Robert P. Goldman, Stephanie Guerlain, Christopher Miller and David J. Musliner
We are developing mixed-initiative systems that provide users with flexible control over evolving processes. For such systems to be feasible, both users and automated agents must be able to operate at multiple levels of abstraction. We are designing mixed initiative systems organized around shared task models to achieve this multi-level task sharing. These task models will allow both human and automated agents to view the ongoing state, to control tasks and to project future requirements. Our task models will have semantics, allowing a suitable interpreter to directly execute tasks for users. We discuss this approach and our initial implementation of a system that assists users in proposal preparation.