This thesis presents an investigation into how to effectively optimize the production of machine generated game AI, exploring the behavior tree model and evolutionary computation.
The optimization methods focus on providing a ‘proof of concept’ that a system can be designed and implemented, through a series of studies, being capable of producing game AIs with alternative behaviors within a playthrough of a game. The construction of these behaviors should be informed by the evaluation of previous behaviors, as well as show a quantifiable improvement in performance.
The studies evaluate the performance of a generated AI for the game XCOM 2, a Turn-Based Tactics video game. The AIs will be evaluated by running combat simulations against the standard AI implemented by its developers. Ultimately, the results of the process led to an user experiment, in which the most successful machine generated game AI won 50% of matches.