Strategic Reasoning in Digital Zero-Sum Games

Digital zero-sum games are a challenging domain for artificial intelligence techniques. In such games, human players often resort to strategies, i.e., memorized sequences of lowlevel actions that guide their behavior. In this research we model this way of playing by introducing the algorithm selection metagame, in which agents select algorithms to perform low-level game actions on their behalf. The metagame provides a formal basis for algorithm selection in adversarial settings, presenting a simplified representation for complex games. We instantiate it upon realtime strategy game StarCraft, being able to discuss gametheoretic concepts in the resulting abstract representation, as well as generating a game-playing agent that successfully learns how to select algorithms in AI tournaments.