Monday, August 20, 2007

Value of Theory?

Friends:

I'm curious what folks think about the new article on USChess.org by Joel Benjamin. The topic is computers and the relative value of theory. (see below)

My position is that theory is overrated, and that many strong American players give too much weight to this aspect of the game. Further, even with best play theory gives weaker players (such as myself) an advantage that practically speaking has little meaning.

I'm more of a believer in Rowson's idea that for adult players to improve you have to increase skill.

Thoughts?

Carey Theil

http://beta.uschess.org/frontend/askGM_24_67.php

Ask GM Joel

Question: Dear Joel, Your June 4 column addressed a question about humans vs. computers in Fischer Random chess, concluding that, "So Fischer Random chess actually eliminates human advantages and accentuates computer strengths!" What about computers playing normal chess but without an opening book? It seems to me that top-level humans would have a significant advantage since opening theory imparts the "wisdom" of massive experience to the human.

By Jim Eoff 2007-08-18

GM Joel responds: At one time, computers couldn’t play the opening at all without an opening book. Their evaluation functions were poorly developed and calculating powers not strong enough to avoid typical traps. Now that computers are acknowledged to be stronger than even the best grandmasters, I’m not sure computers need an opening book.

Human opening knowledge uses assumptions based on past experience. We know that theory is fallible—how else could players constantly find “improvements” on established theory? Computers may not always find better moves, but they may choose perfectly acceptable ones that have been overlooked or discarded for no good reason.

Even when a computer plays an opening move which is apparently inferior, it can still put the human out of his comfort zone. Since people know openings so well, every move that is still in the grandmaster’s opening “book” is an opportunity for him to use that knowledge. By leaving theoretical territory, the computer, as the stronger player, should be better suited to handle the situation.

In its match with Kramnik, Fritz played an unusual maneuver with Re1-e3. Despite its weird appearance, the move still contained a drop of poison and Kramnik was fairly befuddled about how to deal with it.

The need for an opening book was put to the test in a recent match between GM Jaan Ehlvest and Rybka, generally acknowledged to be the world’s strongest computer. Rybka was allowed only a few moves of book before it had to play completely on its own. Ehlvest was given White in each game, and more time on the clock as well.

Rybka scored three wins and three draws. The results are open to interpretation. Rybka won handily, but perhaps it would have won by a greater margin with book openings? Ehlvest was also pressing in a number of games and could well have scored some wins.

In August I played a match with Rybka (ending 4.5-3.5 in Rybka's favor) where I received a pawn handicap in all eight games, as Ehlvest had in a match back in March. With the pawn missing, the games in its opening database simply don't register. Some of its opening moves seemed a bit strange, though perhaps not objectively weaker under the circumstances.

I would say that the absence of an opening book imparts a slight handicap to a computer, but not enough in itself to enable grandmasters to beat the top programs.

No comments: