Human-Model Interaction in Public Sectors

Computational models—from probabilistic forecasts to AI foundation models—are increasingly shaping decisions in public life. It is critical to ensure that individuals and groups, from the general public to domain experts and data scientists, can perceive, use, and develop these models effectively and responsibly. In this talk, I will share my research on human–model interaction across three domains: AI in education, AI in decision-making, and election forecasting. I will begin with findings from our systematic survey of Human–AI decision-making to give an overview of this field. I will then present our work on understanding the needs of K–12 teachers and co-designing an LLM-based classroom assessment authoring tool. Finally, if time permits, I will also discuss our experiments using uncertainty visualizations to build appropriate trust in probabilistic election forecasts.