Kasparov vs Deep Blue
by Mark Crowther
Between May 3rd and May 11th Gary Kasparov will contest a match over 6 games against a computer program called Deep Blue. The winner will take $700,000 the loser
$400,000. Kasparov defeated a program by the same team last year by 4-2 but which saw Kasparov lose the first game and not establish his winning margin until the final two games
of the match.
The biggest chess story of the decade?
Like it or not the match between Gary Kasparov and Deep Blue is likely to be the biggest chess story in many years. Down in my local pub the landlord said to me "Did you see
Tomorrows World tonight?" it was about the match. He was impressed, "you should see the machine when they took it to the venue" he said "it was massive and it calculates
2,000,000 moves a second ... or something. But why are they putting all this effort in? When the match is over they'll just put it in a cupboard, all that effort going to waste."
I explained that the chess computer was simply testing the computer. It was a powerful machine which was new and being prepared for more important projects. Besides, the publicity
would pay for the effort put into the project. Even I, who could be said to be anti-computer chess, had got caught up in it all.
The media are going to find it much easier to cover this match than recent World Championships matches. Chess gets covered when there is an angle. Fischer-Spassky 1972 (east vs
west), Karpov-Korchnoi (communist party man vs defector), Karpov-Kasparov (communist vs reformer) are all examples. The labels often didn't quite fit but chess relies on such
stories for main stream coverage. Man vs machine, HAL from 2001 and technophobia will all fill space in newspapers. Champion vs rather nice Indian challenger didn't play so well in
the media during Kasparov's last proper chess match. The more colourful Kamsky would have played better in the media.
I've done a lot of reading. Computer chess is not something I have put much effort in to understanding prior to this. I have even gone as far as to buy a text book on Artificial
Intellegence (AI). [One nugget I dug out is that one field of AI is aimed at proving it is impossible.]
To be fair to IBM their publicity and web site is scrupulously fair. Their branch of artificial intelligence is simply aimed at doing a task better than the best humans. Nothing to do with
intelligence, except that of the programmers and technicians themselves (for instance the computer will not learn from its mistakes, the programmers will have to try and understand any
errors and eliminate them between games). This time is the real push, last year's match accomplished much (the first defeat of Kasparov by a computer at normal time-rates) but this
time they claim they are really ready.
The computer program Deep Blue has not fundamentally changed since last year. It is based on parallel processor technology allowing simultaneous calculations to be made thus vastly
increasing the number of variations per second that can be made. This year there will be a modest doubling of the number of calculations it can make. This is a significant but not an
overwhelming improvement. The problems in getting this all to work are not easy however and having access to the computer in full playing trim has been a problem in the past. No
top level micro-computer would lose contact with its opening book as Deep Blue did last year. But if you have to set up the entire system in a few days then you're always going to
have these problems. This time I expect things to be different, the program will have been tested and the whole system refined to work much better. The computer makes assessments
of each position using an evaluation function. If this evaluation is refined the performance of the computer program is enhanced greatly. The assessments and opening repertoire have
been worked on by the programming team. Grandmaster Joel Benjamin has been aiding them with this for months, there is little doubt that the machine will be improved.
Improved from what however? The only games we have to go on are the 6 that were played in February last year. The program received mixed reviews. The next best programs are
only of international master standard and there are possibly only 2 or 3 of those.Can it be true that the best computer program (all be it running on a vastly more powerful computer)
can not only be of Grandmaster standard, but of Super Grandmaster standard?
In addition a test version of Deep Blue playing in the World Computer Championships only scored 3.5/5 in 1995. However it was only a short event and it was there for testing only.
Nevertheless it showed the computer program had weaknesses, even against other computers.
Philadelphia February 1996
This previous match with Kasparov established Deep Blue in many eyes as the best chess playing computer in the World. A loss, but there were some signs of real improvement for
the program. How did the match progress?
1) Kasparov played poorly in the first game. He was completely thrown by an unusual move from the computer early in the game and never recovered his equilibrium.
2) The next was the game where the opening book did not work properly in the computer. The computer found its way through some of the problems but was short of time.
Kasparov won efficiently in the end. 3) The third game saw another glitch at the start of the game with the computer being rebooted. Kasparov built up a large advantage but the
computer defended very well and held the draw.
4) This was possibly the best game of the series. Both Kasparov and the machine would have been happy with the type of position in the game. Kasparov got an advantage but it was
complicated, in the end he got short of time and had to force a draw in a position where he would have otherwise been much worse.
5) The fifth game showed just how little we know about computers. A level opening produced a draw offer from Kasparov. The computer's handlers turned this down even though the
computer thought itself slightly worse. It then proceeded to play dreadfully and lose.
6) The final game saw Kasparov get an opening that the computer could not handle well. It misplaced all its pieces and was crushed.
The last match was too short to fully assess the computer, or indeed the approach that Kasparov used. He will be much better prepared this time with the example of the six games
along with the knowledge that the computer is dangerous. In addition his form in international chess was patchy at the time, he has since become excellent with two of his most
convincing victories in international chess.
If one accepts that the last match was a fair reflection of the computer's strength then there isn't a large distance between 4-2 and a tied match.
The efforts of the programming team have been far greater this time, and they have had a good working machine to start with. They indicate that the new version of Deep Blue (they
call it Deeper Blue in private) is beating the old version handily. They have made the machine more flexible so that it will be easier to make changes to the settings between games and
it will even play quicker if Kasparov runs short of time. The programmers exude quiet confidence (mind you they did last time). A prediction of Kasparov versus a computer which
whilst better than last year has no known strength is hard. However this is not black magic. The computer is not going to gain in strength to the point where it is unplayable for instance.
With Kasparov better prepared and in better form the result ought to be the same or better for him. So I go for 4-2, possibly with no losses this time but with the standard of play
increased. We'll know better after game one.