Learnable Buddy: Learnable Supportive Ai In Commercial Mmorpg

In commercial massively-multiplayer online role-playing games (MMORPG), players usually play in a populated environments with simple non-player game characters. These non-player characters have fix behaviour. They cannot learn from what they experience in the game. However, MMORPG environments are believed to be greatly suitable for training AI, with plenty of players to provide tremendous amount of feedback, and persistent worlds to provide learning environments. This paper presents an experiment to find out the potential of MMORPG environments for fast learning evolutionary AI. The genetic algorithm is chosen as our learning method to train a non-player character to assist real players. We use a game server emulator and custom game clients to simulate and run a commercial MMORPG. Clients are divided into two groups, real players and "helpers". The results show that helpers can learn to assist real players effectively in small amount of time. This confirms that evolutionary learning can be used to provide efficient learning in commercial MMORPG. It also verifies that MMORPG provide great platforms for research in evolutionary learning.