Adaptive Interfaces for Value-Driven Agents

As autonomous agents perform increasingly important and complex tasks, they have the potential to greatly expand human capabilities. For this collaboration to be successful, a human must be able to effectively communicate goals and preferences to the agent, so the agent's choices will be consistent with those of the human user. We present our plans for integrating two lines of research towards this end: an architecture for value-based agents, called Icarus, which employs a numeric utility model to resolve choices in skills; and research on adaptive user interfaces that acquire a numerical model of user preferences from experience. Taken together, these capabilities offer a means of tailoring autonomous agent behavior in complex domains.