SYMBIOTIC LEARNING IN COMMERCIAL COMPUTER GAMES

This paper proposes an approach for learning team behaviour to improve agent intelligence in team-oriented commercial computer games. The approach, named 'symbiotic learning', focuses on the exploitation of relevant gameplay experiences. The results of an experiment in the game QUAKE III show the symbiotic learning approach to be able to successfully learn effective agent behaviour. We conclude that symbiotic learning can be used during game development practice to automatically validate and produce AI, and, provided a good balance is found between exploitation and exploration, the approach can be applied in practice for the purpose of online learning in commercial computer games.