- Program Overview ›
- How To Apply ›
- Facts + Figures ›
- Research Groups
- Center for Bits and Atoms
- Degrees of Study
- Lab Life ›
Jeff Orkin Thesis Defense
Jeff Orkin Thesis Defense: "Collective Artificial Intelligence: Crowdsourced Social Interaction for Digital Actors"
Thursday, December 13, 2012 | 2:00pm - 4:00pm
Collective Artificial Intelligence is an end-to-end process for automating digital actors in videogames from thousands of recorded human demonstrations. Videogames are a dynamic story telling medium, unique in allowing players to influence the story being told. While today's games afford players incredible freedom to interact physically with other characters and the environment, the ability to interact socially–using language as action–is much more limited.
Current approaches to game development restrict social interaction to a relatively small, discreet set of hand-crafted branches, and do not scale to the thousands of possible patterns of actions and utterances observed in actual human interaction. By embracing data-driven approaches, it is possible to overcome this authoring bottleneck, and take a step toward realizing the full potential of games as an interactive story telling medium that robustly supports player choice. Doing so requires rethinking both the technology and the division of labor in videogame production.
Collective Artificial Intelligence combines crowdsourcing, automatic pattern discovery, and case-based planning. Content creation is crowdsourced by recording players online. Browser-based tools allow non-experts anywhere in the world to annotate events and long-term dependencies. Patterns discovered from this meta-data power a novel planning system, which combines plan recognition with case-based planning. This system has been evaluated by recording over 10,000 demonstrations in The Restaurant Game, and automating an AI-controlled waitress who can interact in the 3D world, and converse with a human customer via typed text or speech. Quantitative results show that the ability to infer context leads to a system significantly more capable of open-ended interaction with humans, while focus groups reveal additional factors to consider in the future to improve player engagement. Robust support for natural language interaction in open-ended scenarios has the potential to impact not only entertainment, but also other fields including education, therapy, and robotics.