Up: Contents Previous: 5.5 The Help menu

6 How GIB works

GIB uses techniques from artificial intelligence to build the world's first expert-level computer bridge player. It is written in C and is based on an extremely fast double dummy solver. The principal technique used to construct the double dummy solver is partition search, which is described in a technical paper of the same name that was presented at the 1996 conference of the American Association for Artificial Intelligence.

When it is its turn to play a card, GIB decides what to do by dealing out the opponents' hands at random, consistent with their bidding and play thus far. The double-dummy engine is then used to find the best play on the deal that is produced. Of course, this particular play may be bizarre, catering to the specific distribution generated. But GIB deals the opponents' hands many times, and selects the play that works best on average.

Bidding is similar. When faced with a decision, GIB once again deals out random hands for partner and for the opponents. It then considers each bid it might make, and projects how the auction would proceed on each sample deal. It also analyses the deals double dummy, and then makes the bid that appears likely to lead to the best overall contract.


Up: Contents Previous: 5.5 The Help menu