Modeling Humans in Collaborative AI Systems

As AI algorithms and technologies become reliable and robust, they are being employed in a variety of domains to solve complex problems in transportation, health, education, banking, etc. Not surprisingly, the questions related to relationships between intelligent agents or systems and humans have taken center stage in AI research. While human-agent interaction research has made some useful strides towards natural interactive interfaces for AI systems, it crucially misses a challenge that is critical for robust human-agent collaboration - modeling, learning, and reasoning about the human.

In this talk, Dr. Shiwali Mohan characterizes the space of human-agent collaboration to develop an understanding of various modeling needs. She grounds the discussion in various AI systems projects that she has worked on in the past few years including interactive task learning for robots, intelligent planning for sustainable transport, and intelligent scaffolds for health and well being. She summarizes an approach to building such systems with a focus on the challenges of evaluating collaborative human-AI technology, and concludes the talk with lessons learned and an outlook for future work as well as a brief summary of exciting new DARPA projects at PARC.