Michael Hieb, David Hille, and Gheorghe Tecuci
This paper presents a general approach to designing a computer opponent for nondeterministic adversarial games which is able to learn from interaction with a human expert. This approach is based on an integration of planning, knowledge acquisition and learning. We illustrate this approach with WARGLES (WARGame LEarning System), which plays strategic level wargames. WARGLES consists of a game playing system, a computer opponent and a learning component for the computer opponent. The computer opponent utilizes an adversarial planner with an incomplete and partially incorrect knowledge base. The learning component revises this knowledge base to improve the quality of play of the computer opponent based on interactions with an expert.