Author: Sune Fischer
Date: 05:37:42 03/01/02
Go up one level in this thread
On February 28, 2002 at 16:55:14, Amir Ban wrote: >On February 22, 2002 at 14:21:01, Paul Doire wrote: > >>When chess is "figured out" what ELO will that equate to? >> >>Paul > >There was a recent post that said 3300. This seems reasonable. > >I think 4000 which was guessed on this thread is a wild exaggeration. > >If perfect play is 3300, Kasparov would score about 7%, and the average GM 2-4%. >If perfect play is 4000, then these percentages should be divided by 40, which >stretches belief. I think 3300 is too low, the super GMs are already having great problems with the top mini progs, just look at Rebels last match. Since there is not yet any major signs of diminishing returns, and since we're only at ply ~12 there is still a lot of room for improvement. I think a 20 ply searcher would "tear Kasparov to pieces". Notice how even the top players say "this position is unclear" or "black has good chances". It seems the positional understand of man can compensate for about ~14 ply of tactics, but not far beyond that I think. If going from 1 ply to 12 plies is 2700-1000 = 1700 elo, then 12-24 plies should give at least another 1000 elo, a rating of 3700, and still a long way from perfect play. Pure speculation of cause. >The game outcome is not relevant for this discussion. If the game is won, the >perfect player will always win with the right color, but not with the wrong >color, and conceivably he may lose. If the game is drawn, the perfect player >needs a mistake to win any game. It's not the theoretical outcome but the level >of opposition that is important. I agree, therefore we might as well assume it is drawn, it makes it easier to talk about, and it probably is IMHO. >It's possible to get an approximation by having a strong player play against an >EGTB in a level position and looking how well he scores. That has also been suggested, I think it would be a good test. Perhaps we do not even need top players to do the test. We take one 1400, one 1600 etc. up to 2200 or as high as we can get. That would give us some good data need for extrapolation (ie. how does the frequency of wrong moves relate to elo). -S. >Amir
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.