Dynamically adapting an AI game engine based on players' eye movements and strategies

Artificial intelligence (AI) game engines have frequently been used to drive computational antagonists when playing games against humans. Limited work exists, however, on using human players' psychophysical measures to directly parametrise AI game engines. Instead, parameters to optimise AI performance are usually derived from general play-related data or user models. This paper presents novel research on using eye movement data in addition to data on users' strategies to adapt the live play of a computational antagonist in the visuo-spatial strategy game, Hex. It offers a set of suitable parameters for both types of data. A systematic evaluation of the approach showed, among other things, that using eye movement data led to significantly better gameplay experience for human players, as they experienced less frustration with sufficient challenge. Findings are discussed not only with regard to designing gameplay experience, but also their more general ramifications on using live psychophysical data for intelligent interactive systems.