This position was taken from Game 1 of the match. Deep Blue's move23 was P-Q5 (or d5 in algebraic notation). This strong move completed thedemolition of Kasparov's pawn structure; all Black's pawns were soon isolated and unable to support each other. Kasparov knew that 23. P-Q5 was a strong move, but he did not expect it from a computer, because it involved a pawn sacrifice -- something computers are often reluctant to do. However, Deep Blue, in analyzing the position, saw deeply enough to realize that 23. P-Q5 was only a temporary pawn sacrifice; that is, it saw that it would later win back the pawn and retain all the other advantages.

As figure 5.7 illustrates, however, computers can sometimes lack basic chess concepts that are understood even by amateur players. The diagram shows the final position in Game 6 of the match. Although Deep Blue was actually a pawn ahead, its pieces were all trapped, or immobilized. Deep Blue had not recognized the danger in this position many moves earlier, when there was still a chance to avoid it. If Deep Blue had not resigned, Kasparov could have won easily by, for example, opening up the king side and attacking the undefended king. The human ability to reason about permanently trapped pieces was a deciding factor in this game.

Although the competitive aspects of human-versus-computer play attract considerable attention, cooperation between man and machine is becoming more and more common. Many grandmasters use PC chess programs to help them analyze chess positions. And players can now learn more about chess endgames by studying computer-generated endgame databases that demonstrate perfect play in positions with five or fewer pieces on the board. But, perhaps most notably, Kasparov feels that the 1996 match with Deep Blue helped him understand more about chess. This may be a sign of things to come.


The Future of Computer Chess

Early optimism in the field of artificial intelligence led people to believe that the chess problem would be relatively easy to solve. In the late 1950s, Herbert Simon, one of the founding fathers of artificial intelligence, predicted that it would take only ten years for a machine to become world champion. Despite his expertise in the field, Simon was off by at least thirty years. After Kasparov lost a regulation game to Deep Blue, many people mistakenly assumed that the chess-playing problem had finally been solved. It is becoming more and more apparent, however, that chess mastery requires an intriguing mixture of skills: pure calculation, sophisticated evaluation, learning, and a generalized reasoning capability. Although a machine like Deep Blue excels in calculation, at present it still lacks many other skills essential to consistent world-class chess play. Until computers possess the ability to reason, strong human chess players will always have a chance to defeat a computer-style opponent.

Given recent advances in hardware speed and algorithms, I believe Kasparov's loss to a machine in a regulation match was inevitable. Kasparov still has the advantage in that he has the ability to adapt quickly to weaknesses in a computer opponent, a skill that current chess-playing machines lack. With continued progress, however, it is likely that we will see the end of competitive matches between man and machine sometime in the next century. Certainly competitive chess will continue: man against man; machine against machine. Ultimately, though, the computer's superiority over human players will be so great that the only value in man-versus-machine play would be the instructional benefit it provides human players, or -- as in 2001 -- ts recreational use on journeys to faraway planets. The applications that have been, and will continue to be derived from developing a world-class chess machine will advance our use of computers as tools for solving other complex problems. Even so we are still decades away from creating a computer with HAL's capabilities.


Acknowledgments

I would like to acknowledge the other members of the Deep Blue team: Feng-hsiung Hsu, the principal designer of Deep Blue, and A. Joseph Hoane, Jr. Other IBM Research staff that supported the project include C.J. Tan and Jerry Brody. My thanks to tcrain@s2.sonnet.com for directing me to the article by Grandmaster Larry Evans that includes the Frank Poole-HAL game in its entirety.