Author: Robert Hyatt
Date: 19:48:13 05/01/01
Go up one level in this thread
On May 01, 2001 at 17:29:32, Dann Corbit wrote: >On May 01, 2001 at 17:15:39, Uri Blass wrote: >[snip] >>When a lot of programs play in a tournament opening preperation against specific >>opponents is more important. >> >>The number of games of every program in a tournament is also usually smaller >>than the number of games in a match so I do not think that I can get better idea >>about the question which program is better from a tournament. > >I think it answers a different question. > >The first set (against a large pool of different players) answers the question: >"Against this set of opponents, which program seems the strongest?" > >Against the second set (a large number of games against a single opponent) it >answers the question: >"Between these two foes, which one will fare the best head-to-head" > >Extrapolation beyond either of these questions is largely speculation. > >I think if you want the best program against humans, it would be best to play >against a pool of humans. And if you want the best program against Kramnik, you >should play a large number of games with each prospective engine against >Kraminik. > >Now, I don't think we can realistically realize this, but I think that is the >best way to secure the right outcome. Of course, that assumes that Kraminik >fully cooperates by playing his hardest against each engine, and that he never >has a sequence of mental lapses or indifference. > >In short, it is very hard to design the experiment properly. I don't think it is hard at all. The only problem is to frame the question you want to answer _first_. Then construct an experimental set-up that will answer that question. I think a two-program match is the _worst_ way to answer this. One program might know one thing about the game that the other doesn't, and that is enough to beat it regularly without being the best program of all available programs. A computer tournament is better. Because at least you get a mix of opponents and the program with the best overall understanding of the game will do the best, on average. A computer vs human tournament would be the _ideal_ way to see which program does the best against humans, and since Kramnik is a human, this is a good way to pick his best silicon opposition. BGN asked the wrong question, then constructed the wrong experimental setup to answer that wrong question. As a result, they got Fritz. Whether it is the best opponent for Kramnik or not, who knows? The experiment didn't answer that question. Is it better than Junior? Probably, although not by a big margin. Will it do better than Junior against Kramnik? That question wasn't posed nor answered.
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.