Basic Reasons to Distrust the Engine
This post is about some easily demonstrated reasons the engine isn't necessarily showing you the best moves, and even if it is why following the engine's advice isn't always useful. A forum post of mine ended up being so long that I decided to turn this into a blog.
First I'll make the engine calculate the same line twice. Here's a position I came up with for that:
-
-
White's best move is to capture the bishop, and it doesn't matter what piece he promotes to, black will recapture... but since the pawn can promote to different pieces the engine will treat that as different lines and so it will calculate the same line twice. I'll set my engine to show the top 2 moves. This way we'll see that even with:
Identical hardware
Identical software
Identical thinking time
Identical depth
Due to the inner workings of an engine, the evaluation can fluctuate about 0.2
-
-
You'll also notice depending on the depth, you'll get different answers and different evaluations.
This leads to the first two tips:
1) In the engine's opinion, moves that are 0.2 apart are essentially equal.
2) The engine can change its mind (both the move and the evaluation) depending on how long you let it think.
It can also change its mind depending on whether you play out some moves on the board or just do all the calculation on the initial position, but I'm not going to set up a demonstration of that.
In chess, there are only 3 real evaluations a move can have: winning, drawing, or losing. Next I'll give justification for the following two tips:
3) An engine may show you a winning move, but that doesn't mean it will show you the most logical winning move. In fact
4) For human play, practical considerations can override the engine's analysis.
To demonstrate this, look at this endgame tablebase (with link below the picture). We'll compare it to what an engine thinks, and then use a slightly different position to second guess the engine.
-
http://www.k4it.de/?topic=egtb&lang=en
-
White has many moves that win, and the top 4 moves all win equally as fast.
Here's what the engine thinks:
-
-
-
First of all you'll notice that the engine's top 4 isn't the real top 4. The EGTB showed that the top 4 are: Qb2, Qf3, Qb6, and, Qa7. The engine's top 4 (at this depth) only includes one of those (Qb6).
But more importantly, even though the engine knows this is winning, in a practical game most non-masters will not be able to win this as white. Not only is breaking down the false fortress very hard (the defender only has the shuffle the king and rook), but the ensuing pawnless Q vs R endgame is also much more difficult for the stronger side to win than it is the weaker side to draw.
So knowing the practical human side of things, let's look at this position.
-
-
A human could sacrifice the knight for the last 2 pawns and will almost certainly draw with ease at the sub-master level. The engine recommends 1...d5, but that doesn't necessarily increase your drawing chances.
-
In summary, the engine is not an oracle. Its top moves can be essentially equal, and it changes its mind. Additionally, chess is a game where most positions have more than one winning or drawing move. To play well as a human you should choose the more logical paths to victory, and take practical elements into consideration even when that means disagreeing with the engine.
