Author: Dave Gomboc
Date: 20:19:20 09/07/01
Go up one level in this thread
On September 07, 2001 at 13:34:53, Robert Hyatt wrote: >On September 07, 2001 at 00:39:39, Dave Gomboc wrote: > >>On September 06, 2001 at 13:13:22, Robert Hyatt wrote: >> >>>Here is my take: >>> >>>If DF wins convincingly (IE by a more than one-game margin) then I will have >>>to re-think my GM evaluation. If DF wins by .5 or 1.0 points, it is possible >>>that Kramnik was too conservative, then lost one game by accident, and couldn't >>>catch up. If the match is tied, it won't say much either. If Kramnik wins >>>by some narrow margin, it could mean one of two things. He was lucky. Or he >>>was ultra-conservative. >> >>Don't you think it a bit odd that it would take a convincing win over the world >>champion to convince you that the program is of GM strength? :-) > >No... for the reasons I gave. IE suppose the first 7 games are simple >draws, dictated by the GM's play. And suppose he tries to win the last >game and makes a simple tactical error. I watched a GM game on ICC yesterday >between Polgar and Short (short was white). Polgar defended nicely against >a kingside attack and ended up a pawn up. But in a very hard to win position. >Suddenly, Short played Rg2 and Crafty's score _instantly_ went to -5 (black >was winning). And I mean _instantly_. And the deeper it searched, the better >this got for black. When I looked at the board for about 10 seconds, I could >see the problem too (white was losing a pinned piece due to a queen check >that made the king move and simultaneuosly attacked the pinned piece once more >than it was defended.). > >Is short much worse than Polgar? hardly. That is why a win due to a simple >error won't be very convincinging. If Kramnik comes out swinging, and plays >to win in every game, then a single point victory by fritz will mean a _lot_. > >That is why I said "it all depends on the final score and the chess that was >played in each individual game." A conservative approach would be to play >cautiously in each game, waiting for a positional mistake that you believe >you can exploit to win. If you do this and draw the first 7 games by choice, >then the last game isn't going to be a good indication of who is better. If >you try to win every game by playing either anti-computer or traditional chess, >then the match result will be more revealing. Fair enough. >>My personal opinion is that when this "are computers GM strength?" debate began, >>you were correct to say that no, but that times have since changed. I don't for >>a moment doubt that the leading computer programs can be made to look pathetic >>from time to time, but the flip side is that they can and do play some pretty >>damn good chess games as well. >> >>Dave > >They can play good chess when the humans let them do so. But block the position >and every program I have seen looks like a complete moron. The question is, >will the human choose to do this, or will he choose to play normal chess where >a program can often look just like a super-gm given the right positions. > >If I play you in tennis, and I know you have a dynamite forehand, you are >_never_ going to return a ball from your right side of the court, because I >am never going to hit it there, except when I drive you wide to the other side >and then want to make you run a bit. IE I will _never_ play to your strength, >as it simply makes no sense, other than to occasionally keep you honest. It seems to me that there is a large percentage of GMs who are either unable or unwilling to take this approach when they play programs. Joel Benjamin is excellent at pounding the things (that's why the DB team got him onboard) but most GMs don't have the same ability he does at knowing how the machine thinks and using that against it. We saw Fritz get smoked badly in a late game in the Dutch Championship, but we also saw it smash a GM who was trying to keep things closed just a few rounds earlier. Even when GMs are trying to play anti-computer, they don't always succeed. Add that to those who don't bother or are unable to do so, and you've got a significant number of games where the programs are quite capable of winning. Dave
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.