A computer that learns the game by watching you play

[Terry McCracken]

A computer that learns the rules of a game by watching you play



A Parisian researcher has created a computer system that can learn the rules of games by watching you play — and then beat you at it.

http://www.extremetech.com/extreme/1325 ... g-you-play

http://liafa.jussieu.fr/~kaiser/pub/lea ... lexity.pdf

[kgburcham]

well it is a good start but without the advancements in todays top program code, it will have to weed out GM blunders somehow.

[Terry McCracken]

well it is a good start but without the advancements in todays top program code, it will have to weed out GM blunders somehow.

I would think it's many years away before it could tackle chess with this method.

But it's interesting.

[Daniel Shawul]

I know an effort on general game playing but that does n't learn rules. You have to specify at least the rules (game terminations etc.) . The rest that is the AI is done automatically by the computer selecting the best algorithm depending on the rules. I think it would be very hard to learn what a fifty move rule is just by looking (visual inspection). It would take a lot of training but with big database and smart rule learning algorithms it could be possible. Tic-tac-toe/connect 4 have very simple rules however. And also other 'place a stone' games such as Go could be easy to learn the rules but probably not chess...

[Harald]

You can always win against this computer.

If you are in trouble just capture any piece of your opponent with any of
your own pieces regardless of the piece positions or do another move you
invent on the fly. If the computer "learns" the move and tries to use it
against you tell him it is forbidden now since it is not Tuesday afternoon
07:35 and you are not facing south.

This method is well known:
http://en.memory-alpha.org/wiki/Fizzbin
http://en.wikipedia.org/wiki/Dragon_Poker
http://www.ephraimkishon.de/en/my_favorite_stories.htm

Harald

[Daniel Shawul]

You can always win against this computer.

If you are in trouble just capture any piece of your opponent with any of
your own pieces regardless of the piece positions or do another move you
invent on the fly. If the computer "learns" the move and tries to use it
against you tell him it is forbidden now since it is not Tuesday afternoon
07:35 and you are not facing south.

This method is well known:
http://en.memory-alpha.org/wiki/Fizzbin
http://en.wikipedia.org/wiki/Dragon_Poker
http://www.ephraimkishon.de/en/my_favorite_stories.htm

Harald

Lol. You wanna try some jewish poker ? I have a number in my head that is not 12 ... On a serious note, general game playing is serious buisness http://games.stanford.edu/. There are some strong engines that play a game where the rules are specified at run time. All that this new paper did is to try and infur the rules using visual recognition which it already does well for tic-tac-toe and connect-4. I tell you this the future robocop..