next up previous contents
Next: Machine learning Up: Previous Computer Science Studies Previous: Previous Computer Science Studies

Findler

One of the first studies of computer poker was done by Nicolas Findler Findler77,Findler78. The variation of poker used in his research was a simplified version of five-card draw poker. During the years Findler's project was carried out, he developed a variety poker-playing programs, each different in its structure and approach to decision-making. Virtually all of the early investigations into computer poker were done by Findler and his associates. He would use the game of poker as a model for a variety of studies which is now known as Artificial Intelligence. Findler was one of the first people to recognise the potential benefits of research into poker, as a model for decision making under uncertainty, but since his computer poker research was inspired by the social sciences, his approaches were based significantly on the psychological processes of human thought, rather than a mathematically oriented analysis. The poker variation addressed by Findler was not entirely realistic, being a form of two-person Draw poker without any opening requirements or ante, but it is still closer to real poker than any previously existing mathematical model at the time. Findler used a combination of a probabilistic assessment of hand strength, with the collection of frequency data for opponent behaviour to support the refinement of the models of opponent. The frequency data was collected separately for each opponent and each round of play. He considered that:

``In order to program a computer to play poker well it is necessary to understand the cognitive processes employed when human beings play poker. (The mathematical theory of games can only treat simplified versions of the game).'' Findler's goal was not to create a world-class poker-playing program and, as a result, none of his programs were very strong players. The algorithms developed also incorporated certain weak strategies, such as betting in exact proportion to the strength of its hand (which would convey too much information to an opponent). Findler's research was not geared toward producing the strongest possible poker program, but was focused on simulating the thought processes of human players. While these may be interesting problems in their own right, they impose severe limitations to the performance of a program based on these ideas and is of limited usefulness to the future development of a computer playing program. While the work Finder did was worthwhile, the goal of producing high performance poker programs will require not only different techniques, but a completely new, more scientific and mathematical framework from which to build upon.


next up previous contents
Next: Machine learning Up: Previous Computer Science Studies Previous: Previous Computer Science Studies
Jason R Carlton
2000-11-13