Lecture by UMBC Chess Coach Igor Epshteyn
In February 1996, Gary Kasparov defeated IBM's Deep Blue computer chess program 4-2 in a regulation-style match held in Philadelphia, as part of the 1996 ACM Computer Science Conference.
In Game 1, the computer outplayed Kasparov positionally and tactically, when Kasparov played in his usual aggressive style. In Game 2, Kasparov changed his style and allowed the computer to proceed according to its own plans (or lack thereof). Kasparov won in an interesting ending, though due to programming errors the computer in that game played without any opening database. Two draws followed. Kasparov's second win came in Game 5 after the IBM programmers refused a draw offer even though Deep Blue considered the position roughly equal. In this game, Deep Blue demonstrated its lack of understanding of the danger of a kingside pawn majority. In Game 6, Kasparov totally outplayed Deep Blue by gaining a decisive space advantage and by avoiding any weaknesses. In this game, Kasparov trapped the computer's rook and bishop, in part because the IBM programmers had never properly adjusted a parameter that signals when bishops are trapped.
Deep Blue is an improved version of the older Deep Thought, augmented by parallel special-purpose hardware. Deep Blue uses a selectively deepening search strategy, using improvements of the alpha-beta search strategy, with powerful evaluation functions. Transposition tables help avoid unnecessarily calculating the same position more than once. Two powerful databases further augment Deep Blue's play: an opening database includes all recorded grandmaster games from the past one hundred years, and an endgame database enables Deep Blue to play all endgames with five or fewer pieces perfectly.
Contrary to some media reports, Game 1 was not the first time a computer defeated a Grandmaster in a slow game. For example, on March 24, 1995, Grandmaster Gennady Sagalchik lost to *Socrates at UMBC's Man versus Machine Match.
Detailed information about the match is available from IBM's WWW pages, including game scores.
For more about computer chess, see UMBC's graduate seminar in computer chess.