Author: Amir Ban
Date: 07:30:00 02/15/03
1) Why did you choose black for the first game? Kasparov was the favorite to win in the general opinion as well as in his own opinion, and to do that you have to win with white. We did not want him to have a black game to "test the waters" but to put the burden of proof on him right at the start. 2) What computer hardware was used in each game? We had three computers in New-York: Intel Xeon 4 x 1900 GHz Intel Xeon 8 x 1600 GHz AMD Athlon 2 x MP 2200+ In practical performace the three machines were quite close to each other, with the quad showing a slight superiority. All games were played on the quad machine, with the other two taking turns at being the backup machine. 3) How this match and contact with Garry Kasparov will influence the further development of the next versions of Junior ? No changes , big changes, little changes ? This question is a bit premature as Junior 8 has not been released yet. I hope that the public will take an interest in it. As for me, my mind is usually somewhat blank after completing a big challenge, and I can only quote Scarlett O'Hara "Tomorrow will be another day". I have learned a lot from the experience, but I'm not ready yet to take action on it. The challenge for Junior will continue to be as it has been for some time now: To transcend the goal of matching the best human play and demonstrate things that were not previously known about chess. 4) What was the last book move in each of the six games? Some comments about this 6 positions ? I already answered this in my posted comments on the games. 5) Did you learn something about junior in this match or did it performed as expected ? It's not possible to isolate the performance of Junior in this match from its opponent: Garry Kasparov. We have had enough exposure to grandmasters in public and test games and if Kasparov had been an ordinary GM, things would have been quite different. But he was an unknown, and a scary one at that. My conclusions about different aspects of the game are a bit of a surprise to me too: Opening preparations: Kasparov is no. 1 in this regard, and he showed what this means in game 1. However, Boris put up an excellent fight and in the second part of the match it appeared that we were getting the upper hand. Tactics: Strangely enough, towards the end of the match I realized that Junior is not outplaying Kasparov tactically. I have seen enough games against grandmasters where this happened to realize that nothing of the sort is happening against this particular opponent. For this reason I was less enthusiastic about game 5 than others arguing that early complications lead us nowhere. Positional understanding: perhaps even more surprising than tactics, Junior was holding its own in positional play, the best examples of which were games 3 & 4. We have had our experiments in the past with so called anti-GM strategies and with avoiding closed positions and everything else that according to conventional wisdom is supposed to be unsuitable for computers and all these were left in the wastebin for this match. Endgames: This aspect of the game was hardly discussed in this match. In the end Deep Junior played not only as strongly as a super-GM (a debate that has been going in this forum for several years too long), but actually as a super-GM. I've seen opinions expressed that the programs are 2500 in understanding and 2900 in tactics, so you get an overall 2700 performance. I'm not buying it. The same for some comments like "typical computer move" which in some cases were so unthinking that they were seemingly generated by a computer. I believe that the two different ways of playing chess: human chess and computer chess are starting to converge at their highest level. To be sure, the programs are from time to time conceding their silicon origins, but in the same way we humans must admit that when events exceed our understanding, then something mechanical in our own thought processes reveals itself. 6) Did you change parameters of DJ8 (besides the opening book) between the games? If yes, which changes (roughly) did you make? Decrease it's aggressiveness in the last game? There is no aggressiveness parameter in Junior. I varied the contempt factor during the match, starting at 0 in the first game, going up to 15 in the next 4 games, and in the last game contempt was set at 25. The match was a period of intensive activity for us. Five different versions played, and the only time a version was not changed was from game 3 to 4. Opening book work was especially intensive before game 5, when Boris decided to abandon his Semi-Slav preparation and spent two days guiding Junior through the Nimzoindian and Queen's Indian. Incidentally, I expected Kasparov to open 1.e4 in at least one game, until Boris Alterman told me that nobody plays e4 against Junior. This seems to be true. The last one to do so was Leko (Dortmund 2000), and he lost. 7) Is the software in the match programed specifically against human or does it perform well also against computers? The code is not developed with any specific opponent in mind, but when testing and comparing various test versions we put in considerations that were more sophisticated than reading the result, and we recruited Boris Alterman for help in that. His input was often highly non-trivial, e.g. he once dismissed a version that won big against a well-known SSDF star as being based on nothing more than cheap tactics, and recommended a version based on a match that it actually lost. 8) was this version of Junior specially tune for a very fast hardware ? No, and the hardware we ran on was not especially fast. E.g. Shredder had a dual Athlon 2100 in Maastricht, and the hardware here was not more than 30% faster. Amir
This page took 0.03 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.