Our relationship with computing devices is becoming more situated, physical, and embodied. At the same time, everyday people's interaction with artificial intelligence is increasingly commonplace and personal. How can we design - and evaluate - AI that fits non- expert human behavior? How can we build robots and synthetic characters that appropriately learn from people, and work with them more collaboratively, effectively, and fluently?
I will present an array of research studying AI and robotics coming in contact with everyday people. We have investigated how ordinary people want to teach a machine-learning game character, and how traditional machine-learning may have to change to fit these people's expectations. We also found that if a game character anticipates its human partner's moves, it becomes more fluent, more effective, and seems more intelligent and committed. I will present a cognitively plausible embodied framework for collaborative robots, which was shown to have significant effects on untrained human teammates. Finally, I will discuss lessons theater acting has to offer designers of artificially intelligent agents, and a human-robot theatrical performance that resulted from this exploration.
While still some time off, robots are bound to be commonplace in homes, offices, shops, and schools, and may also be central to the 21st-century game industry. I will conclude by introducing the notion of Personal Robot Design as a new interdisciplinary design field, and describe the process of designing a non-anthropomorphic robotic desk lamp.
Guy Hoffman holds an M.Sc. in Computer Science from Tel Aviv University, having graduated from the Interdisciplinary Program for Outstanding Students with concentrations in computer science, film studies, and psychology. His thesis in the Computer Vision and Robotics lab was entitled "A Quadtree Approach to Motion Segmentation into Layers". His industry experience includes the position of Vice President for Product Development at Internet startup uTOK inc., as well as that of Project Manager at Check Point Software Technologies, where he led the design, implementation, and marketing of a cellular security product. He then studied animation and data visualization at Parsons School of Design.
Hoffman received his Ph.D. from the MIT Media Laboratory's Personal Robots Group, with a dissertation entitled "Ensemble: Fluency and Embodiment for Robots Acting with Humans." His research is concerned with the various mechanisms that underly joint action. Recent work includes the formulation and exploration of embodied cognitive architectures for robotic agents that enable them to perform tasks fluently with their human counterparts, and what it means to practice to get better as a team. Other research is concerned with the application of theater acting techniques to artificial intelligence and robotic stage performers. His robot, AUR, is the winner of the 2007 IEEE Ro-Man International Robot Design Award.