Beyond these promising foundations there lies many difficult and interesting problems in poker research, from both a theoretical and practical point of view.
In developing a near-optimal game theoretic strategy, many pragmatic issues must be addressed that have never been properly considered in existing game theory studies. For example, the vast majority of mathematical poker models do not account for the drawing of cards and subsequent betting rounds. This has a radical effect both on theoretically correct strategy and practical considerations. Also at issue is the appropriate use of betting history in earlier rounds and previous actions in the current round.
It is clear that a maximal algorithm must observe opponent behavior and make appropriate strategy adjustments. How to best exploit any perceived weakness or predictability is a non-trivial problem. One of the keys to poker mastery is the ability to handle many different game conditions, and the strongest algorithms must have the ability to smoothly adapt to the prevailing characteristics, which may change during the course of a game.
While we believe limit Hold'em is the most natural choice for early study, new problems arise in other poker variations. For example, Sakaguchi and Sakai have proven that handling partial information, such as the opponent's face-up cards in the game of Seven-Card Stud, can make a game harder than having no information at all. Hi-Low forms of poker also have their own unique theoretical properties. No-limit poker may be the ultimate challenge within the domain, since it seems to emphasize the more nebulous poker skills, such as in-depth knowledge of the opponent and the ability to make fine judgements.
Finally, many other approaches could be viable for producing strong algorithms, such as genetic algorithms or neural nets. Each of these methods can be developed to be compatible with existing techniques, and the relative success of each paradigm can be settled at the virtual poker table.