# Chess positions that are easy for humans but near impossible or confusing for computers

I noticed that one student is performing exceptionally well and beyond his level during an online class. I was wondering if there are particular chess positions that I can use during class to find out about such cheating. I would like to know of some positions that are easy and logical for humans but would make engines go crazy.

• Might it not be easier to do the opposite? Find a position that's really hard for a human but easy for an engine and see what happens? Jan 15, 2021 at 21:55
• Well, the problem is there are other students attending the class as well and I don't want to make this very conspicuous. So I was thinking maybe introduce some logical puzzle that is easy for humans but tricky for computers. Jan 19, 2021 at 18:50
• – BCLC
Jan 20, 2021 at 15:26
• Probably related chess.stackexchange.com/questions/8933/… Jan 21, 2021 at 15:35
• Maybe you can try a different trick: easy positions in which the second best move is easy for humans (and wining), but the best move is somehow obscure. For example, something like a simple endgame with a straightforward mate in 3, but with a mate in 2 possible with a weird underpromotion. I think you can find this kind of positions in middlegames when the king is about to be mated. Jan 26, 2021 at 10:04

One way to fool some computers is to give them a fortress and then offer material to trick them into opening the portcullis. For example:

``````[Title "White to draw"]
[FEN "8/8/pr1k4/Pp2rp1p/1Pp1pPpP/2PpP1P1/3P4/3K4 w - - 0 0"]

1. fxe5+?? (1. Ke1)  (1. Kc1)  1... Kxe5 2. axb6 Kd6 0-1
``````

Many computers won't resist the temptation to win RR for PP, but the resulting pawn ending is lost (Black can play Kb6, a5, a4, and then break through with f4). Instead White can just play random King moves, ignoring any Rook captures, and wait for 50 moves and there's nothing Black can do about it.

• I actually fed the position to Stockfish 12 and it did not fall into the trap. It proposed Kc1 (although with an evaluation of -10.15 at depth 54). Jan 16, 2021 at 3:33
• Interesting. How long did Stockfish 12 take to reject 1 fxe5+? What happens if you also give Black a Bishop on a dark square (on either side of the pawn chain, say +bBg2 or +bBb7)? Jan 16, 2021 at 3:53
• Stockfish suggests fxe5 up to depth 25 (just seconds of calculation) and then Kc1. Adding a bishop on g2 or b7, just improves evaluation for black (-15.49) but still does not trick the engine. But maybe there is a position some moves before where you could trick the engine with black into going into this closed position based on higher evaluation? Jan 20, 2021 at 9:39
• Note that there are many of such fortresses (the most well-known is probably the one where White must reject NxQ and instead gang up on a poor a-pawn to win) and the difficulty for engine is proportional to how far the winning line lies behind the horizon. BTW, got another idea: let them suggest a move in KRN/KNN. Looks perfectly natural, but anyone giving the shortest move to mate (+200 moves in a random setup) is a cheater! Jan 24, 2021 at 22:19

It's not exactly your question but could do the trick for your case. Set up mate in one problem involving en passant. Setting up a position with an en passant square is not completely trivial, so your allegedly cheating student could miss one of this type. Here is an example.

``````[FEN "4rk2/3prp1Q/1p4P1/3Pb2P/2P1pPq1/8/7K/5R1R b - - 0 1"]
``````

For some reason, chess engines also have a problem solving below puzzle, white to play and win

``````[FEN "8/n2P3k/3K3p/2p3n1/1b4N1/2p1p1P1/8/3B4 w - - 0 1"]
``````

Below is an image from chess.com

The puzzle may however be also difficult for humans, or at least those who have not seen it before

• This has been posted before in an answer that apparently since got deleted. It took many of the strongest chess GMs of the time many hours to solve this position. Meanwhile it took a current version of Leela just 10 minutes when I tried. It is difficult for engines, but even more so for humans since it's a very hard puzzle. Jan 21, 2021 at 16:29
• GM Mikhail Tal apparently solved this after a short walk to think source. Jan 28, 2021 at 2:56