What Deep Blue Taught Kasparov - and Us
Rebounding from defeat last year, the Deep Blue supercomputer has won a signal victory over chess champion Garry Kasparov - showing how far it has come, but also how far it still has to go to achieve its enormous potential.
Mr. Kasparov's performance also highlighted strengths and weaknesses, from the amazing ingenuity of the mind to the pitfalls of erroneous thinking that bedevil humans - both on and off the chessboard. The news from this match is good, showing that man-machine interfacing can have salutary effects for the advance of human as well as artificial intelligence.
Deep Blue came to the match with enhanced quantitative computing power. It can now assess 200 million positions per second, about 100 million times human calculating ability. Its tactical powers, move by move, have improved substantially. More important, Deep Blue has also been programmed with a new set of qualitative skills for assessing strategic aspects of the game.
This substantial shift toward conceptual understanding made Deep Blue far stronger than it was last year. Its subtle handling of the complexities of Game 2 led to a strategic masterpiece comparable to the greatest performances of any human chess master. Its tenacious, resourceful defensive maneuvers in Games 3, 4, and 5 also showed a sophisticated awareness of the positional nuances that lie at the heart of chess.
Kasparov also showed signs of growth from last year, clearly prodded by lessons that he learned from his first match with Deep Blue. While his career has been built on tactical sharpness, his understanding that he could not outpunch Deep Blue led him to rely more on strategic judgment than sharp, direct attacks. As a result, the world champion's amazing chess powers have broadened even further. Indeed, Kasparov himself is one of the most devoted users of computers for analysis and match preparation.
As both Kasparov and Deep Blue beefed up their strategic, positional thinking coming into the match, how did their respective capabilities match up?
Despite Deep Blue's magnificent advances in conceptual logic, the human player outshone the intelligent machine as a strategist. In Game 1, Kasparov's profound understanding of the nature of position allowed him to make a sacrifice, which could not be recovered, but which gave him a winning advantage. In Games 3, 4, and 5, Kasparov outplayed Deep Blue each time, with the machine clearly on the defensive and in trouble. In these games all of Deep Blue's quantitative computing power proved of little moment.
If he did so well, why did Kasparov lose?
The answer is not to be found in the machine's improvements, since the champion had made comparable improvements in his own game.
Kasparov lost because he believed - wrongly in this case - that he could not win. In Game 2, he resigned in a position that could have been defended successfully; in the middle games of the match, he overlooked straightforward winning moves, allowing Deep Blue to escape with draws. Thus, this match highlighted less the implacable strength of the intelligent machine than the boundless creativity of the mind.
Some lessons for, and from, chess can be derived from this match.
First, it must be recognized that the adversaries are both man-machine teams - with one sending a man to the board. With this in mind, future matches of this sort must allow for adjournments, as is the practice in professional chess. This would allow the human a reasonable time for rest and introspection, and afford him the opportunity to consult advisers - human and machine - before returning to finish a particular game.
Beyond the chessboard, "Blue 2" affirms the rising importance of the qualitative aspects of high-performance computing relative to brute-force calculating speed. In practice, this means directing research funding to more conceptual projects. If, for instance, path-breaking new military efficiencies are to be realized in an era of budgetary constraint, their seeds will have to be sowed in studies that blend insights with number-crunching.
Finally, this latest chess match should foster the realization that a profound, and yet harmonious, relationship between humans and intelligent machines is emerging. As the computer spurs ever-greater human creativity and insight, humans, in turn, can help their thinking machines begin to contemplate the sublime.
* John Arquilla is professor of information science at the United States Naval Postgraduate School in Monterey, Calif.