Chess engine that would always resign

November 11, 2013 3:05 PM

I remember reading about how, in the early days of computer chess, someone wrote this engine that played pretty poorly, but knew when the game was no longer in his favor and would resign. Does anyone know what I'm talking about? Or would this have been a pretty common feature among many engines?
posted by Busoni to Computers & Internet (6 answers total)

Best answer: I think this is mentioned in Godel, Escher and Bach, and it's interesting because most chess engines never resign.
posted by pompomtom at 3:37 PM on November 11, 2013

Response by poster: OK, so THAT's where I read it.
posted by Busoni at 4:05 PM on November 11, 2013

Response by poster: I don't suppose anyone knows the name of this program or the history behind it though?
posted by Busoni at 4:05 PM on November 11, 2013

Usually, in a modern chess engine, you can set this behavior. The program can resign when the evaluation of the position crosses a threshold. This may be related to the delightfully named "contempt parameter" that chess programmers use.
posted by thelonius at 4:18 PM on November 11, 2013

To summarize GEB, it was a "recent tournament in Canada" (this in 1979), the program was the "weakest", had the "unusual feature of quitting long before the game was over", could "spot a hopeless position" and "resign then and there" (instead of playing the game to completion).

I couldn't find any other obvious references to this anecdote, even though most of the early chess tournaments were written up regularly in journals and popular magazines such as BYTE. I also couldn't find any accounts of computer chess tournaments taking place in Canada other than the 2nd World Computer Chess Championship, which was played in Toronto. The computer chess wiki [not Wikipedia] has a summary of the WCCC 1977 (ranking here) and none of the entrants seems to meet the description.

Possibly it could be a reference to Ostrich, "named Ostrich because of its cowardly 'head in the sand when in a crisis' style of play", but that program won, it appears, 2 out of 3 games and drew the third. Tell was the losingest engine, but is generally described as cruder and clumsier rather than elegant and prescient.

I'm a bit surprised as, while computer chess competitions started earlier than you might think, there really hadn't been all that many by 1979, and the significant ones were certainly well-covered within the appropriate fields. So I'd expect there to be something that pops out more obviously within this limited history. But at least you can browse these resources I've linked.
posted by dhartung at 1:33 AM on November 12, 2013

Usually, in a modern chess engine, you can set this behavior. The program can resign when the evaluation of the position crosses a threshold. This may be related to the delightfully named "contempt parameter" that chess programmers use.

The contempt parameter is a different concept; it represents how much the engine thinks it can outplay its opponent from a disadvantageous position. Say that your program has the option of either forcing an immediate draw (by a perpetual check, say) or going into a position with an evaluation of -0.5 ("half a pawn" down). Without a contempt factor it would take the draw rather than go into a position in which it thought it stood worse. With a contempt factor of over 0.5, it would go for the "dubious" line rather than taking the draw, assuming it could outplay you from that point on.
posted by dfan at 9:32 AM on November 13, 2013