EconCS Seminar

 

Lecturer: 

Prof. Sigal Oren (BGU)

Title: 

AI-Assisted Decision Making with Human Learning

Abstract: 

AI systems are increasingly used to assist human decision-making, often leaving the final decision in human hands. We study AI-assisted decision-making in repeated interactions where the human learns over time from the algorithm’s recommendations. In our framework, the algorithm—aiming to maximize accuracy according to its own model—selects which features the human may consider, while the human makes predictions using their own, less accurate model.

This setting gives rise to a fundamental trade-off: should the algorithm choose features that are most informative, encouraging human learning even at the cost of short-term accuracy, or instead select features aligned with the human’s current understanding to maximize immediate performance? We show that the optimal feature-selection strategy admits a clean combinatorial characterization. As the algorithm becomes more patient or the human learns more effectively, the algorithm increasingly favors informative features, improving both decision quality and human understanding over time.

Joint work with Gali Noti, Kate Donahue and Jon Kleinberg

Location: 

Room 130, Feldman Building, Edmond J. Safra Campus.

Click here to add the EconCS Seminar to your Google Calendar