Wednesday, December 1, 2010

GM Evaluations vs. Engine Evaluations

Chess engines do not understand chess at all; they simply calculate variations and evaluate positions using an evaluation function which consists of material count plus factors like pawn structure, king safety, etc. and then it returns an evaluation. In some cases, the computer's evaluation function will favor a strategically lost position because it is making a static analysis but there may be overriding dynamic factors in the position that actually favor the other side. In some positions GM’s and other strong players will recognize these exceptions. That is why in strong CC tournaments players who rely solely on computer generated moves often lose to players who are strong enough to recognize when an engine’s evaluation is wrong.

I like to play over GM games using an engine because it allows me to quickly play through variations and try out my own lines. On more than one occasion I’ve run into situations where the GM’s evaluation is completely different than the engine evaluation; I always trust the GM.

Of course GM’s have been known to publish errors but these are not apt to be serious and these days most check their analysis with engines. When playing over older games you will often discover tactical errors in analysis, but when it comes to having a position explained or a strategic evaluation, always trust the GM. Also, in quiet positions there is often more than one move that is “best” so it’s pointless to pay any attention to engine analysis.

Recently while looking through The Art of Bisguier Vol. 2, I came upon the following position where Bisguier disagreed with Fritz so I decided to check it out. I think Bisguier was using Fritz 5, so I used a stronger engine, Houdini. What was interesting was Bisguier’s explanation of why he disagreed. Let’s take a look at the game Bisguier – Kane, US Championship, 1973.

Bisguier comments that Kane must have been very happy with his position. He has two pieces developed (to one), two P’s influencing the center (to one) and an open g-file on which to attack the K should White castle K-side. According to Bisguier, Fritz gave Black the advantage worth the better part of Pawn, but he disagreed. According to my Houdini engine, Black stands better by almost a half a P after White plays 9.d3, so what is it in the position that makes Bisguier think it is actually White that stands better? We will see after the moves: 8.d3 Qd6 9.Nf3 Nd7 10.0–0 0–0–0 11.d4 Rhg8 12.Be3 Qg6 13.Qe2



13…cxd4 (Bisguier comments that this is not a mistake by itself, but it is the start of a bad plan that exposes Black’s Q-side weakness on a7. He recommends instead moves like 13…h5, 13…Qg4, …Qf5 or 13…e6. Indeed after each of these moves Houdini continues to rate the position nearly equal.) 14.Bxd4 Kb8 15.Rf2 e6 16.a4 Nb6 (Houdini does not like this move after which its evaluation swings in White’s favor by a half P. Bisguier recommends 16…Bxd4 instead, with an equal position.) 17.Ne5 Bxe5 18.Qxe5+ Kc8 19.Nb5



In this position Houdini gives White a one P advantage. The remaining moves were: 19… Bxb5 20.axb5 Nc4 21.Qe1 and Black resigned. Final position:
In this position Houdini rates the position as one P in White’s favor, so why did Kane resign? Bisguier writes, “At first glance Black’s position looks as though it should still have plenty of play in it, but once again appearance are deceiving. A closer study reveals that Black’s assault on my K has come to a standstill, while his Q-side K position is altogether indefensible.”

And there you have it. In the beginning diagram it appears to us amateurs and the engine that the position is equal, but Bisguier saw beyond the static features and realized that dynamically his position held better possibilities. Even in the final position most of us would have probably evaluated the position either as slightly favorable to White or equal, but certainly not resignable for Black. Of course in a game with players of lesser caliber than Bisguier and Kane it would not make sense for Black to resign, but clearly both players here realized that while Black could play on, his position was hopeless.

No comments:

Post a Comment