Chapter # Policy-based Agent Directability KAREN L. MYERS and DAVID N. MORLEY Artificial Intelligence Center, SRI International, 333 Ravenswood Ave., Menlo Park, CA 94025 Keywords: adjustable autonomy, advisable systems, mixed-initiative, agents Abstract: Many potential applications for agent technology require humans and agents to work together to achieve complex tasks effectively. In contrast, most of the work in the agents community to date has focused on technologies for fully autonomous agent systems. This paper presents a framework for the directability of agents, in which a human supervisor can define policies to influence agent activities at execution time. The framework focuses on the concepts of adjustable autonomy for agents (i.e., varying the degree to which agents make decisions without human intervention) and strategy preference (i.e., recommending how agents should accomplish assigned tasks). These mechanisms enable a human to customize the operations of agents to suit individual preferences and situation dynamics, leading to improved system reliability and increased user confidence over fully automated agent systems. The directability framework has been implemented within a BDI environment, and applied to a multiagent intelligence-gathering domain. 1. INTRODUCTION The technical and public press are filled these days with visions of a not-too-distant future in which humans rely on software and hardware agents to assist with tasks in environments both physical (e.g., smart homes and offices) and virtual (e.g., the Internet). The notion of delegation plays a central role in these visions, with a human off-loading responsibilities to agents that can perform activities in his place. Successful delegation, however, requires more than the mere assignment of tasks. A good manager generally provides directions to subordinates so that tasks are performed to his liking. To ensure effectiveness, the manager will monitor the progress of subordinates, interrupting occasionally to provide advice or resolve problems. Analogously, effective delegation of tasks to intelligent agents will require tools by which a human supervisor can interact with agents and direct their operations. The agents research community has, for the most part, focused on the mechanics of building autonomous agents and techniques for communication and coordination among agents. In contrast, little attention has been paid to supporting human interactions with agents. Most agent frameworks lie at the extremes of the interaction spectrum, either assuming full automation by the agents with no means for user involvement, or requiring human intervention at each step along the way (i.e., teleoperation mode). Recently, however, there has been increased interest in agent systems designed specifically to