Learning adversary modeling from games

Abstract

Since ancient times, adversary modeling has been used during wargaming exercises in which military leaders have recreated past battles or simulated future battles in order to educate military professionals. Although the technology today is much different, adversary modeling still serves the same goals - to help military professionals learn tactics from past successes and mistakes. In the computer age, highly accurate models and simulations of the enemy can be created. However, including the effects of motivations, capabilities, and weaknesses of adversaries in current wars is still extremely difficult. Limit Texas Hold'em poker, with many attributes similar to real-world warfare, is an excellent test-bed to study and improve adversary modeling. For example, stochastic outcomes which deal with multiple independent agents, deception, and acting amidst uncertainty, are some of the aspects of poker that closely resemble important aspects of warfare. These attributes make poker a better choice as a study platform than other traditional games, such as chess, where there is no deception or uncertainty. The defined rules of poker provide researchers with a controlled environment to improve and test adversary-modeling techniques. Perfecting adversary modeling in poker will allow simulators to improve and generate more accurate models for wargames, giving warfighters the advantage in current and future battles.