Author: Pete R.
Date: 16:42:26 05/24/00
Go up one level in this thread
I say "would" to mollify those that always caution about over-extrapolating results. I *want* to speculate. :) So if we *assume* that the returns from ever increasing search depth do diminish, what would that mean in practical terms for chess programmers? Knowledge knowledge knowledge? It's all in the eval tuning? To me, it's intuitive that if a program could evaluate any given position as accurately as a human GM, they would play better than humans all the time. I.e. if you can see with perfect accuracy what is the best position you can force the game to over say 10-12 ply, it would be game over for the humans. "Planning" is simply what humans do to accomplish this same feat, e.g. see that a particular weakness might be forced on the opponent within a few moves, and a program with such an eval function should exhibit apparent planning behavior. That assumes of course that all the long-term implications of a given position such as the pawn structure, space issues, etc are folded into the eval with correct overall weight. Much easier said than done of course. Perhaps the strength of DB owes mostly to the huge numbers of eval terms that were came "free" with the hardware. Just speculating.
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.