Author: Bob Durrett
Date: 17:55:49 01/01/04
Suppose someone wanted to produce a chess-playing program which emulated the play of a human rated 1200 to 2200. Rhetorical Question: "How would one go about doing that?" The first step must surely be to produce a high-fidelity model of the thing that is to be emulated, in this case the human amateur chessplayer, or at least to model the output of that thing. Perhaps it would be sufficient to produce a "black box" which outputs moves which are ***statistically*** similar to those observed from the human amateurs. In other words, a "black box" chess-playing program which outputs moves having the same statistical properties as the outputs of the thing being modelled [the amateur chessplayer] might be sufficient. This could get very involved, but consider one simple example: Perhaps the two most dominant features of the moves of "the thing to be modelled" is the high number of tactical blunders per game and the high number of missed tactical opportunities per game. These two things could be evaluated statistically with measured performance rating as a parameter. To correctly emulate these two items, it would be necessary that the "character" of these be the same. In other words, Turing's Principle needs to come into play. The "tactical blunders" and "missed tactical opportunities" need to be essentially indistinguishable from those produced by "the thing to be modelled." The same would have to apply to all statistically significant properties of the output [i.e. the moves] of "the thing to be modelled" [i.e. the human amateur chessplayer]. Does this sound right? Has any chess software tried to do this? 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.