Sunday, February 17, 2013

Chess Engines' Evaluations

Nowadays powerful chess engines have become routine assistants of competitive chess players. Both amateurs and professionals use them to analyze their games, prepare opening lines, evaluate certain positions, etc. Most websites that broadcast games also offer a built-in engine evaluation to make the viewing process more interesting for chess fans. Such mighty servants as chess engines are of great use, but they also pose a serious danger.
First of all, it’s very easy to lose one’s own tactical skill if one starts following the computer lines without thinking for oneself. Secondly, quite a few players, even very strong ones, start “worshipping” engines and religiously trusting them. However, there are still blank spots in the evaluation mechanisms of the programs, so even at a large depth the first line of a program is not necessarily the best move.  Also, when playing humans we have to try to pose as much difficulties before the opponent as possible, place them under psychological pressure. Meanwhile, computers don’t know such things, and for them a king vs king position is evaluated the same as an insanely complicated draw that can be reached by making 20 one-and-only moves in a row.
Computer engines evaluate positions and offer an aggregate figure to show who is ahead. The number means how much better one side is in terms of material. Of course, in most positions material is not the only factor to consider, so the figure is derived by carefully weighing the tactical variations and positional factors.
Equality: =, from 0 to 0.26
Small advantage for White: +/=, over 0.27 and up to 0.7
Serious advantage for White: +/- over 0.7
Decisive advantage for White: +-, over 1.5
The signs for Black are similar (=, =/+, -/+, -+). A – sign is used to show that Black is ahead. E.g. a -0.8 evaluation means that Black has a serious advantage that is equal to about 0.8 of a pawn.
In some theoretically drawn positions the engines might still be saying that one side is ahead. Therefore, in endgames one should be especially careful when analyzing. The only exception is the endgame (Nalimov) tablebases (6-men are available online; 7-men are harder to find; creation of 32-men would mean that chess is solved). Using those one can instantly find the mathematical evaluation of the position: draw or a win for one of the sides.
Here is an example from the recently played Karjakin vs Topalov game, round 7 of Tata Steel Chess tournament:


The strong sides of chess engines are calculation and defense. They can also come up with unexpected and bizarre-looking ideas in certain positions. The weaknesses (if we can say so about players rated well over 3000) are positional evaluation and long-term planning. Quite often a chess engine would be saying that one side is better for a series of moves, and then all of a sudden treacherously change the evaluation for the opposite.
I use chess engines to check all my games and openings. However, the final decision belongs to me. This is especially true for positions with a few more or less equal options available. In such cases it’s very important to understand the idea/plan behind each of the moves. The difference might become obvious only a couple of moves later. In such situations chess engines are of no use, so you should either rely on your own brain, or check out leading players’ games to make a choice. Don’t be afraid to make a move that is not deemed to be the best by the engine.
Post-mortem: Robert Huebner and Natalia Pogonina. Photo by Martin Chrz
The instructive example of the above-mentioned principles will be my first game against former world #3 Robert Huebner from the Snowdrops vs Oldhands match (the second game was drawn as well). Most of the time the engine I have been using to analyze was claiming that the position is equal. However, by doing so it was neglecting some important features of the position. For example, after 20…dc the computer still says the evaluation is close to 0, but a qualified human would tell you that White is better. Black is obliged to defend passively, while White has some plans associated with pushing the kingside pawns. Therefore, in human terms it’s a “position for two results”: either White will win, or the game will be drawn. Maybe with strong play the position is indeed drawish, but Black is the only side at risk.

Therefore, if you prepare for the game only by memorizing chess engines’ moves and evaluations, at some point you will get in trouble. Always try to understand the ideas behind moves and make sure you understand why the position is evaluated as it is. For us, humans, intuition and experience are more important than brute force calculation. A chess engine is a great assistant, but it can never substitute for using one’s own brain in an over-the-board game

No comments:

Post a Comment