Question about dealing with endgame databases

[Eduard]

Again and again I see in engine games that my engine prefers to play with fewer stones. Example: In the endgame Kp vs Kp, the last pawn of one's own is given away rather than the opponent's. The engine would rather play with a naked king than with pawns. That's a joke, but unfortunately that's the way it is. Is that just coincidentally wanted by the engine (a draw is a draw) or does the engine want to delay the game?

[syzygy]

Again and again I see in engine games that my engine prefers to play with fewer stones. Example: In the endgame Kp vs Kp, the last pawn of one's own is given away rather than the opponent's. The engine would rather play with a naked king than with pawns. That's a joke, but unfortunately that's the way it is. Is that just coincidentally wanted by the engine (a draw is a draw) or does the engine want to delay the game?

With many drawn KPvKP positions this can happen even without tablebases, since the engine will search so deeply that all its lines end up in KvK anyway.

With less trivially drawn positions, it depends on how the engine was programmed to deal with drawn tablebase endings. For example, Stockfish is unlikely to needlessly throw away material (it will basically play the drawn TB position as if it did not have access to TBs, except that it will make sure not to play a losing move). For other engines this can be different.

[Eduard]

Here is a position from practice, just played live on Lichess:

[d]8/8/6k1/7p/5K1P/1b6/4B3/8 b - - 0 1

My engine is black. It's the endgame and White only has a few seconds left on his clock. Nevertheless, the black engine plays Kf6 here and gives away a pawn. Later the bishops are swapped. In the end, white has one pawn more and can even afford to exceed the time without losing. No human plays like that!

Here is an analysis on my smart phone with Cfish:

[53] 0.00 Kf7 Kg5 Kg7 Ba6 Bd5 Kxh5
d: 54/6 1: Kf7 t: 10.81 n: 63320k nps: 5858k tb: 130k h: 1

Here, too, Black gives away his pawns. I don't like this way of playing. The programmers should come up with something against it.

[Ajedrecista]

Hello Eduard:

It could be worse for your taste in this position. Since all moves lead to a draw, 1.- ..., Bc4 or 1.- ..., Bd1 can be played, not losing the pawn but the bishop firstly, but taking advantage of the wrong bishop endgame.

Regards from Spain.

Ajedrecista.