Author: Robert Hyatt
Date: 07:38:24 06/25/01
Go up one level in this thread
On June 25, 2001 at 05:30:04, Mark Young wrote: >On June 25, 2001 at 00:22:15, Robert Hyatt wrote: > >>On June 24, 2001 at 23:06:09, Slater Wold wrote: >> >>>I am holding a qualifing match between ALL the top programs. The time control >>>will be 25/10 and it will be a 3 cycle Round Robin. >>> >>>The purpose of this tournament is to qualify an engine to go against several >>>2500+ GM's in the next 5-6 months. These games will also be played at 25/10. >>> >>>Each game will be played on a Dual Pentium III 1,000Mhz ~ 184MB hash. Pondering >>>will be on, and the default book will be used, at tournament levels. >> >> >>One question: what is the point of playing computers against each other, to >>choose one to play against a human? Isn't this like playing 9 holes of golf >>to choose the challenger for the world champion in the shot put? > >Bob is there any conclusive evidence to show or suggest the strongest computer >playing other computers, is not the strongest or at least a good benchmark when >playing humans? I know this is debatable, but I have seen nothing conclusive to >suggest this is a pure waste of time. > >If you are correct, what was the point in playing Junior 7 Vs. Fritz 7 for the >match with GM Kramnik? Did you make the same objection? I said _exactly_ the same thing there, so yes I made the same objection. The qualifier was worthless. Here is a trivial example to make the point. Let's have a 2 0 tournament on ICC, computer against a GM I will choose. Let's have a qualifier to pick the computer challenger. Btw, the GM is going to play the trojan horse attack in every game. Which commercial program is going to avoid it at 2 0 time controls? Which commercial program will win the qualifier? Will those two programs be the "same program"? Not a chance. If the computers don't play the trojan, then the opponent will never lose to it. But if the human does, then bam. Ditto for kingside attacks and the like. Computers are not as multi-dimensional as humans. Some are very conservative. Some are overly aggressive. None are "ideally balanced". Pick an aggressive program against an anti-computer human and it will have trouble. Pick a passive program against an attacking human, ditto. The best way to choose the best program to play humans, is to play humans. Or barring that, take the current world champion program, as many of us suggested during the fritz/junior debacle. I suppose anyone can do whatever they want. IE they can try to grow pure strains of bacteria in a tropical rain forest. But it won't mean very much.
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.