2

Imagine the following position:

 [FEN "3k4/7p/7K/3B4/8/8/8/8 w - - 0 1"]

White is played by Stockfish in DroidFish and black by human (me). We both have 15 seconds.

After a few moves, white bishop or king gets the black pawn and... "Game over,draw by impossibility of mate!"... It is logical, due to the fact that there isn't mate position with two kings and one bishop.

I repeated the experiment sometimes and the same result happened... So, I wonder, why doesn't Stockfish flag me? Isn't it smarter to wait 50 moves for drawing with the hope to flag me without getting the pawn?

There are many possible mate positions without getting the pawn. I imagined one if pawn is promoted to knight and then: white king g3, white bishop h3, black king h1 and black knight f3. Black to move. Black knight to g1 and white bishop to g2 checkmate.

 [FEN "8/8/8/8/8/5nKB/8/7k b - - 0 1"]

 1...Ng1?? 2. Bg2#

So, it is not more logical in these cases, engines to try to flag the opponent? Maybe in a game between two engines it is meaningless, but in the case of engine vs human, I think it isn't so illogical.

What is your opinion?

  • 1
    I changed the title because, of course, Stockfish doesn't flag. – Allure Dec 16 '19 at 7:34
10

They are programmed to play the best chess, and there is simply no programming added in that attempts to account for the opponent being low on time. The way it played it was the best chess-wise. A human might try to flag you, but not the computer. Personally, I like it that way.

  • 1
    “But not the computer” – well, it certainly isn't impossible to do that for a computer. Only, it wouldn't be useful when it's playing against another computer, which is the setting which engines are optimised for. Winning against a human is easy anyway... – leftaroundabout Dec 16 '19 at 8:48
  • 1
    @leftaroundabout I never said they could not be programmed that way, just they they currently are not. – PhishMaster Dec 16 '19 at 9:11
  • 1
    It might also be worth mentioning that evaluating which moves will be most difficult for a human to respond to is a somewhat subjective thing, and not very easy to quantify in a way that would be useful to an AI. Especially because the human might make non-AI decisions, such as "I want to simplify this game", which means many lines are ruled out and the human won't have to spend much time calculating those, and how is an AI going to be able to take that into account? – Arthur Dec 16 '19 at 10:21
  • 1
    @Arthur I preach that to humans all the time as a matter of practical chess: "How easy is a position to play for a human?" I am not a programmer, so I do not worry about it in that sense. – PhishMaster Dec 16 '19 at 10:29
7

Generally, the computer plays the moves leading to the best objective result. Taking the pawn or doing what you said both lead to a draw objectively, so there's no big reason to prefer one over the other.

There are some adjustments that can be made to the engine, such as getting it to avoid repetitions early on (even if going for the early draw makes sense objectively). However, doing what you said would be hoping for a specific scenario occurring many moves down the road, and it could only really happen if you were trying to lose.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.