Author: Dann Corbit
Date: 15:18:44 03/19/99
Go up one level in this thread
On March 19, 1999 at 17:24:21, KarinsDad wrote: [snip] >What can you conclude from a Yermo/Hiarcs7 3.5-2.5 score? >I would think that this would show that Hiarcs7 is strong, but would not be >enough evidence to conclude that it was at GM strength (since Yermo may just be >playing for the minimal win). > >What can you conclude from a Yermo/Hiarcs7 6-0 score? >Hiarcs7 is not GM strength (yet). > >What can you conclude from a Yermo/Hiarcs7 3-3 or better score? >Hiarcs7 is GM strength since it drew or won in standard time against a very >strong GM. > >It would appear that you can only infer that Hiarcs7 is GM strength if it wins >or draws. And you can only infer that it is not GM strength if it loses badly. >Anything in between is only evidence of a strong program, nothing more. I disagree with all of these, only because you used the word "conclude." Two 2700 GM's could meet and the score is 6-0. What shall we conclude? Not a whole lot. It is no different with programs. Suppose Yermolinsky discovers a tiny crack and exploits it for 6 games? At the end of 6 games, the computer has learned the information and it could not be used again, perhaps. Suppose Yermolinsky is troubled by something or does not get a good nights sleep? Life is full of random variables, and machines are programs are just as affected by them as we are [albeit in different ways]. With 6 games, we will have enough to make a very, very crude provisional rating. If a player who had never played in a tournament before went and won 6 games against top players will FIDE give him a GM title? No, and the reason is that it is still absurdly unproven. But -- it would show _evidence_ that that player _is or may be_ of GM caliber (or owns a radio transmitter & Fritz on a PII 400). ;-) Jut not enough to conclude anything yet.
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.