AI for Player Avatars

When considering game playing, the human becomes a critical element of the system. Rather than regard this human element as stochastic interference in AI research, we believe that an important strand of future research will embrace and use the characteristics of the human game player to create more engaging games, and to advance AI research through capturing the characteristics of human players and responding more intelligently. The recent spectacular rise in popularity of online multiplayer games has been advanced as reducing the importance of AI in games, but this rise provides an excellent opportunity to use captured network data for millions of online game instances and use this data to analyse human player characteristics, to classify, generalise, understand and mimic play actions, and create more engaging and effective AI. AI controlled Non-Player Characters (NPCs), already one of the principal sources of gameplay in current games, will only continue to rise in importance as increased CPU power provides the opportunity to tailor AI behaviour and react to human player characteristics, provided AI research can find a way to deliver research techniques which can engage the human player.

Investigations into AI techniques which automate the capture and modelling of human behaviour have made a promising start with the commercial Forza Motorsport game, where an avatar uses machine learning to copy the driving style of a human player, so that a player’s avatar may continue to play even when the human is not logged on. The idea of player persistence in a virtual game world is likely to increase in importance, especially on massively multiplayer online games, if the AI community can deliver usable technologies to the games industry. An extension of this idea is to consider games where the creation of AI enhanced characters is a major feature of gameplay. Ideas such as this have been explored in the commercial Black and White game, where the player trains an AI character, rather than directly controlling that character. Further ideas for AI creation as a gameplay device are considered in. Increasingly, there is recognition that AI can be used to modify the characteristics of a game during gameplay, in ways which feel natural to the game player, to increase enjoyment and challenge. There will be a special session on this topic at the upcoming AAAI Artificial Intelligence for Interactive Digital Entertainment (AIIDE) conference this year. This is an area where the gap between the goals of AI research and those of the games industry are relatively small, and so it will be a promising area to consider.