Bluff! One-on-One Algorithm Challenge

by Richard Smith 2002

One of the trickiest situations in a game of Bluff! is what to bid when it's down to one dice versus one. Going first is a mixed blessing in this situation, as is getting a star.

Following articles in Variable Pig regarding whether or not there really is a "winning" strategy, it was decided to put it to the test by means of a competition between robot players, each programmed with their own algorithm. The robot players play each other on a round robin to determine the best overall. Subsequently the winning algorithm is tested against human opponents (who do not know what the algorithm is) in real games of Bluff!

The competition works as follows:

Each algorithm "plays" all the others on a one v. one basis. Each "game" consists of drawing up two grids of all 36 combinations, one for each player going first. If there are plenty of algorithms submitted then the game can be played on a single round basis. Alternatively, the game can be run over a number of rounds - the first batch of algorithms play off and the results (but not the details of the algorithms) are printed and other players may then submit algorithms to challenge the existing ones.

Each player submits a set of rules for when bidding first, and one for when going second. These rules can be as simple or complex as the player wishes. An example of very simple algorithm is:
Going first, always open one 5, then challenge thereafter
Going second, always challenge

Rules can be conditional on the value of your own dice and on opponent's bids.

Players' algorithms are secret until after the game.