Page under construction.



Belief and Intention Recognition with Narrative Planning
Narrative planners are typically used generatively: to generate stories given a symbolic domain of actions and a fully grounded initial state. The initial state includes characters' goals, and in our work, their beliefs with complete theory of mind. This project used our belief model to do narrative planning in reverse: to infer the beliefs and goals of agents by observing their actions in a given scenario.
Publication

AIIDE paper on belief and intention recognition


Rachelyn Farrell, Stephen G. Ware. Narrative planning for belief and intention recognition. In Proceedings of the 16th AAAI international conference on Artificial Intelligence and Interactive Digital Entertainment, pp. 52-58, 2020.

Dataset

Intelligence Analyst Workflow Dataset


This was a DoD-funded project and a multi-year collaboration with the Laboratory for Analytic Sciences at North Carolina State University. We developed a fictional insider threat investigation scenario, and built tools for intelligence analysts to explore the evidence and develop hypotheses about who leaked a piece of software. We compared different methodologies to improve their workflow, including using belief and intention recognition to automatically infer possible agent motivations, suggesting possible alternative explanations for the evidence.