Dartmouth Engineering Professor Awarded Grant to Develop AI-Human Trust

In the abstract for the project, "Quantifiable Expectability and Measurable Intentions: Computational Calibrated Trust in Human-AI Joint Activity," Santos writes that without expectability, defined as being anticipatory, understandable, and precautionary with regards to one's behavior, trust cannot be established.

The research will reduce clashes and gaps caused by differences in how each team member is adapting and learning during joint activity and decision-making. Santos aims to do this by learning the rewards that drive the behaviors, decisions, and actions of both human and AI partners.

"Complementarity and diversity can be important strengths in partnerships," said Santos. "Aligning or matching up intentions doesn't mean that they think the same."

He will build a prototype during the year-long grant period that demonstrates how powerful and beneficial a human-AI team can be when trust is calibrated between them. The project seeks to optimize human-AI performance and efficiency through shared joint action for the purposes of digital data analysis and reporting.

The Trusted AI Challenge Series is sponsored by Innovare, the State University of New York, IBM, New York State Technology Enterprise Corporation (NYSTEC), and the National Security Innovation Network (NSIN).