# Can you improve a supercomputer evaluation on your own move?

While watching the Chess World Championship, a few news outlets provide updates on how supercomputers are evaluating the board position. The example I'm looking at is using the Sesse Supercomputer, which evaluates the strength of one side over the other measured in pawns. For example, though I don't remember the exact numbers, at one point in game 8, Caruana had an advantage of roughly +2, however after the ill-advised move h3, the computer reevaluated their position to be roughly equal.

That made me wonder: Can you improve your evaluated position on your own turn, or do you only stand to weaken it?

It seems to me that if you had a move available which strengthened your position, that would be included in the evaluation. My assumption (of which I'm not certain), is that the computer expects you to take the best possible move, so doing so would simply maintain the advantage you already have. Taking any move aside from the strongest would weaken it.

Is this the case?

• one side weakening is another side strengthening. So you improve upon what your opponent allows you to improve – Isac Nov 21 '18 at 20:56
• – itub Nov 21 '18 at 22:49

Yes, all the moves would have been included in the evaluation, but that evaluation isn't perfect. Supercomputer is still a normal computer, just faster.

In your example, the chess god will see that +2 position as "mate in 50", hence it should be evaluated as +infinity, a big improvement.

Whether mortals like those world champions can improve that evaluation is a different matter.

• "the chess god will see that +2 position as mate in 50" It might also see the +2 position as "+0.0" ... I suspect that the draw margin with perfect play is quite large – Zubin Mukerjee Nov 21 '18 at 20:42
• @ZubinMukerjee His point stands though. A computer which can see far enough ahead wont use anything except +infinity, -infinity, or 0. Chess God already knows how the game should end; to Chess God there isn't even a game. But to an imperfect chess god, there is, and that leaves room to make a move better than the engine's best move – Lord Farquaad Nov 21 '18 at 21:08
• @LordFarquaad, yes. Although it's also possible that as you follow engine's best suggestion, you will see the evaluation rising, eventually showing mate in X. – jf328 Nov 22 '18 at 9:31

The point of an evaluation is to rate one sides chances of winning and to select the best chance of winning for that side. All the engine can do is calculate up to a certain point and then rate the position. So yes you can definitely improve your evaluation because the engine may have missed something past its search depth.

If an engine had infinite depth there would be no point of an evaluation.

No way your assumptions are right. If the move is winning, the score may ruse even further. However, if the move is not winning playing it the score might drop to zero.

Stockfish can make mistakes and often do even on a good hardware.

Evaluation can change to either direction every time you make a move. Sorry chess is a hard game.