Distributed Coaching for an Intelligent Learning Environment

1 Institute for the Learning Sciences, Northwestern University {forbus ureel jbaher skuehne } @ils.nwu.edu 2 Xerox Palo Alto Research Center [email protected] 3 Department of Mechanical Engineering, Northwestern University [email protected] Abstract Several barriers hinder the widespread application of AI-based educational software. School and student machines are often underpowered, keeping software and case libraries updated can be difficult, and customization typically requires AI expertise. The widespread growth of Internet access, combined with appropriate AI technologies, enables the creation of distributed coaches that can help overcome these barriers. We describe a distributed coaching system for a deployed intelligent learning environment in engineering thermodynamics. Part of the coach resides on the student's computer, with the rest residing in a server accessed via email. The on-board coach handles common kinds of contradictions in student's assumptions and makes suggestions about parameter values based on its understanding of the teleology of the student's design, derived via Bayesian reasoning. The email coach provides additional analysis help and uses analogy for design coaching, providing step-by-step advice on how principles in a web-based library can be applied to a student's particular design. The distributed coach is currently undergoing field testing.