Author: Mig Greengard
Date: 22:38:23 08/24/01
Go up one level in this thread
On August 25, 2001 at 00:04:53, K. Burcham wrote: > >i cannot imagine how stefan felt watching this game. shredders score > quickly went up 5 points, and maintained a five point advantage for quite > some time. stefan had to feel very confident. somewhere about move 60 > stefan probably witnessed his score drop tremendously. there must have > been a lot of surprised chess enthusiasts in the game room. I talked about this during this game and a few others I commented live at KasparovChess.com. I enjoy comparing how computers play and solve problems compared to how a strong human would go about it. Shredder had an undeniable advantage for most of the game, and a clearly winning advantage after it won the exchange. It then started doing things that a strong human would never consider for purely pragmatic reasons. I mentioned at the time, while Shredder was still well in the plus, that it was playing with fire. (Gromit did the same thing, turning a simple draw into a wild ride that it lost to Junior.) Your first obligation when you have such a position is to eliminate counterplay. This isn't 100% possible against a strong opponent, but you certainly work to simplify the position in order to keep control. The last thing you want is to see the position sharpen so much that, despite your objective advantage, one error in calculation can cost you the half or full point. Which is exactly what happened to Shredder. It's eval did not swing so much because it evaluated the position incorrectly when it was at +5 (although it does tend to be overoptimistic, like many programs, which is why I have so much admiration for Junior's more conservative, and generally more accurate, evals), but because it allowed the position, despite being advantageous, to get so sharp that each half-ply meant life or death. This is often fatal for humans and computers alike. Mikhail Tal would stir up unfathomable complications on the board because he trusted his instincts and because he knew he could calculate better than anyone in the world. But even Tal knew when it was time to simplify into a winning endgame! No matter how well you calculate, you don't give your opponent a batch of passed pawns (if you can help it) if only because nobody is perfect and you just might have missed something. Programs are vulnerable here because they always play the "best move," and not the most practical move. They can't see the inherent complexity of a coming position, just the eval. Don't blame Shredder for the Rd8 blunder; that's the symptom, not the disease that all comps face. Programming common sense is much harder than programming genius. It would be interesting to have an uncertainty variable in the mix that would help a program distinguish between a line that is winning with low risk and another that is a tightrope walk. Saludos, Mig http://www.kasparovchess.com
This page took 0 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.