The opportunity exists to structure poker research in such a way that each person's contributions will be relevant and usable by others in the field. In view of the limited usefulness of previous poker research, we believe a more scientific approach is both desirable and necessary.
In keeping with the well established paradigm of the natural sciences, steady progress can be made within a research discipline through the careful design of objective experiments and quantifiable results. It is the belief of this author that such rigour has often been lacking in previous investigations, although this is somewhat understandable given that computer science, and Artificial Intelligence in particular, are still in their formative years. We now look at some general methodologies and specific techniques applicable to the study of poker that are more akin to the classic scientific method.
As with the physical sciences, an important step is to eliminate the subjective and often unreliable perceptions of humans, in favour of a more objective methodology and interpretation of results.
To some, this might seem to be an absurd idea, since poker is usually regarded to be heavily dependent on human elements, such as psychology and bravado. The key point is that the human approach to poker is not necessarily how poker should be played. Indeed, perfect poker is based on certain probabilistic facts and fundamental strategic principles which may only be clouded by focusing on the particular biases and vagaries of human praxis. By seeking out the underlying truths of poker, it is possible to achieve higher standards than those of current human expertise. By concentrating on the ``physics'' first, we stand to learn more about this fascinating game, and about the abstract problems of decision making in conditions of uncertainty.
Some of the concrete steps we can take in this direction include the standardization of the topic, and the search for general solutions rather than game-specific methods. For example, betting strategies should be based on probabilistic evaluations of hand strength and potential, rather than case-by-case situations which may arise in a particular poker variation. The same betting principles can then be applied directly to any other poker variant, provided the ``front-end'' of the new program produces the same form of numerical evaluations.
Conversely, researchers should avoid rule-based expert systems whenever possible, because the number of subtle and distinct cases which must be considered in a game like Hold'em is prohibitively large. More to the point, even if a set of inference rules can be developed to attain an adequate level of play, it will not teach us much about the general nature of poker or decision processes.
For this reason, we believe future poker researchers should strive to use computer-oriented techniques whenever feasible and reasonable. There are many examples from Artificial Intelligence and computer strategic game playing which demonstrate that steady improvements in performance can be attained by exploiting the natural strengths of computers, while de-emphasizing the human model. Although it may sometimes be a significant challenge to cast problems of imperfect information in terms of high speed computation and perfect memory, these methods promise to eventually produce the most natural solutions to the task, from a digital logic point of view.
We now look at some specific examples of this philosophy in more detail.