Kairos to Study Human-AI Decision-Making for DARPA’s In the Moment (ITM) Program

Kairos Research has kicked off a new DARPA project as a member of a team led by Raytheon BBN Technologies. DARPA’s “In the Moment” program aims to develop AI algorithms whose decision-making attributes align with those of trusted humans experts. Models of human decision-making in difficult domains (i.e., domains where there is no agreed upon right answer) will need to be created in order to facilitate the development of the algorithms. As part of the “Attributes of Decision-makers to Enable Prediction of Trust (ADEPT)” team, Kairos Research will lead quantitative human subjects research to develop decision-making scenarios and measurement approaches that enable modeling of statistical distributions of human decision-making attributes.

Dr. Amy Summerville, a Kairos Research Senior Research Scientist who will lead the company’s efforts on the program, noted the importance of behavioral science research in guiding the team’s overall approach. “Human decision-makers differ in so many ways that trying to model and tune an AI to all of these would be an almost intractable problem. Instead, our approach is informed by theory and empirical research from social and organizational psychology on what attributes drive trust between human teammates. This will allow us to center our efforts to understand what factors impact a human decision maker’s trust in an algorithmic decision tool.”

Dr. Locke Welborn, another Kairos Senior Research Scientist on the ADEPT team, observed that the program tackles a particularly challenging area of AI. “In The Moment goes beyond the problem of having an algorithmic tool produce the right answer to looking at more ambiguous decision contexts, such as battlefield medical triage, where human experts often do not agree on the best course of action.  The Kairos team is excited to have the opportunity to apply our expertise in decision making, moral judgment, and human-machine teaming to advance the science of collaborative human-AI decision-making in these difficult situations.”

Posted in