Joshua Peterson Leads DARPA-Funded Study on AI Trust
Joshua Peterson, Assistant Professor in the Faculty of Computing & Data Sciences (CDS), performs groundbreaking research using data science tools to understand human behavior. This past fall, he secured DARPA I2O funding for the development of a platform for evaluating the trustworthiness of AI systems. The 18-month project, “Predicting Algorithmic Trust at Scale,” is designed to develop a “Trust Score”—a rating that predicts how much trust we will put in any given AI system to make decisions for us. To enable this, Josh and his collaborators will conduct large-scale experiments to model what makes people willing (or not) to defer to an AI decision-maker. These experiments will span diverse domains, including finance, law, and medicine. The research involves generating thousands of decision problems, assessing trust across a diverse set of AI systems, and analyzing human deference behavior. The ultimate goal is a tool that can evaluate new AI systems and predict public trust prior to deployment. The project is part of a multi-institution effort led by Princeton University, in collaboration with NYU and Cornell. While each team will collect data on specific domains, Josh’s effort at BU will focus on tasks related to classic decision-making research in psychology and economics. This work exemplifies convergent research at CDS, advancing both AI for Science and AI in the Public Interest.