What a way to win
14 October 2007 10:30 AM CET | Last modified: 21:00 | By Peter Doggers | Filed under: Uncategorized | Tags: Computer chess, Match, Odd
During during the World Championship in Mexico, the computer match Rybka-Zappa was played. In the fourth game something happened that made me lose interest in this match (and actually in computer-computer matches in general). In Chess Today issue 2530 (which came out last Friday) IM Andrey Deviatkin analysed some games of the Rybka-Zappa match and after reading this, I feel obliged to tell the “truth�? about game 4.
We paid attention to this match right before it started by publishing the press conference, and now we return to it. As you probably know, the outcome was a surprising 5,5-4,5 victory by Zappa. (You can download the PGN file with the ten games here.) The main explanation for their success, given by the Zappa team, was the fact that the engines were playing on 8-core machines. “Zappa is slightly stronger than Rybka on these platforms.”
It’s not that I want to treat Zappa’s victory with disregard – beating Rybka in a match is a fine achievement – but what happened in game 4 resulted in some scepticism on my part about games between computers. First I give IM Deviatkin’s analysis.
The course of events was a bit different. I was standing next to the board and computers, chatting with the operators and the arbiter, while the engines were calculating their moves. Around move 100, the Rybka operator offered a draw to the Zappa team. (An interesting sidenote: if the Zappa team had agreed, the arbiter David Levy would have had to agree on the draw as well. One of his tasks is to check if the engines themselves both think it’s a draw, not just the operators!)
But… the Zappa operator declined the draw offer! Quite surprisingly, since it had zero winning chances. (It did have a reliable fortress, as Deviatkin writes, but certainly not more than that.) The good thing is that I could ask right then and there what was going on, since engines do not have ears.
Here it comes: the Zappa operator knew that if the game would continue for a while, soon 50 moves would be played without pawn moves or captures. He knew that Rybka would then sacrifice a pawn, because the engine understands the 50-move rule and since its evalutation was higher than 0.00 (because of the material plus) sacrificing a pawn was the only way to continue the game. So by declining the draw offer, the Zappa team knew they would soon win a pawn! This remarkable strategy eventually got Zappa the full point and got me leave the playing hall in disbelief.
Epilogue: It’s seems strange that the arbiter has to approve on a draw, but not on offering, accepting or declining a draw offer. One way to improve these engines even more is perhaps to program them to offer a draw themselves. But well, then such a match would be all too inhuman…
OK, lets say that you would live in a country where all...
I think that this is an error of evaluation of Rybka engines.
Sun Tzu said know your enemy, this what Zaapa team did.
Yes, you could say this was an evaluation error, since otherwise Rybka wouldn’t have traded the pawns to avoid 50 moves rule. Normally this 50 move rule play by Rybka is quite ok (look at the first game against GM Fontaine), but in this situation everything went wrong :(.
Whose idea was it to teach computers to play chess?
Get the point, people! The operators won this game, not the program. How about if Anand had someone sitting next to him making the draw offer decisions. It’s absurd!
Computerprogramma’s zijn er om mee te oefenen!! Zo’n computer vs computer match vind ik niks.
Actually this article just shows how unfamiliar the author and the various idiots who commented on it are with computer chess.
The standard algorithm in computer chess is ALWAYS: play on until a forced draw, by repetition/50 move rule/etc.
It was ok to continue to play. Rybka was kind of put to a specific test and failed, in that game. But I think, the true problem was that there were misevaluations involved, of the positions after the critical pawn moves. There certainly isn’t an ‘automatism’ that pawns are sacrificed just to avoid a 50 move draw, not even if the eval is better than 0.00. Decisive is if the engine thinks the position will still remain better AFTER the pawn sac, because if not, it wouldn’t sac it but rather draw.
Like +0.50 before the sacrifice but -0.50 after it: In that case it of course would NOT be played, because 0.00 is better. But Rybka’s evals were for example, 109. h6 {0.86/24 25} and after only 2 white pawns remained, still 123. Kc3 {0.39/20 18}.
In other words, Rybkas eval was too optimistic for Q+pawns versus RRB, and I guess that will be repaired.
Interestingly, if we look at Zappa’s evaluations, we find: 109. h6 {(Df4) 33} Rxh6 {0.73/20 35} (=white advantage too, after the first pawn sac). But: 122. Qe5 {(Dc5) 75} Rxb6+ {-0.08/16 29}, here Zappa’s first evaluation to Black’s advantage in that game, after Rybka had wasted 3 pawns…
(the evaluations are from the Rybka forum and in the Hiarcs forum, where the PGNs have been posted)
Huh? Computers play chess?? Why?
http://www.sonshi.com/learn.html “Sun tzu’s art of war”
“Therefore, know the enemy’s plans and calculate their strengths and weaknesses. ?” The Zappateam discovered a weakness in rybka’s software and they took advantage of it.
I don’t understand why Peter is complaining.
@Martin Why do they still offer draws in such positions then?
@ schaakje Not complaining. Just expressing astonishment.
I still think chesscomputers are just mere tools to practice your game. Just give me a game between these guys :Kramnik, Anand, Topalov and Ivanchuk.
Well Martin just made a huge fool of himself.
Why of course, every world class chess engine abides by the ’standard’ chess algorithms, because they actually exist. Oh definetly.
Silly computers still cannot play chess the correct way. Winner and loser of this match both looking foolish, ha ha ha!
Reconsidering, I might be entirely wrong, since Rybka showed just all-too-human characteristics: unable to reconcile itself with the fact that it had spoiled a winning position, it continued to try beating water from the rock, resulting in the opposite. These engines are perhaps rapibly gaining in humaness…
I believe, the only way is to exclude the operator teams and the arbiter from the process. Programs should have played on a chess server; all the chess community would be happy to watch it…
Remove the opening books too, those are made by humans, anyway, and do not show the real strength or weakness of the software.
Look, you tools are clearly not figuring this out.
A computer doesn’t get tired. There is absolutely no reason not to just let it play until the theoretical end of game. The only reason to offer a draw is if the operators get tired
why astonished peter?
did you never used an opponents weakness to take advantage of so you could beat him (eg openingsvariation X or Y)?