Practical Applications of Reinforcement Learning in Digital Games

Abstract

To explore the social applications of reinforcement learning, research and development of AI capable of playing games that simulate the complexity of the real world is beneficial. However, opportunities to learn from such complex games are rare. We have developed an AI learning interface for a widely played video game that possesses a considerable complexity. To demonstrate the feasibility of reinforcement learning through this interface, we have developed AI which can play the game with reinforcement learning, and the result indicate that the AI can handle the game's complexity. Furthermore, this effort showed the potential to bridge AI and the general public.