Review Article
Total Page:16
File Type:pdf, Size:1020Kb
Amu Ban 201 ARTIFICIAL INTELLIGENCE Review Article A Chronology of Computer Chess and its Literature Hans J. Berliner Computer Science Department, Carnegie-Mellon University, Pittsburgh, PA, U.S.A. on While our main objective in this article is to review a spate of recent books surveying computer chess, we believe this can best be done in the context of the field as a whole, and pointing outvaluable contributions to the literature,both past consider and present. We have appended a bibliography of those items which we to constitute a definitive set of readings on the subject at hand. Pre-History instance was Even before there were computers there was computer chess. The first which played a famous automaton constructed by Baron von Kempclen in 1770, excellent chess. It won most of its games against all comers which supposedly included Napoleon. The automaton consisted of a large box with chess pieces on fashion, top and aproliferation of gear wheels inside, which were shown in magician 1 However, one compartment at a time to the audience before a performance. a skilled (and small) chessplayer was artfullyhidden in the midst ofall this machinery, and this was the actual reason for the "mechanism's" success. In contrast, a genuine piece of science was the electro-mechanical device con- of structed by Torres y Quevedo, a Spanish engineer, in 1890 which was capable mating with king and rook vs. king. Its construction was certainly a marvel of its day. The Dawn few, Shannon In 1950, when computers were becomingavailable to a selccl Claude describing a chess then of Bell Telephone Laboratories wrote an article how Artificial Intelligence 10 0978), 201-214 Copyright © 1978 by North-Holland Publishing Company 202 H. J. BERLINER A CHRONOLOGY OF COMPUTER CHESS 203 program could be constructed [20]. He hypothesized three different types of was to a great degreeresponsible for my quitting industry to seek higher education program: in the field of Artificial Intelligence. The culmination of the new enthusiasm in Type A. A program that did a complete search to a given depth and used a computer chess came with the organization in 1970 by Monty Newborn in New simple terminal evaluation function, York City of the First United States Computer Chess Championship. Since 1970 Type B. A program that searched selectively by eliminating branches (laterknown there have been annual U.S. Computer Chess Championships (held concurrently as forward pruning) that appear to have little promise a priori, with the ACM annual meetings), and a World Computer Championship series was begun Stockholm Type C. A program that was goal oriented, although it was not specified how at IFIPS in 1974. such a thing could be done. Unfortunately, Greenbiatt, who started it all, could not be prevailed upon to enter his program in any of these events. During this period, the Northwestern A year later, Turing [21] wrote a paper detailing his ideas (some very similar but University chess program, programmed by David Slate and Larry Atkin, and arrived at independently), describing and his (hand simulated)program Turochamp, variously named CHESS 3.0 to CHESS 4.2 dominated the chess scene. It won and one of its games. Both writers mentioned theproblems associated with attempt- every U.S. Computer Championship except the 1974 event. In the process it ing to evaluate positions that were not in a quiescent state. convincingly defeated all programs it encountered, leaving only a small doubt about how it would have fared versus Greenbiatt (who was improving his program) Ancient History and the program that was being developed at the Moscow Institute of Control From 1955 to 1958 the first actual computer chess programs appeared. There were Sciences. three programs which corresponded exactly to the three categories that Shannon Part of this doubt was resolved when CHESS 4.0 stubbed its toe in a game against had specified. A Type A program was constructed by a group at the Los Alomos CHAOS (a perennial contender written by I. Ruben et al. originally at UNIVAC) 1 Laboratories (the best understanding ofwhat can be done with a primitive Type A in the first World Computer Championship in 1974. This gamefeatured aprepared program can be gained from [9]), a Type B program at IBM by a group headed by opening variation (a la human masters) and a fine sacrifice by CHAOS. This Bernstein [6], and a Type C program by Newell, Simon and Shaw' [l4] (see also ultimately led to KAISSA, the Moscow program, winning with a perfect score, but some very cogent remarks in [13]). Unfortunately,all three oftheseprograms played without having met CHESS 4.0. A specially arranged game between KAISSA and very mediocre chess, for reasons that are now quite clear, but were hard to CHESS 4.0 ended in a draw after several vicissitudes, thus leaving many questions appreciate at that time. Because of this lack of success, computer chess was for a unresolved. time thought to be too difficult to tackle effectively, although an isolatedBatchelor's Two meaningful books published during this period were [7 and 12]. Botwinnik thesis was produced by Kotok 2 at MIT under the direction of John McCarthy. (a former World Chess Champion and one of the great players ofall time) describes This program, transplanted to Stanford University alongwith Professor McCarthy, a very advanced method ofconceptualizing chess positions in the human style. His played and lost the first Computer Chess match played by mail with work is very crisp and well thought out. Unfortunately,it dependsupon many lesser a program ! from the Institute of Control Sciences in Moscow. problems having already been solved, which happens as yet not to have occurred. I feci confident, however, in asserting that it will be a long respected work in the The Modern Era chess literature. Newborn's book is more historical in nature and deals with what It program The present thrust in computer chess can be dated back clearly to 1966 and the one can call the start of the Modern Era. describes several alternate on an efforts of Richard Greenbiatt [10] at MIT. He wrote the first program that could structures, gives a lot of background the early competitions, and is in short readable, command the respect of actual chess players. It was a fine piece ofengineering and excellent introduction for the educated layman. It is eminently and has treated the quiescence problem better than any ofits predecessors had. MacHack-6 a fine bibliography. (as Grecnblatt's program was named) soon played in human tournaments, estab- In this period other computer programs began to participate in human tourna- lished a performance rating in the 1400-1500 USCFrange (Class "C") and showed ments, a fact which may have encouraged what I consider a rather one-sided bet in that it was capable on occasion of beating tournamentcaliber players. Its achieve- 1968 between Al-persons Michie, McCarthy, Papert and Kozdrowicki on the one ments spawned a broadside of efforts all over the world, and I must admit that it hand, and David Levy a Scottish chess master on the other. The scientists bet that 1 to Levy a years. Kistcr, J. et al., Experiments in chess, Journal of the 4 (2) (April 1957) 174-177. a computer would be able beat in match within 10 2 Kotok, A. (1962) A chess playing program for the IBM 7090, UnpublishedBachelor's Thesis From 1970 on, one could readily perceive computer programs improving [16], MIT 1962. as evidenced by games against each other. However, there were few occasions : ACM, OF CHESS 205 204 H. J. BERLINER A CHRONOLOGY COMPUTER which allows it to generate where programs played in human competitions, and perform such operations in about 40 nano-seconds nothing substantial to indicate any program. Thus it can that measurable moves, update boards, etc., many times faster than other progress in climbing the human ladder was being made. reached by other run as a simple Type A program and achieve depths not yet to make a real The Jet Age programs. However, this has apparently not yet been sufficient This, together with CHESS's experience, would In the fall of 1975, CHESS 4.4, the latest incarnation of thepowerful Northwestern inroad into playing strength. are best accompanied with knowledge University Chess program, again won the U.S. Computer Championship. This appear to indicate that brute force increases confirmed that CHESS 4.4 was back in the saddle again, avenging its loss the increases. , . , , . „ ... to play according to the "Method ofAnalogies [1 J. previous year to RIBBIT (from the University of Waterloo). Nothing else note- KAISSA has been modified program to avoid searching sub-trees that are considered worthy apparently occurred. However, it marked the beginning of the latest era. This method allows the sub-trees which a definitive result is already known. This was the first time that an incarnation of CHESS had run on a super-fast to be very similar to for appears to be a way of thefuture and did work out very well computer (one of the CDC CYBER 170 series). Although it apparently did not Although this method Computer Chess Championship at Toronto, 1977, need this edge to win the tournament, one can in retrospect see an improvement for KAISSA in the 2nd World CHESS 4.6's strength and the latter program won over its earlier versions in the way it man-handled David Levy in a simultaneous it was not sufficient to overcome score, defeating KAISSA in a reciprocal side game exhibition given by that Master after the tournament. Although the gamewas only the tournamentwith a perfect meet regular competition) after the event.