A Goal-Driven Approach to Support Decision-Making with Incomplete Information in Cyber Operations

Semi-autonomous cyber security (“cyber") operations require effective communication between a human operator and the underlying cyber systems that carry out the mission. We show a goal-driven approach to specifying mission objectives of such systems, where the system has controlled autonomy to refine the goals into executable plans. An ontology aligns goal decomposition and data representation with human-level concepts, and probabilistic extensions of Belief-Desire-Intention (BDI) agent architectures allow dealing with uncertainty about the agent’s state and the effects of its actions. Our framework effectively automates parts of cyber operations while conveying the state of operations to a human operator. A case study from the cyber operations domain demonstrates that this approach can outperform the traditional BDI agent approach while requiring minimal additional information by the agent programmer.