Author: P. Massie
Date: 09:23:29 10/08/03
Go up one level in this thread
I think there's a fairly simple explanation for the results we're seeing. If you take a case where Player A is clearly stronger than Player B and they play a number of matches, Player A will win most or all of the matches. If we then postulate that Player B is getting stronger faster than Player A and they continue playing matches, there'll come a point where most matches end in draws. Player A is still stronger than Player B, but not enough so to demonstrate that strength with small numbers of games. At this point it takes a large number of games (which we don't have in the computer vs. human case) to demonstrate the strength difference. If Player B continues to improve faster than Player A, eventually there will come a point where Player B is stronger than Player A, but most matches will still end in draws. Again, we'll need a large number of games to demonstrate Player B's strength advantage. After more time passes with Player B continuing to improve faster than Player A, there comes a point where Player B starts to win most matches, even if not very many games have been played. In essence, there's a band with relatively evenly matched opponents where large numbers of games are required to demonstrate which is really stronger. As long as we don't have those large numbers of games the actual strength changes will be invisible to us until one of the players moves outside that band. I think we moved into this band with computers vs. top humans a few years ago and we're still within the band today, although (as James said) the actual strength relationship has shifted enough so that top players now need to work hard to maintain an even position with the top programs. If the strength progressions continue there'll come a point where the top players start losing most matches even if they try hard. The exact timing of that point is clearly a matter for debate, although it appears to be within the next very few years. Paul
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.