Partners Who Grow Together: Collaborative Machine Learning in Video Game AI Design

Abstract

The majority of research at the intersection of AI and video games focuses on developing agents capable of playing games without human input, or developing AI game enemies. The research in this paper explores a counter approach, whereby a player trains an a AI partner during game play and learns to play cooperatively with the agent. We created a 2D video game that allows the player to cooperate with an AI agent manipulated by two underlying algorithms, either with reinforcement learning or a random process. For our reinforcement learning approach we used a Q-learning table, that is updated based on the player. We found that players engaged strongly with the idea of training their own custom AI agent and believe this shows significant potential for future exploration.