Automated support for human mixed initiative decision and control

We describe a collaborative multi-agent system that uses a dynamic, model-based representation of a control system to provide a substrate for multiple software and human agents. This representation provides the semantics for a set of software agents that dynamically manage the human-machine dialog as it evolves with the situation. The interface design agents use model-based productive reasoning about situations and interaction design, and create specialized user interfaces that meet the current interaction needs of the user. We are successfully applying this approach in the domain of mixed-initiative command and control of teams of semi-autonomous vehicles. We have found that the combination of a productive modeling approach to basic reasoning on situations, together with the ability to synthesize user interfaces based on the state of a situation, provides the ability to automatically design useful interactions. User interface generation technology supports multi-agent systems that must accommodate varying levels of participant autonomy, particularly in dynamic situations with varying requirements for mixed-initiative control by human commanders.