The poker game (at least most variants) has a complexity beyond the reach of computers. However, methods are developed to approximate the perfect strategy (in terms of game theory) in a face to face (two players) game. Furthermore, algorithms more effective are designed for situations where more players are involved. Perfect strategy has multiple meanings in this context:
- for game theory and according to the method of minimax is one that outweighs any other strategy;
- for programs, the problem is that this optimal strategy varies depending on the expertise of the opponent and weaknesses that it becomes possible to exploit at his expense. In this case, the optimal strategy is to model these weaknesses to take advantage.
Some of these systems are based on the Bayes theorem, Nash equilibrium, methods of Monte Carlo and neural networks.
The best known research unit in this area is that of the University of Alberta who developed Poki, PsOpt and Polaris. Among the members of this team include Jonathan Schaeffer, initiator in 1991 of this research group, Neil Burch Darse and Billing, two Polaris designers.
Polaris, playing Texas hold’em, was tested against two world famousAmerican poker players, Phil Laak and Ali Eslami, at the Annual Conference on Artificial Intelligence, which ended July 24, 2007 to Vancouver. The two human players won narrowly after four games with a draw, a victory for the software and two wins for men. To balance the randomness of the game, every human was playing alone against an instance of the computer program, and Phil Laak received a two-card hand the same as that received the Polaris instance against Ali Eslami.