Author: Michael Neish
Date: 21:08:13 01/20/00
Go up one level in this thread
On January 20, 2000 at 23:19:41, Dann Corbit wrote: >Hyatt & Heinz both did "goes deep" experiments that show no matter how deep you >look, you never stop finding improvements. And each new ply is of approximately >the same value as the one before it (but much harder to calculate). > >GM's, also, can be wrong in their analysis. So what is the best move from some >board position? Nobody knows. On the other hand, you *can* rigorously say that >for an examination of all the positions up to n plies forward, some move is >better than others (assuming you can come to agreement on how to calculate >piece, positional, and other value factors consistently). Well, there's only a limited number of moves in a position, and many of these are probably obviously bad. I suppose what you mean by finding improvements is that you can keep switching between possible candidate moves as you search deeper, right? Maybe the computer switches back and forth between axb5 and Qb6 as it thinks. I remember reading in some paper that as you get very deep errors start to creep in the evaluation, so that after a certain depth, if there is no definite tactical objective, you cannot trust the evaluation at all. I have the reference to the paper at home, which I can give you if you're interested. The author said that Chess programmers tend to ignore this phenomenon, which is negligible at shallow search depths but becomes more and more serious with increasing depth, but that it's there all the same. But we're digressing. By the way, how about the move Qb6, which (I think) is what Kasparov thought every honest computer should play. Is it some sort of trap that Kasparov had laid, which would eventually lead to a draw, or does it also win for White? If Qb6 is good enough to win, then why would the IBM team override a winning move with another more elegant winning move selected by a grandmaster? Cheers, Mike.
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.