Prototype for Increasing the Robustness of AI-Generated Plans

Artificial Intelligence methods can rapidly generate plans for complex situations. However, these plans may be rejected by military experts, who judge dimensions such as robustness using operational rather than computational criteria. This project is designed to capture the tactical and strategic concerns of the users, and incorporate these into the planning technology to filter out the unacceptable options, and highlight the preferred plans. Phase I will employ powerful new methods of Cognitive Task Analysis to elicit the factors underlying judgments of plan robustness. The knowledge elicitation will use the powerful set of cognitive probes with domain experts. The results will be represented first using a series of matrices to contrast robustness factors for different domain experts, and then for different scenarios. Then results will be represented using a hierarchical framework to present the relationships between robustness factors, and to reflect the important contextual factors. The entities and relationships will be codified and implemented in the form of a limited prototype which quantifies plan robustness and uses it to generate more robust plans. Phase II will expand the prototype to generate plans under realistic conditions in complicated domains so that the robustness criteria can be incorporated into other AI planning systems.