4 ms·

Related question: Has anyone ever experimented with taking a really strong engine for one of these games and then simulating the sort of mistakes that humans ma

by ilyanep 11y ago

Related question: Has anyone ever experimented with taking a really strong engine for one of these games and then simulating the sort of mistakes that humans make (overlooking certain positions, looking ahead fewer steps, etc) to try and simulate different skill levels of opponents? I don't play Go, but I dabbled a little bit in Chess, and it seems like the engines with "difficulty settings" don't really make the sort of mistakes that humans make. They can, of course, just over-prune the look-ahead tree, but that's not how the human brain works.

Laaw 11y ago
Yeah, IIRC most engines just blunder every once in a while depending on the difficulty setting. rather than playing consistently less strong moves, or something like that.
noir_lord 11y ago
TheKing (the engine behind Chessmaster) had the most human seeming play to me and I played most of the engines over the years. Most of them are more "play like God then hang their queen". The engines now are so strong I can't beat them with Queen and Rook odds so even that doesn't really help.
bazzargh 11y ago
I recall in ~1990 I read a paper which described exactly this (attempt to make the same mistakes as humans). The argument for doing so in the paper was, IIRC that computers beating humans at chess was just a matter of time, but no longer interesting for AI research. It seems quite likely that the book that paper was in was Computers, Chess, and Cognition https://www.springer.com/us/book/9781461390824 https://www.springer.com/us/book/9781461390824 (revised contributions from the WCCC 1989 Workshop New Directions in Game-Tree Search, May 29-30, 1989)...and was probably the one by John McCarthy (yes, that McCarthy) Edited to add: Found it! The paper I recall reading was 'Artifical Stupidity' by William Hartson, in Advances in Computer Chess 4 (1986) https://chessprogramming.wikispaces.com/Advances+in+Computer+Chess+4 https://chessprogramming.wikispaces.com/Advances+in+Computer... ... perhaps nothing concrete came of this though - Hartson was a player and author, not an AI researcher.
zem 11y ago
it's on my nebulous "someday" todo list to add this sort of thing to quackle [one of the two current leading-edge scrabble programs, elise being the other one]. ideas include simming opponent play with a reduced vocabulary, attempting plausible phonies, and adapting the search tree to account for how well opponent's previous moves matched the simulated best move for them. on the other side of the coin, that would also allow quackle to have realistic skill levels (right now the only skill adjustment is essentially "stop thinking after a certain time and just play the best move found till then")

by ilyanep 11y ago

Laaw 11y ago

Yeah, IIRC most engines just blunder every once in a while depending on the difficulty setting. rather than playing consistently less strong moves, or something like that.

noir_lord 11y ago

TheKing (the engine behind Chessmaster) had the most human seeming play to me and I played most of the engines over the years. Most of them are more "play like God then hang their queen". The engines now are so strong I can't beat them with Queen and Rook odds so even that doesn't really help.

bazzargh 11y ago

I recall in ~1990 I read a paper which described exactly this (attempt to make the same mistakes as humans). The argument for doing so in the paper was, IIRC that computers beating humans at chess was just a matter of time, but no longer interesting for AI research. It seems quite likely that the book that paper was in was Computers, Chess, and Cognition https://www.springer.com/us/book/9781461390824 https://www.springer.com/us/book/9781461390824 (revised contributions from the WCCC 1989 Workshop New Directions in Game-Tree Search, May 29-30, 1989)...and was probably the one by John McCarthy (yes, that McCarthy) Edited to add: Found it! The paper I recall reading was 'Artifical Stupidity' by William Hartson, in Advances in Computer Chess 4 (1986) https://chessprogramming.wikispaces.com/Advances+in+Computer+Chess+4 https://chessprogramming.wikispaces.com/Advances+in+Computer... ... perhaps nothing concrete came of this though - Hartson was a player and author, not an AI researcher.

zem 11y ago

it's on my nebulous "someday" todo list to add this sort of thing to quackle [one of the two current leading-edge scrabble programs, elise being the other one]. ideas include simming opponent play with a reduced vocabulary, attempting plausible phonies, and adapting the search tree to account for how well opponent's previous moves matched the simulated best move for them. on the other side of the coin, that would also allow quackle to have realistic skill levels (right now the only skill adjustment is essentially "stop thinking after a certain time and just play the best move found till then")