Timeline for Why do chess engines cause immediate draws rather than giving the opponent the opportunity to blunder or run out of time?
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 30, 2021 at 7:12 | comment | added | bof | @DarrenH Which of these two minuscule chances is more minuscule: that the human player will walk into that helpmate, or that the computer will lose as a result of a hardware failure while waiting for the human to helpmate himself? | |
| Jun 29, 2021 at 19:15 | comment | added | Mobeus Zoom | @patbarron Let's put paid to the 'possibly walk away with nothing' stuff. There is an obvious way to avoid that (just keep an eye on the queening square) and no probability whatsoever of a white loss if the engine is careful (which the engine knows it will be), whereas there is a non-zero probability of a black loss (if black is particularly 'non-optimal', which the engine doesn't know he will not be!). Hence playing on till ...g8=Q is indeed better. That engine contempt which stretches this far doesn't currently exist is noteworthy. Why is an extremely interesting question (to me). | |
| Jun 29, 2021 at 17:58 | comment | added | Kevin | @DarrenH: More generally: Chess is a deterministic game. Engines don't "think" in terms of chances or probabilities, they think in terms of the evaluation function. In this particular example, the game is clearly drawn at the start, clearly continues to be drawn for most of the rest of the game, and the evaluation function should be approximately zero at all points except right after 14. ... Na2 (after which it should be the numerical equivalent of "mate in one"). Because Black has several other moves it could play which do not lose mate in one, the engine discounts this blunder entirely. | |
| Jun 29, 2021 at 9:11 | comment | added | Hauke Reddmann | @patbarron: The problem obviously being, that computers have only an evaluation function, and it is a nontrivial job to include a fallible opponent into it. How many pawn units shall you "sacrifice" for a possible reward? | |
| Jun 28, 2021 at 18:56 | comment | added | patbarron | @DarrenH - the engine will prefer to take the guaranteed half-point, rather than play a "hope chess" line and possibly walk away with nothing. All this being said, there are certainly chess programs/engines that will deliberately play something other than the best available move (even to the extent of purposely making outright mistakes), just to kind of mix things up or simulate a less skilled player. But an engine that is trying to work towards the best available result is going to just take the half-point and call it a day. | |
| Jun 28, 2021 at 13:25 | comment | added | wizzwizz4 | @DarrenH Engines like Stockfish don't “comprehend” of that as a possibility. | |
| Jun 28, 2021 at 11:31 | comment | added | Gareth McCaughan | Isn't the point about the possibility of a helpmate not that there's any realistic chance that black will help white deliver mate but rather that because mate is still possible in principle, if black runs out of time then the result will be a W win rather than a draw? | |
| Jun 28, 2021 at 9:35 | comment | added | Darren H | @eyeballfrog it's a miniscule chance, but non-zero. Surely an engine would see almost zero as preferable to zero | |
| Jun 28, 2021 at 1:53 | comment | added | eyeballfrog | @JosephSible-ReinstateMonica You'd need a lot of contempt for your opponent to think they might helpmate with an underpromotion. | |
| Jun 28, 2021 at 0:45 | comment | added | Joseph Sible-Reinstate Monica | Isn't the concept of contempt in chess engines exactly that they don't assume their opponent is perfect? | |
| Jun 27, 2021 at 21:27 | history | answered | patbarron | CC BY-SA 4.0 |