SBIR/STTR Award attributes
The Air Force vision for the coming decades calls for operational agility strongly supported by artificial intelligence (AI), such as autonomous uninhabited aircraft systems (UAS). To make operations involving human-autonomy teaming a reality will require not only sophisticated autonomy but extraordinary trust and ease of interaction by the humans with their autonomous teammates. Critical to these is common ground, the shared conceptualization of the team’s operational domain. Current AI possesses impoverished conceptual structures relative to a human teammate, and this is one reason human-AI common ground has been limited. To address this limitation, Charles River Analytics proposes to design and demonstrate the feasibility of an Embodied Common Concept Ontology (ECCO), a system that gives software agents the ability to establish, maintain, and repair common ground with human teammates via shared conceptual structures. ECCO uses a types-based probabilistic knowledgebase to give an agent human-like features of conceptual structure and reasoning. Applied to the domain of spatial relationships among objects in Phase I, these conceptual features include conceptual blending to recognize novel relationships among known objects, situated categorization to enable contextual reasoning about relationships, and concept degeneracy to enable the system to recognize and work with ambiguous relationships.
