Kasparov to take on Deep Blue again in May

Mark Crowther reports

In New York Gary Kasparov announced that he would play another match against IBM's Deep Blue. The winner of the match will take $700,000 and the lost $400,000 the prizefund reflecting the level of publicity that the last event obtained.

The contest "IBM Chess Challenge Rematch" will probably place as much strain upon the team responsible for Deep Blue as it will upon Gary Kasparov. They will hope to at least equal the one victory obtained last time.

The match will take place in "The Millennium" 44th and Broadway, New York City USA between the 3rd and 10th of May 1997. The match will be covered on the www.

Kasparov played Deep Blue in February 1996 and got off to the worst possible start by losing the first game. He equalised in the second game and was under pressure throughout the entire match. He won game 5 only after his draw offer was turned down. The computer proceeded almost immediately to wreck its position and lose extremely badly. In game six it was beaten very convincingly by Kasparov who managed to employ a very pure form of anti-computer strategy in this final game.

The match reflected the general understanding on how strong computers have become. Deep Blue played well in positions where tactics predominated, although Kasparov is also a highly gifted in this area he did not manage to cope with the relentless nature of the tactics in game one. His wins in games 2,5 and 6 were all as a result of the computer's short term tactical approach being the wrong type for assessing a position correctly.

The team of Deep Blue scientists is lead by senior team manager Chung-Jen (CJ) Tan, research scientist Feng Hsiung Hsu (whose name, certainly to those who have been on the internet for a few years, is most closely associated with the project), Murray Campbell, Joseph Hoane and Jerry Brody. Joel Benjamin is the resident Grandmaster associated with the project.

The incarnation of Deep Blue used in February calculated 200 million moves per second. The team believe that this speed will be enough and will be concentrating on making the machine more flexible in its response to Kasparov's approach and on "improving its chess knowledge". This last phrase is almost certain to mean a lot of hard work on the construction of its opening repertoire. Deep Blue's last outing before playing Kasparov last time was in the World Computer Chess Championships where its inferior opening repertoire (aswell as the short nature of that event) were widely held responsible for it not taking the title. If Deep Blue can avoid the kind of position it got out of the opening in game 6 of the match it will have a chance, one almost might say a puncher's chance, otherwise it will be Kasparov who wins, possibly by an even bigger margin than last time.

Yesterday the programers were bullish about their chances. Referring to the losers prizemoney Tan said "The $400,000? You'd have to ask Garry." He regards the last match as an experiment whereas this match will all be about winning. In the last match the final configuration was only available to the programers a short time before the match, this time they will probably have more time to test the hardware and software together. However their confidence in their machine has not yet extended to allowing it to play in public since the last match. Part of the psychological problem that Kasparov will have to overcome is that he will be playing an opponent of unknown strength, he may in fact, give it too much credit. Certainly his win in game five in the must have shocked him, he offered a draw believing that the computer could not err in that position, within a few moves it was lost. It almost seems like a lottery whether it plays well or not. If Deep Blue were to play a training match and be soundly beaten then not only would Kasparov learn from that, he would be less afraid. IBM's confidence in its technology will be interesting.

A prediction? I take Kasparov to win, I think he will have learned much from the last match and the computer will have to be much better to even score the same result.