[AyunuS]

I was playing a tourney a few days ago when GIB (v 19, probably not fixed though) did something...understandable, but still clearly inferior. We were going for 6 hearts and had lost 1 trick so far. I had a QJ of spades, the A was played already and it had none left. It needed to slough off one card from its hand in order to be able to make the contract. It played one of the QJ and got rid of the card it needed to, and neither opponent played the K. I assume it assumed the opponent to play last must have the K. It could have been guaranteed to make the contract, but it played the last spade. When the first opponent did not play the K, it just got rid of some random card from its hand like it was assuming it had made the contract and the player who went last couldn't have the K, except that that player did, took the trick, and then the contract failed. I admit the player with the K should have taken it the first time, but still, GIB had a guaranteed making of the contract with sure information at that point that didn't need to rely on likely information.

[BunnyGo]

AyunuS, on 2011-December-15, 17:37, said:

I was playing a tourney a few days ago when GIB (v 19, probably not fixed though) did something...understandable, but still clearly inferior. We were going for 6 hearts and had lost 1 trick so far. I had a QJ of spades, the A was played already and it had none left. It needed to slough off one card from its hand in order to be able to make the contract. It played one of the QJ and got rid of the card it needed to, and neither opponent played the K. I assume it assumed the opponent to play last must have the K. It could have been guaranteed to make the contract, but it played the last spade. When the first opponent did not play the K, it just got rid of some random card from its hand like it was assuming it had made the contract and the player who went last couldn't have the K, except that that player did, took the trick, and then the contract failed. I admit the player with the K should have taken it the first time, but still, GIB had a guaranteed making of the contract with sure information at that point that didn't need to rely on likely information.

This is particularly strange considering GIB has been known to not repeat a finesse when it *does* need it.

[AyunuS]

Now that is interesting. I know how to program and thus I've dealt with tons of bugs before. This sounds like the if/else statements in one place are switched.

[Antrax]

That's not really how GIB (or any AI) works. If you're interested, someone linked me once to a really excellent article about GIB: http://www.jair.org/...0-1957-jair.pdf