Results of the nineteenth ACM North American computer chess championship

Reviewer: Teodoro Oniga

In a final note of his book Cybernetics [1], Norbert Wiener anticipated by two decades the realization of computer chess contests; in November 1988 the 19th ACM North American Computer Chess Championship was held in Orlando, FL. Commenting on the results of this intersoftware chess competition, the authors report that DEEP THOUGHT 0.02 (D.T.), playing “tough” chess (equivalent to a rating of at least 2580), won three of its four games and drew one against CHESS CHALLENGER X, capturing first place. The paper presents five of the games, two of them with detailed commentaries (and also with some misprints, such as “21. g x h5 b x h7, instead of “21. . . . g x h5 22. B x h7” and “23. . . . xhg4,” instead of “h x g4,” both on page 1228). The report is rather terse and does not make clear, for instance, if there were only four rounds, or 11 as would be normal among 12 players, if the final column of the score table indicates real points won by each program (10 points for the first classified would then be equivalent to 9 wins and 2 draws), or if the rules were exactly those adopted by the International Chess Federation for international chess contests. The information given is too sparse for the reconstruction of the whole picture. The fact that in the week following its success in Orlando D.T. finished in a first-place tie with Grandmaster Anthony Miles, ahead of five other Grandmasters (including the former World Champion Mikhail Tal), does not mean that it is invincible. In a two-game match via electronic mail it lost both games to Mike Valvo (who served as tournament director in Orlando) and, more recently (in October 1989), it was defeated in 90 moves by the current world champion Gary Kasparov, notwithstanding the extraordinary analytic power of the SUJ 4 plus two special processors, which searched up to 720,000 chess positions each second. These defeats show that D.T. is not so deep as to outpace human imagination and combining power, even if, as Richard Re´ti demonstrated long ago, a human player usually cannot anticipate more than three moves in normal legal time (with only three possibilities for each move, this means 3 6=729 variants to be analyzed), and it seems inconceivable that an artificial player could ever imitate Andersen in his immortal game against Kieseritsky in London in 1851 [2].