Standard Economic Models in Nonstandard Settings – StarCraft: Brood War

This paper reviews an entrant to this year's Star-Craft: Brood War AI tournament, CUNYbot. CUNYbot makes strategic decisions using a low-dimensional economic model traditionally used to describe the behavior of countries, but has applications for any real-time-strategy game (RTS) where the capital/labor ratio (k) and technology/labor ratio (t) are critical. CUNYbot first tunes the economic model parameters between games using a genetic algorithm, allowing it to learn an optimal static k for each built-in AI race. In the second step of the project, CUNYbot models the opponent during the game and is permitted to react to the opponent's inferred choices in k. The reactive model adopts a greedy "tit-for-tat" style strategy against all three races. This paper reviews the features of the capital-augmenting Cobb-Douglas model, the game theory behind the reactive strategy, and demonstrates adaptation to particular opponents in the RTS context.