Multi-Agent Distributed Continuous and Asynchronous Planning (MADCAP)

Human adversaries, rapidly evolving malware and pandemics, large scale natural disasters, and disruptions to the national infrastructure all require complex, dynamic defensive responses. While our ability to communicate has increased, our response remains sequential: our plans are based on a priori knowledge and assumptions, and our execution follows our plans. Additionally, we do not take advantage of the information available from first responders: namely, detailed knowledge of the situation; current and anticipated response needs; and the effectiveness of current response tactics. While some response tasks can be implemented by machines, complete automation is impractical for large-scale crises where an effective response requires diverse knowledge and skills not resident in one place. Here, planners and responders alike would benefit from a means to continuously assess, collaborate, and replan defenses to efficiently use the resources available. To address this need, we propose to demonstrate the feasibility of developing a Multi-Agent Distributed, Continuous, and Asynchronous Planning (MADCAP) architecture for the cyber-defense domain. MADCAP enables human and computer agents to perform cyber attack response planning by creating a distributed plan workspace, employing a partial-order plan representation, and using a novel approach to plan filtering and selection using multi-objective optimization. BENEFIT: A MADCAP-based planning system will have immediate and tangible benefit for military and government-sponsored cyber security defense and planning programs. MADCAP incorporates up-to-the-minute information about the precise nature of a cyber attack and remediation success to enable responders and planners alike to collaboratively and continuously replan in a rapidly evolving and distributed work environment. Such a system will benefit not only cyber defenders but those in other domains where plans must be continuously and efficiently adapted.