Author: Robert Hyatt
Date: 11:06:58 06/13/01
Go up one level in this thread
On June 13, 2001 at 13:31:43, Mark Young wrote: >On June 13, 2001 at 12:24:12, Robert Hyatt wrote: > >>On June 13, 2001 at 10:24:56, Mark Young wrote: >> >>>In this upcoming match next month Deep Fritz tops on the SSDF list will be >>>playing a 6 game match with GM Hubner (2620 Elo). >>> >>>This match should be good indication if computers are now grandmasters. GM >>>Hubner at 2620 is very close to the performance rating of the computers that >>>have played grandmasters at tournament time controls. >>> >>>Deep Fritz is a well-known program, so GM Hubner should have ample time to find >>>holes in the program and exploit them if he is able. >>> >>>Being match play should also help the Grandmaster if Bob Hyatt is correct. (I >>>also think this is correct). >>> >>>If Fritz puts in a Grandmaster performance in this upcoming match, the evidence >>>that computers are grandmasters start to become overwhelming. >> >> >>I would not disagree, unless Hubner goes hog-wild. IE the Kramnik match is for >>a small fortune. Kramnik will have a huge incentive to win. But he may well >>win by one game only, since that is all that is needed (if I were playing such >>a match against a computer, I would take all the 'easy' draws that came along >>until I reached a position that looked like it was winnable without having any >>unnecessary opportunities to lose as well). > >I agree, that why I think the Deep Fritz match will be more telling. Lucky for >us we can disagee all we want...but the data is coming whoever is right. > >> >>The Kramnik match will be interesting. I think he +could+ probably overwhelm >>DF. I don't think he will because the strategy for winning a match is to not >>try to win every game. Probably he would want to draw every game with black >>and play for reasonable winning chances with white. If this was not a match, >>but a series of 6 games with $100,000 per game for each win, the strategy >>would change. > >Yes I also agree, also Kramnik rating is so high DF only needs to draw a few >games to have a GM performance. What will be more telling in this match is if >DF-7 can win a game, somthing even GM Kasparov was unable to do. If Kramnik >Draws every game with black DF-7 earns a GM performance with ease. That is why >you need to take a closer look at Deep Junior at Dortmund were Deep Junior >played all, and every game was important. IF DF draws all games as black, and loses all games as white, I wouldn't _begin_ to say that is a GM performance. Rather, I would say it was just good match strategy by the GM to not try to overcome the disadvantage of moving second in those games. > >> >>Which means that in the Hubner match, Hubner might win every game, he might >>barely win the match (by playing very safe) or he might lose marginally or >>by a whopping score. But winning every game is not the goal in a 6 game >>match, and a GM will likely keep that in mind. > >I think for you to be correct, GM Hubner must win this match. We can argue later >about how much DF lost by if that is the case. The only useful information will be learned dependiong on which of the following happens: 1. Hubner wins handily. Fritz is "suspect". 2. Hubner wins. Hubner is better than fritz. Could be several hundred rating points better, since the draws could be strategy for winning the match at the cost of .5 points here and there. 3. Fritz barely wins. Fritz is very likely a GM-level player. Not necessarily anywhere near Hubner's rating, but still probably a GM, unless we all see something very ugly going on. 4. Fritz wins easily. Then it really must be a GM of some sort. > >GM Hubner has the advantages you subscribe in your thoery that kills computers >from being GM in your mind. ex. Knowing the computers weaknesses, match play, >etc. I don't know enough about his computer skills to make a comment there. But if the games are really 40/2hrs, with a secondary time control as well, then if he loses, it will make a big point IMHO. If it is just game in 2 hours, then things might be interpreted differently depending on how the human loses. IE ahead in the game, but blunders in the last few minutes of time scramble.. > >> >>This means that it is _much_ more difficult to judge the strength of the >>computer, since there is no way to compare match play. The machine will >>play every game as if it is the _only_ game. The human uses a different >>approach to attempt to maximize match-winning chances. > >Thats fine I concede match tactics...but GM Hubner better win this match. A win >or draw match by DF and your position will be suspect. I would agree. > >> >>If the computer wins, it might win because the human was better in most every >>game but he went for the safe "draw" only to lose the last game (IE DB/Kasparov >>in 1997). If the human wins by 1/2 point, it could be because they were very >>close, or it could be because the human was very conservative. > >No backing off now....If you are correct in your postings GM Hubner should have >no problem winning this match. I'm not backing off a bit. I am simply saying that _if_ the human wins, it doesn't matter whether he wins 6-0 or 3.5-2.5, since it is possible he could win 6 0 but he chose to take the 'safe road' with black whenever possible. IE if Hubner wins, he is better. How much better can _not_ be determined by the final match score. Which is why matches generally are not rated. IE if I win the first game, I will try _very_ hard to draw the remainder and win the match, even if I could win every game most likely. Because trying to win can clutch defeat from the jaws of victory, which is foolish in match play. > >> >>> >>>If fritz gets crush, and I would think this is what Bob Hyatt theory would >>>indicate from his postings, it will be time for us that think computers are >>>grandmasters to reassess. >> >> >>I don't think "getting crushed" is likely in a match. Because the strategy to >>win a match is not the same as the strategy to win the maximum number of games >>out of a set of N. Of course, the computer knows nothing about this so it is >>certainly possible that the human gets crushed. :) > >If you are correct, GM Hubner should be able to crush DF by playing closed >position in every game, if he so wishes. But I do concede match tactics. "crushing" is a relative term. Quite often it means locking the position up and waiting for your opponent to make a mistake. One cute strategy is to give a pawn to the computer and lock things up. It will totally wreck its position trying to hang on to that one pawn advantage without drawing. > >> >>If Fritz can beat Hubner in a 6 game match, it will definitely say something >>about the computer being a GM player. Not a final and convincing statement, >>but a strong one for sure. If the human wins, then the conclusion will be less >>informative. > >I agree, and will say more, if GM Hubner has a easy match were he is never in >trouble or crushes DF I will concede the point. But if I see DF drawing or >winning this match....other people need to take a hard look.
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.