Collective artificial intelligence for games

AI in games has evolved significantly in the past decade. While our characters have proven themselves as competent soldiers on the battlefield, it would be a stretch to call them digital actors. Despite the growing interest in co-op gameplay and interactive drama, today’s characters lack spontaneity, and lack the ability to improvisationally communicate and work together with human players. The issue is not simply a question of manpower or processing power, but a deep-rooted problem related to our current development approaches -- labor intensive authoring processes, requiring skilled engineers to create character behaviors from scratch.

The rising popularity of online games has opened the door to a potentially better approach called Collective Artificial Intelligence -- a combination of crowd-sourcing, data mining, and case-based planning, which empowers non-experts to create socially intelligent characters without programming.

In this talk, I will describe progress on a research project called The Restaurant Game (http://theRestaurantGame.net), where we have collected data from over 15,000 people playing the roles of customers and waitress online, and are using this data to automate digital actors. I will relate this new approach to my experiences in industry developing AI for games, including F.E.A.R. and No One Lives Forever 2.