Title: Complementarity in Human-AI Collaborations

Abstract: Multi-agent decision pipelines (including human-AI collaboration) can outperform single agent workflows when complementarity holds, i.e., different agents bring unique information to the table to inform a final decision. In this talk, I will present a set of decision-theoretic frameworks we developed to measure and leverage complementarity in multi-agent decision-making. I will share examples of how these tools help diagnose when people should rely on AI, communicate what information is missing, and train language models to surface complementary signals from text to support better decisions.