Research Directions in Interactive AI

The MAPLE (Multi-Agent Planning and LEarning) research group at UMBC has several ongoing projects that empha-size user interaction with AI methods. This position paper briefly summarizes these projects: (1) representing and incor-porating a user’s qualitative domain knowledge into machine learning algorithms; (2) creating understandable visual rep-resentations of high-dimensional probabilistic domain mod-els (Bayesian networks); (3) developing cost-sensitive con-straint satisfaction methods that allow the system to reason about the cognitive costs of involving a user in checking constraints; and (4) integrating automated planning and con-straint satisfaction methods with interactive plan editing tech-niques to permit mixed-initiative plan generation in complex, real-world domains