Don't Worship Your Chess Engine!

Chess computer software is extremely popular, and has been for a long time. A chess engine can analyze your games and give you an idea of how well or poorly you played. With ratings topping the 3400 mark, these monsters are several hundred points stronger than any human chess player, dead or alive.

…that is abused by less experienced players

Beware blindly following the output of a chess computer! An engine does not “understand” chess the same way a human does, and we cannot achieve the near-perfection in play that a computer can. When analyzing tactics the computer sees nearly everything, but what if you want to understand a position where pieces aren’t flying everywhere?

Here’s an example. You enter a game into ChessBase (or open one from a database). Then you open your favorite chess engine to analyze it, such as Fritz or Stockfish. How helpful might this be?

Let’s take the classic game Evans—Opsahl from the 1950 Dubrovnik Olympiad.

Evans-Opsahl, 1950, after 17 moves. I have opened Stockfish 11. How helpful is the engine, really?

The screenshot is not easy to see, so I’ll fill you in on some details. I turned on Stockfish 11 after black’s 17th move and let it analyze for awhile.

At 36 ply (half moves) or 18 full moves, it considers white’s best move to be 18.Rb2 for some strange reason, giving an evaluation of +/= 0.68. This suggests white is slightly better. In a real game between humans, I totally disagree!

Situations like this usually arise from the Exchange Variation of the Queen’s Gambit Declined (1.d4 d5 2.c4 e6 3.Nc3 Nf6 4.cxd5 exd5; other sequences of moves can reach this same position). After both players support their d-pawns, and castle, we get a Karlsbad pawn structure, like this:

Both sides hope to start a minority attack on the flank where they have less pawns! White plays on the queenside, and black on the kingside. The idea is to create weaknesses to attack later. White’s play is quicker and easier, but if black succeeds the reward is a dangerous assault on white’s king.

What the engine can’t tell you

Let’s take another look at the game position:

White is ready to play 18.b5! to break apart black’s queenside. Notice that white’s pieces are in position to pounce. Of course, the future five-time U.S. Champion did just that.

Black has not played in the most accurate manner, and his attack is nowhere near threatening enough to disturb white seriously. Things would look better for him if his knight was on a more threatening post.

But wait, there’s more…

I’d like to mention that if black had the move here, an interesting possibility would be to play 18…b5!? himself. That makes is much harder for white to break through, and black has only one weakness to defend, on c6, though it’s a very serious one. Slowing white’s queenside play would also give black time to organize counterplay against white’s kingside. Either that, or try to land the black knight on c4 where it shields the weak c6-pawn from white’s pieces:

After the computer suggestion 18.Rb2 the move 18…b5 gains even more punch, because if white now follows with 19.axb5 axb5:

You can’t go wrong with this classic by Ludek Pachman.

White’s heavy pieces trip over each other and struggle to fight for the newly-opened a-file!

A chess engine can’t explain all of that to you. You have to either read a middle game textbook (such as Pachman’s Modern Chess Strategy), study well-annotated games from a database, or hire a coach. You can also read this blog regularly!

I could have left the engine on even longer, and maybe it would have chosen 18.b5 after all. It was the second-choice move with an evaluation of +/= 0.63. The point is, the computer couldn’t tell an inexperienced player the ideas behind any moves it suggests!

Evans was more than “slightly better” after 18.b5

The second player had to passively defend a weak structure for the rest of the game. In a practical game, this is a nightmare scenario. Opsahl finally succumbed after 81 moves.