- and a set of concrete steps that they can take to im- in the evaluation process. This tutorial is designed termediate-level tutorial should be familiar with prove their ability to successfully develop knowl- for the systems developer or manager interested in basic issues and methods in artificial intelligence, edge-based systems. The tutorial will stress diverse some of the primary techniques and approaches to particularly knowledge representation, search, corporate and government examples, and will verification and validation, and quality manage- planning, and natural language. Some knowledge make use of numerous case studies. ment. of learning methods for classification would be Prerequisite Knowledge: No prerequisites are re- Daniel E. O’Leary is an associate professor at useful but is not necessary. quired or assumed in this introductory tutorial, al- the University of Southern California. He received Pat Langley is a senior scientist at NASA Ames though familiarity with knowledge-based systems his Ph.D. from Case Western Reserve University Research Center, where he carries out research on is helpful. and his Master’s degree from the University of machine learning and intelligent agents. Before Tod Hayes Loofbourrow is president and CEO of Michigan. Professor O’Leary has published papers coming to NASA, Dr. Langley was an associate Foundation Technologies, Inc., a leading knowl- in Expert Systems with Applications, IEEE Expert, professor of computer science at the University of edge technology consulting firm. He has per- International Journal of Expert Systems: Research California, Irvine, and a research scientist at formed strategic consulting for clients worldwide and Applications, International Journal of Intelli- Carnegie Mellon University, where he received his and teaches graduate courses in AI at Harvard gent Systems, International Journal of Man-Ma- Ph.D. in cognitive psychology. He has taught nu- University, where he created the university’s first chine Studies and elsewhere. In , Dan was the merous courses in AI, including graduate seminars courses on expert systems. Mr. Loofbourrow is au- chairperson of the AAAI Workshop on Verifica- and tutorials on machine learning. Dr. Langley has thor of the “Managing Knowledge” column in Ex- tion, Validation and Testing of Knowledge-Based published papers on scientific discovery, concept pert Systems magazine, is a featured columnist for Systems. formation, heuristics learning, motor learning, the Software Engineering journal, and served as a Kirstie Bellman is a research scientist with The and language acquisition. He is coauthor of Scien- columnist for Globetech Magazine. Aerospace Corporation. In , Dr. Bellman was tific Discovery and coeditor of Production System Ed Mahler is currently program manager for AI the chairperson at the IJCAI Workshop on Verifi- Models of Learning and Development, Computation at DuPont with responsibility for leading their im- cation, Validation and Testing of Knowledge-Based Models of Scientific Discovery and Theory Forma- plementation program worldwide. Widely known Systems. Dr. Bellman has published papers in a tion, and Concept Formation: Knowledge and Expe- for his highly successful no nonsense business-ori- number of journals and proceedings, including rience in Unsupervised Learning. ented approach, he has frequently been quoted in Expert Systems with Applications, Proceedings of the Raymond Mooney is an assistant professor in such diverse publications as Time, Computerworld, Space Quality Conference and Proceedings on AI the Department of Computer Sciences at the Uni- and EDP Analyzer. His program at DuPont won and Simulation. versity of Texas at Austin. He was educated at the the award for innovation from High Technology University of Illinois at Urbana-Champaign where magazine in 1987. he earned a Ph.D. in computer science in . His Ed received his B.S. and Ph.D. degrees in chem- dissertation was published in A General Explana- ical engineering from the University of Texas. He tion-Based Learning Mechanism and Its Application worked for a small chemical company before join- to Narrative Understanding. He is author of over ing DuPont in 1969. He held numerous manageri- twenty-five papers in machine learning and was al positions in research and manufacturing prior to cochair of the Seventh International Conference a five year assignment in Corporate strategic plan- on Machine Learning. His recent research has con- ning. While serving as liaison for emerging tech- cerned comparing symbolic and neural-network nologies, he founded an ad hoc group, the Corpo- learning, combining explanation-based and empir- rate Artificial Intelligence Task Force, to exploit AI ical learning, abductive reasoning, and knowledge- in DuPont. This activity precipitated the creation base and theory refinement. of his current position in . Tutorials 13

MP (Monday, ‒  PM, July ) MP (Monday,  ‒  PM, July ) MP (Monday,  ‒  PM, July ) Natural Language Generation Experimental Methods for Distributed Artificial Intelligence Tools

Kathleen F. McCoy, Associate Professor, Department of Evaluating AI Systems Edmund H. Durfee, Assistant Professor, University of Computer and Information Sciences, University of Delaware Michigan and Katia P. Sycara, Research Scientist, Carnegie and Johanna D. Moore, Assistant Professor, Department of Paul Cohen, Associate Professor, Computer Science, Univer- Mellon University Computer Science and Learning Research and Development sity of Massachusetts at Amherst and Bruce Porter, Associate Center, University of Pittsburgh Professor, Computer Science, University of Texas at Austin This tutorial will be a thorough survey of prob- This tutorial will introduce, through case studies, lems, techniques and applications in contemporary This tutorial is intended to provide an introduc- DAI, in preparation for building DAI systems or as tion to natural language generation for those who designs for exploratory and confirmatory experi- ments, a few statistical techniques for analyzing re- background for doing research to advance the state wish to undertake research in this field, and to of DAI practice. Particular emphasis will be given serve as a practical decision-making tool for prod- sults, and techniques for evaluating what results mean to research and development programs. Our to DAI tools and to methodologies for building uct developers who have applications that could and evaluating DAI systems. Real-world applica- benefit from this technology. We will introduce the objective is to provide methods by which re- searchers and practitioners can test hypotheses and tions of DAI are only beginning to emerge, and an field's theoretical problems as well as some solu- objective of this tutorial is to speed this process tions, and will also describe well-established tech- substantiate claims of system performance. We will focus on experiments in knowledge-based systems, along by introducing DAI algorithms and niques that can and are being used in real applica- paradigms to attendees in practical, rather than tions. We will examine several general approaches machine learning and planning, although much of our discussion will apply to broader areas of AI. only theoretical, terms. to generation and the particular tasks to which Prerequisite Knowledge: The tutorial presumes these approaches can be applied. We will then delve The case studies illustrate techniques for measur- ing and comparing levels of performance, analyz- knowledge of AI at the level of an introductory into three important application areas: Report gen- course, and familiarity with such general concepts eration, expert system explanation generation, and ing the effects of adding knowledge to a system, finding interactions between components of sys- as object-oriented systems, planning, heuristic tutorial dialogues. Finally, we will characterize the search, knowledge-based systems, reasoning under needs of several different classes of applications and tems, and isolating causes of poor performance. We will discuss some tricky problems, including uncertainty, and so on. also identify the capabilities offered by different Edmund H. Durfee received his Ph.D. degree in generation techniques in order to enable attendees designing representative test sets and getting repre- sentative samples of data; and we will describe computer and information science from the Uni- to choose a technique suited to their needs. versity of Massachusetts, Amherst, in . His Prerequisite Knowledge: Familiarity with basic AI some open problems, for which convincing tech- niques are not yet available, such as generalizing Ph.D. research developed an approach for plan- concepts, techniques, and algorithms such as search ning coordinated actions and interactions in a net- techniques, early work in planning (STRIPS and results from one system to others. Prerequisite Knowledge: Some familiarity with work of distributed AI problem-solving systems. NOAH), pattern matching, unification, production He is currently an assistant professor in the De- systems, and deductive retrieval is assumed. No knowledge-based systems and basic machine learning and planning techniques is helpful but partment of Electrical Engineering and Computer background in linguistics or computational linguis- Science at the University of Michigan, where his tics is required in this introductory tutorial. not essential. No knowledge of statistics is as- sumed. This is an introductory-level tutorial. interests are in distributed artificial intelligence, Kathleen F. McCoy is an associate professor in planning, blackboard systems, and real-time prob- the Department of Computer and Information Paul Cohen is an associate professor at the Uni- versity of Massachusetts at Amherst, where he di- lem solving. He is author of Coordination of Dis- Sciences at the University of Delaware where she tributed Problem Solvers (Kluwer Academic Press) holds a joint appointment in the Department of rects the Experimental Knowledge Systems Lab. His research concerns planning in uncertain, real- and received a Presidential Young Investigator Linguistics. She has published articles in the area award from the National Science Foundation. of natural language generation in the Artificial In- time environments. Cohen has been involved in the development of research methods for AI and Katia P. Sycara is a research scientist in the telligence, Computational Linguistics, and Compu- School of Computer Science at Carnegie Mellon tational Intelligence journals as well as in confer- computer science in general for several years, and in 1990 published a critical survey of AI research University. She is also the director of the Laborato- ence proceedings. Her current interest is in com- ry for Enterprise Integration. She is directing and bining natural language processing techniques papers, and suggested a new methodology. Dr. Cohen received his Ph.D. from Stanford Uni- conducting research in investigating and integrat- with the field of augmentative communication in ing decision making across the manufacturing order to develop technology to enhance the com- versity, and joined the faculty at the University of Massachusetts in 1983. He is an editor of The product life cycle. She is currently involved in re- municative capabilities of people with various search on integrating planning, scheduling and Handbook of Artificial Intelligence, volumes III and kinds of disabilities. case-based learning, developing an information in- Dr. Johanna D. Moore holds interdisciplinary IV. Bruce Porter is an associate professor of com- frastructure for manufacturing decision making, appointments as an assistant professor of comput- and concurrent engineering design. er science and as a research scientist at the Learn- puter sciences at the University of Texas at Austin. His current research is developing large knowledge She holds a Ph.D. in computer science from Geor- ing Research and Development Center at the Uni- gia Institute of Technology. Her Ph.D. dissertation versity of Pittsburgh. Her research interests are to bases and methods for generating coherent expla- nations to answer questions using them. His earlier research resulted in the development of an ap- improve human-computer interaction through the proach for multi-agent conflict resolution and ne- research with Ray Bareiss produced the PROTOS use of explanation and natural language genera- gotiation that integrates case-based reasoning and tion. Dr. Moore has worked extensively in the area knowledge acquisition system and a comprehen- sive evaluation of its performance. Dr. Porter has decision theoretic techniques. While continuing of explanation for expert systems and is now ex- work on case-based reasoning, she has investigated tending this work to providing explanation capa- taught many short courses on AI, machine learn- ing, and expert systems, and he has consulted with extensively distributed factory scheduling, fusing bilities for information-giving and intelligent tu- distributed expertise, and distributed decision sup- toring systems. numerous companies on AI projects. Dr. Porter received his Ph.D. from the Universi- port for enterprise integration. She has published Dr. Moore has written and lectured extensively extensively in these areas and delivered seminars in on the topic explanation for expert systems, and ty of California at Irvine in 1984 following his dis- sertation on machine learning. Dr. Porter was hon- universities and companies in the U.S., Europe and an implementation of an explanation component Japan. She is also the AI editor for Group Decision for an expert system shell she developed with her ored with a Presidential Young Investigator Award from the National Science Foundation in 1988. and Negotiation, a journal published by The Insti- colleagues Drs. Cecile Paris and William Swartout tute for Management Science (TIMS). at USC/Information Sciences Institute. 14 IAAI‒ Preliminary Program

‒ Customer Service Applications : ‒ :  IAAI SlurryMINDER: A Rational Oil Well : ‒ :  Preliminary Program Completion Design Module A Knowledge-Based System Within A E. B. Kelly, P. Caillot, R. Roemer and T. Simien, (Subject to change) Cooperative Processing Environment Dowell Schlumberger Dale Danilewitz, Whirlpool Corporation; Frederick E. Freiheit IV, Technology Solutions Corporation : ‒ :   Microsoft Product Manager’s Workbench, Monday, July : ‒ :  Inventing Technology with an Integrated : ‒ :  SMART, Support Management Automated Expert System Opening Remarks Reasoning Technology for Compaq William P. Shields, Microsoft Corporation; Carli Scott, IAAI Conference Chair Customer Service Brian R. Watkins, Andersen Consulting Timothy L. Acorn, Compaq Computer Corporation; Sherry H. Walden, Inference Corporation : ‒ :  Finance Applications Meet the Authors : ‒ :  : ‒ :  HelpDesk: Using AI to Improve Customer The Credit Assistant: The Second Leg in the Service Knowledge Highway for American Express Jeffrey Kenyon and Debra Logan, Wednesday, July  James Dzierzanowski and Susan Lawson, American Carnegie Group Inc. Express; Eric Hestenes, Inference Corporation Data Analysis : ‒ :  : ‒ :  Meet the Authors : ‒ :  Making Sense of Gigabytes: A System for MOCCA: A Set of Instruments to Support : ‒ :  Mortgage Credit Granting Knowledge-Based Market Analysis Steve Hottiger and Dieter Wenger, IAAI Opening Reception Tej Anand and Gary Kahn, A. C. Nielsen Swiss Bank Corporation : ‒ :  : ‒ :  TPF Dump Analyzer: A System to Provide Break Tuesday, July  Expert Assistance to Analysts in Solving Run-time Program Exceptions by Deriving : ‒ :  AI-on-Line Panel Program Intention from a TPF Assembly PHAROS – The Single European Market :  ‒ :  Language Program Adviser R. Greg Arbon, Laurie Atkinson, James Chen Ebby Adhami and Malcolm McKenzie, Developing Case-Based Reasoning Systems: and Chris Guida, Covia Technologies Ernst & Young Management Consultants; Approaches, Applications, Advantages, and Mike Thornley, National Westminster NatWest Tools : ‒ :  :  ‒ :  Organized by Cognitive Systems MARVEL: A Distributed Real-time Monitoring and Analysis Application CRESUS: An Integrated Expert System for : ‒ :  Ursula M. Schwuttke, Raymond Yeung, Alan G. Cash Management Lunch Quan, Robert Angelino, Cynthia L. Childs, John R. Pete Shell, Carnegie Mellon University; Gonzalo Veregge and Monica B. Rivera, Jet Propulsion Quiroga, Juan A. Hernandez-Rubio and Javier Laboratory, California Institute of Technology Berbiela, Union Fenosa S.A.; Jose Garcia, Industrial Applications Norsistemas Consultores S.A. : ‒ :  : ‒ :  : ‒ :  Break DMCM: A Knowledge Based Cost Meet the Authors Estimation Tool : ‒ :  Norman Crowfoot, Scott Hatfield and Mike Swank, Xerox Corporation Lunch : ‒ :  AI-on-Line Panel The Ford Motor Company European Automotive Operations Computer Aided : ‒ :  Parts Estimating System Technology Transfer to Corporate America Adam Cunningham, Ford Motor Company; Organized by Spang Robinson and Inference Robert Smart, Inference Europe Ltd. Corporation : ‒ :  : ‒ :  Break Break : ‒ :  An Application for Model Based Reasoning in Experiment Design Andrew B. Parker, Sun Microsystems Inc.; W. Scott Spangler, General Motors IAAI‒ Preliminary Program 15

Thursday, July  AI-on-Line Panel : ‒ :  Successful Deployment of LISP Applications Organized by Lucid Inc. : ‒ :  Break

Routing Applications : ‒ :  Arachne: Weaving the Telephone Network at NYNEX Elissa Gilbert, Rangnath Salgame, Afshin Goodarzi, Yuling Lin, Sanjeev Sardana, Jim Euchner, NYNEX Science and Technology, Inc. : ‒ :  Hub SlAAshing: A Knowledge-Based System Join us in San Jose, California this summer for IAAI–92 / AAAI–92. Located approximately sixty for Severe, Temporary Airline Schedule miles south of San Francisco, San Jose is nestled in the heart of the Santa Clara Valley. Daily maxi- Reduction mum July temperatures average 81, with minimum temperatures averaging 55. Trish Dutton, American Airlines :  ‒ :  Meet the Authors Software Development Regulatory Applications : ‒ :  : ‒ :  : ‒ :  Lunch Knowledge-Based Code Inspection with The CARE System ICICLE J. P. Little and Mark Gingrich, United HealthCare AI-on-Line Panel L. Brothers, V. Sembugamoorthy and A. Irgon, Corporation : ‒ :  Bellcore : ‒ :  Hidden Experts: Knowledge-Based Systems :  ‒ :  A Truly MAGIC Solution in Action Automatic Programming for Sequence Rita C. Kidd, Merced County Human Services Agen- Organized by Aion Corporation and Covia Technologies Control cy; Robert J. Carslon, Andersen Consulting Hiroyuki Mizutani, Yasuko Nakayama, Satoshi Ito, : ‒ :  Yasuo Namioka and Takayuki Matsudaira, Toshiba Corporation Meet the Authors : ‒ :  Meet the Authors Special IAAI Plenary Session : ‒ :  : ‒ : , July ,  Lunch Japan Watch 1992 Panel: Expert System Applications and Advanced Knowledge- Based Systems Research AI-on-Line Panel Panelists: Edward Feigenbaum (Chair), Stanford : ‒ :  University; Robert S. Engelmore, Stanford Universi- ty; Peter Friedland, NASA Ames Research Center; The Art of Pioneering a Neural Network in Bruce Johnson, Andersen Consulting; Penny Nii, Your Organization Stanford University; Herbert Schorr, USC/Informa- Organized by NeuralWare tion Sciences Institute; and Howard E. Shrobe, Sym- bolics, Inc./Massachusetts Institute of Technology : ‒ :  Break 16 AAAI‒ Preliminary Program

AAAI– : ‒ :  : ‒ :  Classifier Learning from Noisy Data as Session: Learning: Inductive II Preliminary Program Probabilistic Evidence Combination : ‒ :  (Subject to Change) Steven W. Norton, Rutgers University & Siemens Corporate Research; Haym Hirsh, Rutgers Polynomial-Time Learning with Version University Spaces Tuesday, July  :  ‒ :  Haym Hirsh, Rutgers University : ‒ :  Learning in FOL with a Similarity Measure : ‒ :  Plenary Session: Keynote Address Gilles Bisson, Université Paris-sud COGIN: Symbolic Induction with Genetic The Gardens of Learning: The Vision of AI :  ‒ :  Algorithms Oliver G. Selfridge, Senior Staff Scientist, David Perry Greene and Stephen F. Smith, Carnegie GTE Laboratories Session: Problem Solving Mellon University : ‒ :  : ‒ :  : ‒ :  Break Systematic and Nonsystematic Search TDAG: An Algorithm for Learning to Predict Strategies Discrete Sequences :  ‒ :  Pat Langley, NASA Ames Research Center Philip Laird, NASA Ames Research Center Session: Learning—Inductive I : ‒ :  : ‒ :  : ‒ :  Linear-Space Best-First Search: Summary of A Personal Learning Apprentice ChiMerge: Automatic Discretization of Results Lisa Dent, Jesus Boticario, Tom Mitchell and Numeric Features Richard E. Korf, University of California, Los Angeles David Zabowski, Carnegie Mellon University; Randy Kerber, Lockheed AI Center John McDermott, Digital Equipment Corporation : ‒ :  : ‒ :  An Analysis of Branch-and-Bound with : ‒ :  The Feature Selection Problem: Traditional Applications: Summary of Results Session: Problem Solving— Methods and a New Algorithm Weixiong Zhang and Richard E. Korf; University of Search II and Expert Systems Kenji Kira, Mitsubishi Electric Corporation; Larry A. California, Los Angeles Rendell, University of Illinois : ‒ :  :  ‒ :  Minimax Is Not Optimal for Imperfect Game Performance of IDA on Trees and Graphs Players A. Mahanti, S. Ghosh, D. S. Nau, L. N. Kanal, Eric B. Baum, NEC Research Institute Invited Talks University of Maryland; A. K. Pal, IIM, Calcutta : ‒ :  T, J  ‒ T, J  :  ‒ :  (Schedule to be announced) Improved Decision-Making in Game Trees: Session: Representation & Recovering from Pathology Reasoning—Belief I Learning Visual Behaviors Arthur L. Delcher, Loyola College in Maryland; Simon Kasif, The Johns Hopkins University Dana H. Ballard, University of Rochester : ‒ :  Learning to Act: A Perspective from Con- From Statistics to Beliefs : ‒ :  trol Theory Fahiem Bacchus, University of Waterloo; Adam Moving Target Search with Intelligence Andrew Barto, University of Massachusetts Grove and Daphne Koller, Stanford University, Toru Ishida, NTT Communication Science Joseph Y. Halpern, IBM Almaden Research Center The Distributed AI Melting Pot Laboratories Edmund H. Durfee, University of Michigan : ‒ :  : ‒ :  Reasoning as Remembering: The Theory A Belief-Function Logic Modeling Accounting Systems to Support and Practice of CBR Alessandro Saffiotti, Université Libre de Bruxelles Multiple Tasks: A Progress Report Kristian Hammond, University of Chicago : ‒ :  Walter Hamscher, Price Waterhouse Technology AI and Molecular Biology Centre Abstract States of Belief and Lawrence Hunter, National Library of Medicine Conditionalization : ‒ :  Artificial Life Adnan Y. Darwiche and Matthew L. Ginsberg, Session: Representation & Reasoning Christopher G. Langton, Los Alamos National Stanford University Laboratory —Belief II Progress in Legal Reasoning :  ‒ :  : ‒ :  Edwina L. Rissland, University of Massachusetts Lexical Imprecision and Fuzzy Constraint Combining Circumscription and Modal and Harvard Law School Networks Logic AI and Multimedia James Bowen, Robert Lai and Dennis Bahler, North Jacques Wainer, University of Colorado Roger C. Schank, Northwestern University Carolina University : ‒ :  Machine Translation—Now : ‒ :  A Logic for Subjunctive Queries Yorick Wilks, New Mexico State University Lunch Craig Boutilier, University of British Columbia AAAI‒ Preliminary Program 17

: ‒ :  : ‒ :  : ‒ :  Ideal Introspective Belief Common Sense Retrieval Real-time Metareasoning with Dynamic Kurt Konolige, SRI International A. Julian Craddock, University of British Columbia Trade-off Evaluation Ursula M. Schwuttke, Jet Propulsion Laboratory; : ‒ :  : ‒ :  Les Gasser, University of Southern California Logic of Knowledge and Belief for Recursive When Should a Cheetah Remind You of a Modeling – Preliminary Report Bat? Reminding in Case-Based Teaching : ‒ :  Piotr J. Gmytrasiewicz and Edmund H. Durfee, Daniel C. Edelson, Northwestern University Run-Time Prediction for Production Systems University of Michigan Franz Barachini and Hans Mistelberger, Alcatel- : ‒ :  ELIN Research Center; Anoop Gupta, : ‒ :  Model-Based Case Adaptation Stanford University Break Eric K. Jones, Victoria University of Wellington : ‒ :  : ‒ :  : ‒ :  Comparison of Three Algorithms for Session: Learning—Inductive III Generating Dialectical Examples Ensuring Serializable Executions in Parallel Production Systems : ‒ :  Automatically Kevin D. Ashley and Vincent Aleven, James G. Schmolze, Tufts University; Daniel The Attribute Selection Problem in Decision University of Pittsburgh Neiman, University of Massachusetts Tree Generation : ‒ :  Usama M. Fayyad, Jet Propulsion Laboratory, : ‒ :  California Institute of Technology; Keki B. Irani, AAAI‒ Opening Reception Session: Representation and The University of Michigan Reasoning—Terminological : ‒ :  : ‒ :  Sparse Data and the Effect of Overfitting Wednesday, July  Recognition Algorithms for the LOOM Avoidance in Decision Tree Induction Classifier Cullen Schaffer, CUNY/Hunter College : ‒ :  Robert M. MacGregor and David Brill, USC/Information Sciences Institute : ‒ :  Session: Learning—Neural Network and Complementary Discrimination Learning Hybrid : ‒ :  with Decision Lists Computing Least Common Subsumers in : ‒ :  Wei-Min Shen, Microelectronics and Computer Description Logics Technology Corporation Using Knowledge-Based Neural Networks to William W. Cohen, AT&T Bell Laboratories; Alex Improve Algorithms: Refining the Chou- Borgida and Haym Hirsh, Rutgers University : ‒ :  Fasman Algorithm for Protein Folding Learning to Learn Decision Trees Richard Maclin and Jude W. Shavlik, University of : ‒ :  Vlad G. Dabija, Stanford University; Katsuhiko Wisconsin A Non-Well-Founded Approach to Tsujino and Shogo Nishida, Mitsubishi Electric Terminological Cycles Corporation : ‒ :  Robert Dionne, Eric Mays and Frank J. Oles, Using Symbolic Inductive Learning to : ‒ :  IBM T. J. Watson Research Center Improve Knowledge-Based Neural Networks Session: Problem Solving—Hardness Geoffrey G. Towell, Siemens Corporate Research, : ‒ :  and Easiness and Jude W. Shavlik, University of Wisconsin An Empirical Analysis of Terminological Representation Systems : ‒ :  : ‒ :  Jochen Heinsohn, Daniel Kudenko, Bernhard Nebel Hard and Easy Distributions of SAT A Framework for Integrating Fault Diagnosis and Hans-Jurgen Profitlich, German Research Problems and Incremental Knowledge Acquisition in Center for Artificial Intelligence David Mitchell, Simon Fraser University; Connectionist Expert Systems : ‒ :  Bart Selman, AT&T Bell Laboratories; Joo-Hwee Lim, Ho-Chung Lui and Pei-Zhuang Hector Levesque, University of Toronto Wang, National University of Singapore Break : ‒ :  : ‒ :  :  ‒ :  When Heuristic Solutions Work Adapting Bias by Gradient Descent: An Session: Learning—Constructive Ron Musick and Stuart Russell, University of Incremental Version of Delta-Bar-Delta & Linguistic California, Berkeley Richard S. Sutton, GTE Laboratories Incorporated : ‒ :  : ‒ :  : ‒ :  Learning to Disambiguate Relative Pronouns Using Deep Structure to Locate Hard Session: Problem Solving—Real-Time Claire Cardie, University of Massachusetts Problems Colin P. Williams and Tad Hogg, Xerox Palo Alto : ‒ :  : ‒ :  Research Center Can Real-Time Search Algorithms Meet A Connectionist Parser with Recursive Sen- : ‒ :  Deadlines? tence Structure and Lexical Disambiguation Babak Hamidzadeh and Shashi Shekhar, George Berg, State University of New York at Albany Session: Representation and Reasoning University of Minnesota —Case-Based 18 AAAI‒ Preliminary Program

: ‒ :  : ‒ :  : ‒ :  Learning Relations by Pathfinding Session: Learning—Utility and Bias Qualitative Simulation Based on a Logical Bradley L. Richards and Raymond J. Mooney, Uni- Formalism of Space and Time versity of Texas at Austin : ‒ :  Z. Cui, A. G. Cohen and D. A. Randell, Empirical Analysis of the General Utility University of Leeds :  ‒ :  Problem in Machine Learning : ‒ :  Discrimination-Based Constructive Lawrence B. Holder, University of Texas at Arlington Induction of Logic Programs On the Qualitative Structure of a Boonserm Kijsirikul, Masayuki Numao, and : ‒ :  Mechanical Assembly Masamichi Shimura, Tokyo Institute of Technology COMPOSER: A Probabilistic Solution to the Randall H. Wilson and Jean-Claude Latombe, Stanford University :  ‒ :  Utility Problem in Speed-Up Learning Jonathan Gratch and Gerald DeJong, University of : ‒ :  Session: Robot Navigation Illinois at Urbana-Champaign Self-Explanatory Simulations: Scaling Up to : ‒ :  : ‒ :  Large Models Integrating Planning and Reacting in a A Solution to the EBL Utility Problem Kenneth D. Forbus, Northwestern University; Brian Heterogeneous Asynchronous Architecture Russell Greiner and Igor Jurisica, University of Falkenhainer, Xerox Palo Alto Research Center for Controlling Real-World Mobile Robots Toronto : ‒ :  Erann Gat, Jet Propulsion Laboratory, California Institute of Technology : ‒ :  Break Inductive Policy : ‒ :  : ‒ :  Foster John Provost and Bruce G. Buchanan, Uni- Reactive Navigation through Rough Terrain: versity of Pittsburgh Session: Scaling Up Experimental Results : ‒ :  : ‒ :  David P. Miller, Rajiv S. Desai, Erann Gat, Robert Ivlev and John Loch, Jet Propulsion Laboratory, Session: Multi-Agent Coordination Building Large-Scale and Corporate-Wide California Institute of Technology Case-Based Systems: Integration of the Orga- : ‒ :  nization and Machine Executable Algorithms : ‒ :  A General Equilibrium Approach to Hiroaki Kitano, Akihiro Shibata, Juichirou Kajihara, A Reactive Robot System for Find and Fetch Distributed Transportation Planning and Atsumi Sato, NEC Corporation Tasks in an Outdoor Environment Michael P. Wellman, USAF Wright Laboratory : ‒ :  H. James Antonisse, R. Peter Bonasso and Marc G. Slack, The MITRE Corporation : ‒ :  Wafer Scale Integration for Massively Parallel Memory-Based Reasoning :  ‒ :  Using Joint Responsibility to Coordinate Collaborative Problem Solving in Dynamic Hiroaki Kitano and Moritoshi Yasunaga, Carnegie Landmark-Based Robot Navigation Mellon University Environments Anthony Lazanas and Jean-Claude Latombe, Stanford University N. R. Jennings, Queen Mary & Westfield College : ‒ :  : ‒ :  Learning 10,000 Chunks: :  ‒ :  What’s It Like Out There? Constrained Intelligent Action: Planning Session: Representation and Bob Doorenbos, Milind Tambe, and , Under the Influence of a Master Agent Reasoning—Tractability Carnegie Mellon University Eithan Ephrati and Jeffrey S. Rosenschein, Hebrew : ‒ :  University : ‒ :  The Complexity of Propositional Default : ‒ :  Mega-Classification: Discovering Motifs in Massive Datastreams Logics On the Synthesis of Useful Social Laws for Nomi L. Harris, Lawrence Hunter, and David J. Jonathan Stillman, General Electric Research and Artificial Agents Societies Development Center States, National Institutes of Health (Preliminary Report) : ‒ :  Yoav Shoham and Moshe Tennenholtz, Stanford : ‒ :  University Speeding Inference by Acquiring New Session: Perception Concepts : ‒ :  : ‒ :  Henry Kautz and Bart Selman, AT&T Bell Laboratories Session: Representation Grouping Iso-Velocity Points for and Reasoning: Qualitative Ego-Motion Recovery : ‒ :  : ‒ :  Yibing Yang, Harvard University A New Incremental Algorithm for Generating Prime Implicates Towards a Qualitative Lagrangian Theory of : ‒ :  Johan de Kleer, Xerox Palo Alto Research Center Fluid Flow Computational Model for Face Location Gordon Skorstad, University of Illinois Based on Cognitive Principles : ‒ :  Venu Govindaraju, Sargur N. Srihari and David Lunch Sher, State University of New York at Buffalo AAAI‒ Preliminary Program 19

: ‒ :  : ‒ :  :  ‒ :  Computation of Upper-Bounds for Solving Constraint Satisfaction Problems Session: Learning—Robotic Stochastic Context-Free Languages Using Finite State Automata A. Corazza, Istituto per la Ricerca Scientifica e Vempaty, Nageshwara Rao, University of Central : ‒ :  Tecnologica; R. De Mori, McGill University; Florida Reinforcement Learning with Perceptual G. Satta, University of Pennsylvania Aliasing: The Predictive Distinctions : ‒ :  : ‒ :  Approach On the Minimality and Decomposability of Lonnie Chrisman, Carnegie Mellon University Session: Representation and Reasoning Constraint Networks —Qualitative Model Construction Peter van Beek, University of Alberta : ‒ :  Reinforcement Learning with a Hierarchy of : ‒ :  : ‒ :  Abstract Models Automated Model Selection Using Context- Session: Representation and Satinder P. Singh, University of Massachusetts Dependent Behaviors Reasoning—Action and Change P. Pandurang Nayak, Stanford University : ‒ :  : ‒ :  Acquisition of Automatic Activity through : ‒ :  Formalizing Reasoning about Change: A Practice: Changes in Sensory Input Automatic Abduction of Qualitative Models Qualitative Reasoning Approach Jack Gelfond, Marshall Flax, Raymond Endres, Bradley L. Richards and Benjamin J. Kuipers, (Preliminary Report) Stephen Lane and David Handelman, Princeton University of Texas, Austin; Ina Kraan, University of James M. Crawford and David W. Etherington, University Edinburgh AT&T Bell Laboratories :  ‒ :  : ‒ :  : ‒ :  Automatic Programming of Robots Using Causal Approximations Concurrent Actions in the Situation Calculus Genetic Programming P. Pandurang Nayak, Stanford University; Leo Fangzhen Lin and Yoav Shoham, Stanford John R. Koza and James P. Rice, Stanford University Joskowicz and Sanjaya Addanka, IBM T. J. Watson University Research Center :  ‒ :  : ‒ :  : ‒ :  Session: Problem Solving—Constraint Deriving Properties of Belief Update from Learning Engineering Models with the Satisfaction II Theories of Action Minimum Description Length Principle Alvaro del Val and Yoav Shoham, Stanford University : ‒ :  R. Bharat Rao and Stephen C-Y. Lu, University of On the Density of Solutions in Equilibrium Illinois at Urbana-Champaign : ‒ :  Points for the Queens Problem : ‒ :  Nonmonotonic Sorts for Feature Structures Paul Morris, IntelliCorp Session: Special IAAI Plenary Panel Mark A. Young, The University of Michigan : ‒ :  Panel: Japan Watch 1992: Expert System : ‒ :  A New Method for Solving Hard Satisfiability Applications and Advanced Knowledge- Session: Natural Language—Parsing Problems Based Systems Research : ‒ :  Bart Selman, AT&T Bell Laboratories; Hector Panelists: Edward Feigenbaum (Chair), Stanford Levesque, University of Toronto; and David Mitchell, University; Robert S. Engelmore, Stanford Universi- A Probabilistic Parser Applied to Software Simon-Fraser University ty; Peter Friedland, NASA Ames Research Center; Testing Documents Bruce Johnson, Andersen Consulting; Penny Nii, : ‒ :  Stanford University; Herbert Schorr, USC/Informa- Mark A. Jones, AT&T Bell Laboratories; tion Sciences Institute; and Howard E. Shrobe, Sym- Jason Eisner, Emmanuel College A New Improved Connectionist Activation bolics, Inc./Massachusetts Institute of Technology Function for Energy Minimization : ‒ :  Gadi Pinkas, Washington University; Rina Dechter, Parsing Rum Amok: Relation-Driven University of California, Irvine Control for Text Analysis :  ‒ :   Paul S. Jacobs, GE Research & Development Center Thursday, July Semantic Evaluation as Constraint Network : ‒ :  : ‒ :  Consistency Shipping Departments vs. Shipping Nicholas J. Haddock, Hewlett Packard Laboratories Session: Problem Solving—Constraint Pacemakers: Using Thematic Analysis to Satisfaction I Improve Tagging Accuracy :  ‒ :  : ‒ :  Uri Zernik, General Electric Research Session: Representation and & Development Center Reasoning—Temporal Efficient Propositional Constraint Propagation Mukesh Dalal, Rutgers University : ‒ :  : ‒ :  Classifying Texts Using Relevancy Signatures : ‒ :  On the Computational Complexity of Ellen Riloff and Wendy Lehnert, University of Mas- Temporal Projection and Plan Validation An Efficient Cross Product Representation of sachusetts Bernhard Nebel, German Research Center for the Constraint Satisfaction Problem Search : ‒ :  Artificial Intelligence; Christer Backstrom, Space Linkoeping University Paul D. Hubbe and Eugene C. Freuder, University of Break New Hampshire 20 AAAI‒ Preliminary Program

: ‒ :  : ‒ :  : ‒ :  Algorithms and Complexity for Reasoning Session: Planning I Discovery of Equations: Experimental about Time (Extended Abstract) Evaluation of Convergence Martin Charles Golumbic, IBM Israel Scientific : ‒ :  Robert Zembowicz and Jan M. Zytkow, Wichita Center; Ron Shamir, Tel Aviv University The Expected Value of Hierarchical State University Problem-Solving : ‒ :  : ‒ :  Fahiem Bacchus and Qiang Yang, Complexity Results for Serial Decomposability University of Waterloo Conjecture of Hidden Entities in Theory- Tom Bylander, The Ohio State University Driven Discovery: The MECHEM System : ‒ :  Raul E. Valdes-Perez, Carnegie Mellon University :  ‒ :  Achieving the Functionality of Filter Temporal Reasoning in Sequence Graphs Conditions in a Partial Order Planner : ‒ :  Jurgen Dorn, Technical University Vienna Gregg Collins and Louise Pryor, Northwestern Symmetry as Bias: Rediscovering University Special Relativity :  ‒ :  Michael Lowry, Kestrel Institute Session: Natural Language—Interpretation : ‒ :  On the Complexity of Domain-Independent : ‒ :  : ‒ :  Planning Session: Planning II An On-Line Computational Model of Kutluhan Erol, Dana S. Nau, and V. S. Human Sentence Interpretation Subrahmanian, University of Maryland : ‒ :  Daniel Jurafsky, University of California, Berkeley Learning from Goal Interactions in Planning: : ‒ :  Goal Stack Analysis and Generalization : ‒ :  Constrained Decision Revision Kwang Ryel Ryu and Keki B. Irani, The University of Literal Meaning and the Comprehension of Charles Petrie, MCC AI Lab Michigan Metaphors : ‒ :  Steven L. Lytinen, Jeffrey D. Kirtner, and Robert : ‒ :  R. Burridge, The University of Michigan Session: Representation and Analyzing Failure Recovery to Improve Reasoning—Abduction and Diagnosis Planner Design : ‒ :  Adele E. Howe, University of Massachusetts Actions, Beliefs and Intentions in Rationale : ‒ :  Clauses and Means Clauses Consistency-Based Diagnosis in : ‒ :  Cecile T. Balkanski, Harvard University Physiological Domains Cultural Support for Improvisation Keith L. Downing, Linkoeping University Philip E. Agre, University of California, San Diego; :  ‒ :  Ian D. Horswill, Massachusetts Institute of Technology An Approach to the Representation of : ‒ :  Iterative Situations Adaptive Model-Based Diagnostic Mecha- : ‒ :  Michael J. Almeida, SUNY Plattsburgh nism Using a Hierarchical Model Scheme Session: Explanation and Tutoring Yoichiro Nakakuki, Yoshiyuki Koseki, and Midori : ‒ :  Tanaka, NEC Corporation : ‒ :  Lunch Understanding Causal Descriptions of : ‒ :  : ‒ :  Physical Systems Dynamic MAP Calculations for Abduction Gary C. Borchardt, Massachusetts Institute of Session: Learning—Theory Eugene Charniak, Brown University Technology : ‒ :  : ‒ :  : ‒ :  An Analysis of Bayesian Classifiers Reasoning MPE to Multiply Connected Belief Steps from Explanation Planning to Model Pat Langley, Wayne Iba, and Kevin Thompson; Networks Using Message Passing Construction Dialogues NASA Ames Research Center Bon K. Sy, Queens College, City University of New Daniel Suthers, Beverly Woolf, and York Matthew Cornell, University of Massachusetts : ‒ :  A Theory of Unsupervised Speedup Learning : ‒ :  : ‒ :  Prasad Tadepalli, Oregon State University Break Generating Cross-References for Multimedia Explanation : ‒ :  : ‒ :  Kathleen R. McKeown, Steven K. Feiner, Jacques Inferring Finite Automata with Stochastic Session: Learning—Discovery Robin, Doree D. Seligmann, Michael Tanenblatt, Output Functions and an Application to Columbia University Map Learning : ‒ :  : ‒ :  Thomas Dean, Kenneth Basye, Leslie Kaelbling, Operational Definition Refinement: Evangelos Kokkevis, and Oded Maron, Brown A Discovery Process Encoding Domain and Tutoring Knowledge University Jan M. Zytkow, Jieming Zhu, and Via a Tutor Construction Kit Robert Zembowicz, Wichita State University Tom Murray and Beverly Park Woolf, University of : ‒ :  Massachusetts Oblivious PAC Learning of Concept Hierarchies Michael J. Kearns, AT&T Bell Laboratories Registration Information 21

AAAI‒ Technical Tutorial Program Program Registration Registration July -,  July ‒,  Your AAAI– technical program registra- The Tutorial Program Registration includes tion includes admission to technical sessions, admission to one tutorial, the AAAI– / invited talks, the AAAI– / IAAI– joint IAAI– Joint Exhibition, the AI-on-Line exhibition, AI-on-Line panels and the panels at IAAI–, and the tutorial syllabus. Wednesday evening plenary session at Prices quoted are per tutorial. A maximum of IAAI–, the AAAI– opening reception, four may be taken due to parallel schedules. and the AAAI– conference Proceedings. Tutorial Fee Schedule Early Registration (Postmarked by  May) ‒ AAAI Members IAAI Registration Regular $195 Student $70 July -,  Nonmembers Regular $240 Student $90 Your IAAI– registration includes admission to IAAI–, the AI-on-Line sessions, the Late Registration (Postmarked by  June) AAAI– / IAAI– joint exhibition, the AAAI Members AAAI– opening address, the IAAI– open- Regular $235 Student $95 ing reception, and Innovative Applications 4. Nonmembers See you in San Jose this July! Regular $275 Student $115 ‒ ‒ On-Site Registration (Postmarked after  June Registration forms and inquiries should IAAI / AAAI or onsite. Hours are listed below.) be directed to: Registration Fees AAAI Members AAAI– / IAAI– Regular $295 Student $125  Burgess Drive  Early Registration (Postmarked by May) Nonmembers Menlo Park, California - USA AAAI Members Regular $335 Student $145 ⁄-; Fax: ⁄- Regular $245 Student $95 Email [email protected]. Nonmembers On-Site Registration will be located on Regular $290 Student $155 the lower level of the San Jose Convention Workshop Registration Center,  Almaden Avenue, San Jose, Cali- Late Registration (Postmarked by  June) fornia. AAAI Members July -,  Registration hours will be Sunday, July  Regular $295 Student $95 Workshop registration is limited to active through Thursday, July  from : AM – Nonmembers participants determined by the organizer pri- : PM. All attendees must pick up their reg- Regular $340 Student $155 or to the conference. Those individuals at- istration packets for admittance to programs. tending workshops only are subject to a On-Site Registration (Postmarked after  June $100.00 registration fee. or onsite. Hours below) AAAI Members Regular $360 Student $100 Housing Nonmembers Payment & Registration AAAI has reserved a block of rooms in San Regular $400 Student $170 Information Jose hotel properties at reduced conference ‒ ‒ rates. To qualify for these rates, housing AAAI / IAAI Prepayment of registration fees is required. reservations, including requests for suites, Combined Registration Checks, international money orders, bank must be made with the San Jose Convention & Visitors Bureau housing office. A deposit To attend both the AAAI– and the IAAI– transfers and traveler’s checks must be in must accompany the housing form to guar- Conferences, and to receive the proceedings US$. Amex, MasterCard, Visa, and govern- antee a reservation. Checks and money or- of both, please add $95.00 to the appropriate ment purchase orders are also accepted. Reg- ders should be made payable to the San Jose registration fee above. istrations postmarked after the June 12 deadline will be subject to on-site registra- Convention & Visitors Bureau. American Ex- tion fees. The deadline for refund requests is press, Diner’s Club, MasterCard and Visa are June , . All refund requests must be also accepted. made in writing. A $75.00 processing fee will The San Jose Convention & Visitors Bu- be assessed for all refunds. Student registra- reau Housing Office will acknowledge the re- tions must be accompanied by proof of full- ceipt of reservations. If an acknowledgement time student status. is not received within a reasonable period (two – three weeks), please resubmit the 22 Housing & Transportation original request with a confirming letter. If July  Breakfast & dinner book early. In order to qualify for these spe- choice(s) for housing are not available, reser- July  Breakfast cial discounted fares, travel must be round- vations will be forwarded to a comparable Rates per person trip within the continental United States and property. Confirmations will be issued by travel must take place between July  and Double $191.27 hotels. July , . Single $236.27 The deadline for housing reservations is Additional nights can be reserved for $26.50 June , . Reservations received after this (Double), $35.50 (Single). No meals are in- date cannot be guaranteed housing availabil- cluded. ity or special rates. Please act early to guaran- Ground Transportation The deadline for student housing reserva- tee space availability. Changes or cancella- tions is June , . Reservations received The following information provided is the tions of reservations must be made in writ- after June  cannot be guaranteed availabili- best available at press time. Please confirm ing with the San Jose Convention & Visitors ty or special rates. Prepayment of housing fares when making reservations. Bureau until June 12, 1992. After June , all fees is required. Checks, international money changes and cancellations must be made di- orders, bank transfers, and traveler’s checks Airport Connections rectly with the hotels. All rates quoted are per must be in US currency. American Express, Several companies provide service from San person and subject to a room tax of %. MasterCard, and Visa, are also accepted. Re- Francisco International (SFO) and San Jose Please see maps on pages twenty-four and fund requests must be made in writing. The International (SJC) Airports to downtown thirty-one. deadline for refund requests is June , . San Jose. A sampling of companies and their Headquarters Hotel A $50.00 processing fee will be assessed for one-way rates are shown below. Contact the all refunds. company directly for reservations. The Fairmont Hotel Student housing is restricted to full-time In addition to the following firms, all offi-  South Market Street graduate or undergraduate students enrolled cial convention hotels offer complimentary $108.00 (S/D/T) in an accredited college or university pro- shuttle service to and from the San Jose In- Distance to Center: One block gram. Proof of full-time student status must ternational Airport. accompany the student housing form (see Other Hotels page twenty-nine). Housing forms and in- Bay Porter Express quiries should be directed to ⁄- Holiday Inn-Park Center Plaza AAAI– / IAAI– SFO to San Jose: $30.00 first person,  Almaden Boulevard  Burgess Drive $10.00 each additional person $75.00 (S/D) Menlo Park, California - USA Distance to Center: One block ⁄-; Fax: ⁄- The Airport Connection ⁄- Hotel De Anza Email [email protected].  West Santa Clara Street SFO to Fairmont Hotel: $105.00 (S); $120.00 (D) $17.00 each person Distance to Center: Three blocks Air Transportation South Bay Shuttle Radisson Hotel The American Association for Artificial In- ⁄-  North Fourth Street telligence has selected American Airlines as SFO to San Jose: $21.00 first person, $91.00 (S/D) the official carrier. Negotiated fares will re- $5.00 each additional person Distance to Center: Three miles flect a savings of five percent off any applica- SJC to San Jose: $12.50 first person, Located on Light Rail Line. ble discounted coach fare or forty-five per- $5.00 each additional person cent off full, unrestricted, round-trip coach Red Lion Hotel Express Airport Shuttle  Gateway Place class (y/yn/y28) fares in effect. Tickets must ⁄- $80.00 (S/D) be purchased at least three days before the Distance to Center: Three and one-half miles conference and reservations must be made in SFO to San Jose: $21.00 first person, Complimentary shuttle to Light Rail Line. M class for full coach fares or appropriate $5.00 each additional person class of service, whatever applies. AAAI’s pre- SJC to San Jose: $12.00 first person, Student Housing ferred travel agent is Custom Travel, phone $5.00 each additional person ⁽⁾ - ⁽⁾ - AAAI has reserved a block of dormitory or . You or your travel agent may also call American Airlines rooms at Santa Clara University, Santa Clara ⁽⁾ - : : for student housing during the conference. at between AM and midnight Central Standard Time. Please give Car Rentals Accommodations include five nights stay    (July -) with linen service, and the fol- the special discount code: Star #S Z U . AAAI has selected Hertz Rental Car as the of- lowing meals: Restrictions: Reservations for flights re- ficial car rental agency for AAAI– / quiring advance purchase must adhere to all IAAI–. Special rates will be honored from July  Dinner restrictions that apply to that fare. Regular July ‒, , and include unlimited July  Breakfast & dinner coach fares and some other nonrestrictive mileage. For reservations, please call Custom July  Breakfast (No dinner due to fares do not require advance booking but all Travel at ⁄- or ⁄-. You AAAI– opening reception) tickets must be issued at least three days be- may also contact the Hertz convention desk July  Breakfast & dinner fore the conference. To be sure of availability, at ⁄-, identify yourself as a Transportation & Tours 23

AAAI– registrant, and give the special pacity of agent for the Suppliers which are tioning in the real world. Reasoning about code#CV. You may also ask for special the providers of the service. Because AAAI the physical world involves, among other rates available at time of booking. has no control over the personnel, equip- things, the modeling of, and reasoning ment or operations of providers of accom- about, space and time, kinematics and dy- Terms and Conditions: modations or other services included as part namics, and machine vision. Work on agent Applicable charges for taxes, optional refuel- of the AAAI– or IAAI– program, AAAI architectures involves foundational questions ing service, PAI, PEC and LIS are extra. Op- assumes no responsibility for and will not be in the representation of time, action, belief tional LDW may be purchased at $13 or less liable for any personal delay, inconveniences and similar notions, investigating various per day. Rates are nondiscountable nor us- or other damage suffered by conference at- forms of inference, machine learning, and able with any promotion or coupon, with the tendees which may arise by reason of (1) any reconciling problem-solving with real-time exception of Hertz gift certificates with wrongful or negligent acts or omissions on performance requirements. Throughout the PC#. Rentals are subject to Hertz mini- the part of any Supplier or its employees, (2) projects, emphasis is placed on development mal rental and driving age of twenty-five, any defect in or failure of any vehicle, equip- of rigorous theories, coupled with substantial driver’s license and credit requirements as ment or instrumentality owned, operated or experimental validation. The research in the well as car availability. Weekend rentals are otherwise used by any Supplier, or (3) any laboratory will be explained through live available for pick-up between noon Thurs- wrongful or negligent acts or omissions on robotic demonstrations, videos, posters, and day and noon Sunday and must be returned the part of any other party not under the personal explanations. no later than 11:59 PM Monday. Weekend control, direct or otherwise, of AAAI. minimum rentals are as follows: Thursday Neuron Data, Palo Alto pick-ups, three days; Friday pick-ups, two Wednesday, July , : ‒ : PM, days; Saturday and Sunday pick-ups, one day. $20.00 per person (includes box dinner). Weekly rentals are available from five to sev- Tours  en days and must be kept over a Saturday Tours are scheduled for Wednesday, July , Neuron Data develops and markets software  night. Saturday Night Keep rentals are avail- and Thursday, July , and include trans- tools and libraries for building and deploying able for pick-up any day of the week and portation from the San Jose Convention smarter applications. The company's major must be kept over a Saturday night. Center and return. Tour times do not include current products are NEXTPERT OBJECT™, an travel, estimated to be forty-five minutes to Parking expert system development tool installed one hour each way. Departure and return worldwide, and Neuron Data Open Inter- Parking will be available at the San Jose Con- times will be indicated with confirmation of face™, a development tool for building vention Center parking garage for a one-time reservation. Register early as space is limited portable graphical interfaces across operating entrance fee of $4 per day or $6 per day for for each tour. Please see registration form on systems and windowing systems. The NEX- in/out privileges. page twenty-five. PERT and Open Interface libraries are embed- Additional parking is available across the ded in a wide variety of software applications street from the Convention Center at the IBM’s Santa Teresa Laboratory, and products. corner of San Carlos and Almaden Streets. San Jose Neuron Data staff will discuss, through Rates for this lot are seventy-five cents per Wednesday, July , :-: PM existing case studies, the new role of AI and half-hour for the first hour and fifty cents GUI software as critical tools for competi- $20.00 per person (includes box dinner). per half-hour every hour thereafter for a tiveness. The presentation will take place at maximum of $9.50 per day. For Sunday (all Neuron Data's world headquarters in Palo The Santa Teresa Laboratory (STL) is one of day) and weekdays (after 6:00 pm ) there is a Alto, California. $3.00 flat fee. IBM’s major programming development sites. More than 1,800 programmers work at Bus this site on products as diverse as database, Teleos Research, Palo Alto high level languages and knowledge based The San Jose Greyhound station is located at , :-: systems. The tour will be hosted by Dr. Kevin Thursday, July PM  Almaden Avenue, approximately two Seppi and several other members of the STL $20.00 per person (includes box dinner). blocks from the Convention Center. ⁄- . programming community. Part of the tour will include visits to 10 STL Knowledge Min- Teleos Research is working in the fields of in- Rail ing Centers™ and product demonstrations. telligent, embedded, real-time systems and machine vision. Teleos will present an The Amtrak station is located at  Cahill Robotics Laboratory & overview of the current research program Street, approximately one-half mile from the and demonstrations on the robot testbed as Convention Center. ⁄-. Computer Science Department, Stanford University well as PRISM-3: a real-time stereo vision Disclaimer , :-: system; Rex and Gapps: software tools for Thursday, July PM building intelligent, real-time systems; and In offering American Airlines, Hertz Rental $20.00 per person (includes box dinner). Iconicode: a visual programming environ- Cars, KiddieCorp, and all other service ment. providers (hereinafter referred to as “Suppli- Work in the Robotics Laboratory centers er(s)”) for the Innovative Applications Con- around two interwoven themes: modeling ference and the National Conference of Arti- and reasoning about the physical world, and ficial Intelligence, AAAI acts only in the ca- modeling and reasoning about agents func- 24 Tours, Childcare, & Map

The Tech Museum of Innovation, San Jose Special Ticket Offer Discounted Tech Museum tickets are avail- able for purchase through AAAI. Located di- rectly across the street from the San Jose Convention Center, the Tech Museum of In- novation brings Silicon Valley ingenuity to life with imaginative exhibits and labs. The Tech Museum features interactive exhibits in six areas: Microelectronics, Space Explo- ration, High Tech Bikes, Robotics, Materials and Biotechnology. Prices are $5.00 (adults) and $3.00 (children, ‒). Tickets valid for unlimited entry Tuesday through Sunday, : AM–: PM. Closed Monday. Please see registration form on page twenty-five to order.

Child Care Child care will be provided by KiddieCorp. Their experienced staff will offer CLUB KID, a program geared towards infants, preschool and school-age children up to age eleven. Fees include age-appropriate games and toys; arts and crafts; a separate, supervised nap room; movies and videos; and snacks. Lunch provided for children in care between : and : PM. The schedule for CLUB KID is: Sunday, July : : AM – : PM Monday, July : : AM – : PM Tuesday, July : : AM – : PM Wednesday, July : : AM – : PM Thursday, July : : AM – : PM Cost is $6.00 per hour per child. To regis- ter, complete the form on page twenty-six and return to AAAI by June , . In- quiries regarding CLUB KID should be di- rected to Kiddiecorp at ⁄-.

Get Your Registration Form in Early!

The first , preregistrants will receive a commemorative AAAI Tenth Anniversary Pin Map 31 33 AAAI–91 Registration Brochure