Author: Mark Young
Date: 18:22:16 06/25/01
Go up one level in this thread
On June 25, 2001 at 21:05:42, Robert Hyatt wrote: >On June 25, 2001 at 15:16:11, Mark Young wrote: > >>On June 25, 2001 at 12:03:03, Robert Hyatt wrote: >> >>>On June 25, 2001 at 09:30:37, Graham Laight wrote: >>> >>>> >>>>I think that people who've been around a long time might disagree with this. >>>> >>>>The first victory by a computer against a GM was at G5 time control. The GM >>>>famously said, "Bloody iron monster" when he realised that he'd made the >>>>tactical mistake that would cost him the game. >>>> >>>>If my memory serves me correctly, that was in the late 1960s. I don't think that >>>>any computers beat GMs at tournament time controls until the 1990s. They simply >>>>weren't good enough. >>>> >>>>-g >>>> >>> >>> >>>If someone wants to _seriously_ wager that if a computer plays a GM at >>>1 minute per game, 5 minutes per game, 10 minutes per game, 15 minutes >>>per game and 30 minutes per game, and the computer will do better and >>>better as the games become longer, I will set the experiment up on ICC >>>anytime they want. >>> >>>Of course, I don't think anyone _seriously_ believes that. >>> >>>I can't imagine anyone being so foolish. >>> >>>One good example was back around 1991 or 1992. IM Mike Valvo played deep >>>thought lots of blitz games at various ACM events and he simply got mopped up >>>badly, never winning a game and barely drawing one here and there. He played >>>it two 24-hour-per-move games in rec.games.chess, and he blew it out in both >>>games, with the machine never having any sort of a chance at all. >>> >>>That "gets better with more time theory" is correct in one regard. More time >>>means better analysis by the machine. >> >>This was what I was saying... >> >> But it is also wrong in a more important >>>regard. The human gets better at a faster rate of change than the computer, >>>as time controls are lengthened. >> >>The question is how much if any, and for what strength of players....did not >>some expert corr. player play some top computers in corr. chess....from the last >>I heard the computers were doing ok. >> >>The point I was making...search depth does correct some positional flaws and it >>is risky to assume the computers will play the same mistakes at fast and slow >>time controls...as some GM's have found this out the hard way. > >I totally agree. Practicing blitz games to learn weaknesses for standard games >is foolish. Just about as foolish as Kasparov using Fritz to prepare for Deep >Blue in 1997. I don't even consider myself a good chess player, but If I were Garry I would have laughed at such a suggestion. Knowing what I know about computer chess. This is like going to the go cart track to practice for the Indy 500. :) > >However, _if_ you play blitz the right way, and I know of at least one GM that >can do this, then this can work. But you play not to find tactical weaknesses, >but to find positional weaknesses, which won't go away with deeper searches. > >Some might disappear, but significant weaknesses will remain through deeper >searches. And these are the things that a good GM are going to locate, study, >and drive into the back of the program's head. > > > >> >> >> >> >>> >>>It has _always_ been so. It will be so for a _long_ time. More time reduces >>>the importance of tactics, which is what the computer lives and survives on. >>>As tactics dwindle, reliance on "knowledge" increases. And here, the computer >>>is woefully behind.
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.