Panel: The Role of Chess in Artificial Intelligence Research Robert Levinson (Chairperson) Computer and Information Sciences Applied Sciences Building University of California at Santa Cruz Santa Cruz, CA 95064 (408)459-2087 ARPANET:[email protected],edu FAX:429-0146 Feng-hsiung Hsu Jonathan Schaeffer IBM T.J. Watson Research Center Computing Science PO Box 704 University of Alberta York town Heights Edmonton NY 10598 Canada T6G 2H1 T. Anthony Marsland David E* Wilkins Computing Science SRI International EJ227 University of Alberta 333 Ravenswood Ave Edmonton Menlo Park Canada T6G 2H1 CA 94025 Abstract Panel Summary The factors that make chess an excellent domain for AI Our eminent researchers including John McCarthy, research include: Allen Newell, Claude Shannon, Herb Simon, Ken Thompson and Alan Turing put significant effort • Richness of the problem-solving domain. into computer chess research. Now that comput• • Ability to monitor and record progress accurately ers have reached the grandmaster level, and are through competition and rating, because of its well- beginning to vie for the World Championship, the defined structure. AI community should pause to evaluate the sig• nificance of chess in the evolving objectives of AI, • Chess has been around for centuries - the basics are evaluate the contributions made to date, and assess well-understood internationally, expertise is readily what can be expected in the future. Despite the available and is (generally!) beyond proprietary or general interest in chess amongst computer scien• nationalistic interests. Has been considered a "game tists and the significant progress in the last twenty of intelligence." Many players of the game feel men- years, there seems to be a Jack of appreciation for tally "stretched." the field in the AI community. On one hand this is • Detailed psychological studies of chess playing exist. the fruit of success (brute force works, why study These studies suggest that human players use differ• anything else?), but also the result of a focus on ent reasoning modes from those in current chess pro- performance above all else in the chess community. grams. Further, the reasoning modes are also used in Also, chess has proved to be too challenging for many other problem-solving domains. many of the AI techniques that have been thrown at it. We wish to promote chess as the fundamen• • Excellent test bed for uncertainty management tal test bed recognized by our founding researchers schemes - the basis of most expert problem-solving. and increase awareness of its contribution to date. The well-definedness and discreteness of the game have led many to ignore this. Levinson, et at. 547 The above factors make chess a useful tool regardless • The effectiveness of brute-force search. Chess has of the strength of the current programs. Because of clearly demonstrated that simple, brute-force ap• the success of the current methods there remains a vast proaches should not be quickly discarded. arena of other methods that have not been explored. • Iterative search. Some of the ideas developed for The most obvious lack is in the application and develop- alpha-beta search, iterative deepening in particular, ment of machine learning techniques to chess, but other are applicable to other search domains. areas, including knowledge representation and compila- tion, planning and control, also seem to be applicable. • The inadequacy of conventional AI techniques for re• AI researchers should be encouraged to use chess as a altime computation. No competitive computer chess test bed for their techniques, with the understanding program uses AI languages or knowledge representa- that chess is not the end in itself. Chess may provide tion methods. Why? They are too slow for a real- the avenue by which bridges may be built between cog- time, high performance application. nitive science, AI and connectionist modeling- Although these (and other, lesser contributions) have Wit h the current and future battle for the World enhanced our knowledge, it is not clear whether the Human-Computer Championship the AI community effort expended justifies the results obtained. should be made more sensitive to the issues involved It is easy to question the usefulness of computer chess and their bearing on intelligence research: Is search research. It is important to distinguish between com• sufficient? How much detailed chess knowledge is re- puter chess research and research using chess as a test quired? How is this knowledge implemented and incor• bed. Unfortunately, the latter has evolved into the for• porated with search? We are fortunate to have a World mer. An entirely new field of "computer chess" has Champion who promotes creativity over the chess board evolved, with the emphasis on chess performance and and is willing to face the challenge from computers chess research - not generally of much interest to the AI head-on. community. There is a much deserved credibility prob• The members of the panel and the presentations have lem here. The unfortunate correlation between program been designed to address these topics in a way that sup• speed and performance encourages short-term projects ports our objectives to make chess an important and (speeding up a move generator 10%) at the sacrifice respected AI tool in this new decade. Jonathan Scha- of long-term research projects (such as chess programs effer will emphasize those areas of computer chess re• that learn). search that have been ignored, because the approach After over 30 years of work on chess programs, where has been a competitive/engineering one instead of sci• are the scientific advances in: entific. Feng-hsiung Esu of the Deep Thought team will discuss the role of knowledge in current chess pro- • knowledge-based search algorithms? There has been gramming and argue that more responsibility for the some good work in this area, but none has progressed knowledge should be put on the machines themselves, enough to be used in competitive chess programs. Tony Marsland will present specific open research issues Alpha-beta simplifies the programming task, but the in computer chess that will require AI solutions. Robert exponential search limits what can be achieved. Levinson will describe an alternative model of chess • knowledge representation and acquisition? These ar- computation, a self-learning pattern-oriented chess pro• eas are of considerable importance to chess programs, gram ("Morph") whose knowledge is learned incremen• yet the computer chess community has done embar• tally from experience, without many examples being rassing little research in this area. stored (and with little guidance about relevant fea• • error analysis? While extensive error analysis has tures). David Wilkins will provide balance to the dis• been done on search algorithms, little has been done cussion by pointing out the limitations of chess and to quantify errors in evaluation functions and how claiming that Go is a better domain. He will also de- they interact with the search. scribe a new type of games tournament that prevents the human tailoring of evaluation functions and encour• • tool development? With the right tool, work that ages the use of learning and more robust approaches. might take days could be done in minutes. No tools The timing for this panel is particularly good with are being developed to help build chess programs. the current World Championship having completed, a For example, why isn't someone working on tools for more powerful Deep Thought on the scene, a recent defining chess knowledge? article in Scientific American [Hsu et al, 1990] and new If the community were committed to research, many books by Levy and Newborn [1991], and by Marsland of these problems would have been addressed by now. and Schaeffer [1990]. Sadly, much of the work currently being done on com• puter chess programs is engineering, not science. For Presentations example, the engineering of special-purpose VLSI chips to increase the speed of a chess program only underlines Computer Chess: Science or Engineering? the importance chess programmers attach to speed. In my opinion, conventional computer-chess methods Jonathan Schaeffer will yield little of further interest to the AI community. University of Alberta I believe they will be inadequate to defeat the human Research into artificial intelligence using chess as World Champion in a match for a long time to come. the application domain has produced several important It is still very easy to set up a position for which the contributions to AI: computer has no idea what is going on - even if you 548 Panels speed up the machine 1000-fold. The current computer Some fundamental AI questions that will remain are: chess work will only underscore the need for better ways • Given a patient and seemingly perfect teacher (that of adding and manipulating knowledge reliably. is a superior chess-playing machine), how should one The defeat of the human World Chess Champion use it to "teach" an Al-based learning program about sooner rather than later will help artificial intelligence. strategies for playing chess (given that the rules of This will help to re-establish chess as an ideal problem chess themselves are already perfectly known)? domain for experimenting with the fundamental prob• lems of artificial intelligence, as elaborated more fully • A related but perhaps simpler problem comes from by Donskoy and Schaeffer [1989]. the realm of endgame play. Given a perfect N-piece database holding an optimal move for each position "Expert Inputs" are Sometimes Harmful (or perhaps only the length of the optimal sequence Feng-hsiung Hsu from that position, or even less, whether the position IBM TJ. Watson Research Center was won), develop a program that can deduce a sound set of rules or strategies for playing the endgame per• Experience from the chess machine Deep Thought sug• fectly (or at lea#t better than any other expert).