Author: Bob Durrett
Date: 16:25:41 02/14/04
Go up one level in this thread
On February 14, 2004 at 19:00:53, horst meyer wrote: >comps performance can't be compared to human ratings since there r far too less >games human vs. comp. another reason is that the rating differences between >comps wouldn't hardly be affirmed in comp vs. human play, since it's not all >just about good and bad moves to express it in simple words. I do not disagree with the idea that precise estimation of the strength of a chess-playing machine requires many games. On the other hand, estimation need not be precise. One can make an estimation without any data at all. The problem is that no one would have any confidence in that estimate. The estimator would be said to be "just guessing." The estimate would be labeled as "pure speculation." Nevertheless, an estimate is an estimate no matter how little the justification for it. If the estimate were based on just a few games, then is is more reliable than an estimate based on pure speculation. An estimate based on a few dozen games might also be held in low esteem but it surely would be better than one based on a few games. The number of games increases our confidence in the estimate but we would be wrong to assume that the rating estimate were "right on the money." It could be off a little bit. If we have enough credible and relevant data to base our estimate on, then we may feel confident that the estimated rating "is quite close to the true strength. You can always compare FIDE elo to comps "elo" because there is some data. What we cannot do is conclude with absolute certainty that the comparison data is completely accurate. And . . . life goes on. : ) Bob D.
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.