My original research area was theoretical computer science, and I worked on the complexity of counting classes as well as doing some analysis of the properties of abstract machines. My current research lies at the intersection of artificial intelligence and human-computer interaction. It focuses on the integration of intelligent agents and virtual environments. These interests include research on agent architectures, virtual worlds (especially MUDs), online community building, and human-computer interfaces for entertainment and education.

I have also spent several years in industry creating commercial Web systems, and I am very interested in the development of the hypertext transfer protocol, D/HTML, style sheets, and generally both the technical and social issues relating to Web systems and Web design.

Current Research

I am presently a member of the Virtual Theater project at Stanford University, under the direction of Dr. Barbara Hayes-Roth. The purpose of the project is to develop intelligent agents and environments in which a user can take place in entertaining stories, which are either scripted in advance or improvised. The user controls some of the agents in the stories, and gives high-level instructions to the avatars. They use their own knowledge of the world, their desires, and their moods to decide how to implement those instructions.

My work is influenced by, though not directly related to, the main goals of the project. I am concerned with building intelligent agents with personality and affect (also often called synthetic characters or believable agents) that can serve as guides and inhabitants in virtual environments.

This work encompasses two major problems. The first is understanding the design principles for building believable agents; that is, agents that create the illusion of life. Part of my research is the beginnings of an operational theory of believability.

The second issue is in creating agents that can travel among many different environments with little or no a priori knowledge of their contents, but which are still able to provide useful and entertaining interactions. By introducing the concept of annotated environments, we hope to put enough abstract information into each world so that any adequately intelligent agent can query the environment for details or explanations of meaning, context, and feasible activities. Such annotations range from simple, factual information to procedural knowledge the agent can use to gauge its own status as well as a user's or interactor's status.

Here's a simple example. I would like to be able to build a room with a chessboard on a table. Any agent prepared to receive annotations should then be able to walk into the room and play chess against a user. Some agents may only need to be told that the game of chess is available -- they already know how to play. Others might understand the structure of a two-player, turn-taking game, but have to be given a model of the state space so they can search for good moves. Still others might simply have to be given advice by the room as to what move to make.

There are a growing number of virtual environments, most of them for games or education. By separating the domain's knowledge from the visiting agents, we can make the agents more adaptable, while making their knowledge of the environment controllable by the environment's designers. I see this as a practical, scalable way to improve existing characters.

The section on Publications contains my recent papers on believable agents in annotated worlds.

Related Work

There is growing interest in believable agents research. Here are a few of the many academic projects that work on these issues.

Our research is similar to the work that was done in the Oz Project at Carneige-Mellon University, though their work did not emphasize improvisation but guided, designed narrative experiences. Pattie Maes' Software Agents Group and Bruce Blumberg's Synthetic Characters Group both have worked on interactive, intelligent characters, either in virtual worlds or embodied in reality. Clark Elliott's Affective Reasoner project at DePaul is exploring a computational model of emotion. The IntelliMedia Initiative at North Carolina State specializes in characters embedded in educational software, exploring many of the same believability issues that we do.

A list of other projects attempting to create intelligent, believable agents is available on the Virtual Theater's Other Shows page.

Acknowledgments

This seems an appropriate place to thank both Intel and Microsoft for support in my research. From Intel, I have received an Intel Foundation Graduate Fellowship for 1997-1998 as well as a portable computer. From Microsoft I've received development software and documentation. Many thanks to both corporations.