We evaluate LLMs as tools for predicting human behavior at two scales: estimated population-level causal effects to guide theory and research, and individual-level agent models to predict behavior and model heterogeneity.
We develop and experimentally evaluate a nonpartisan AI voter guide that delivers trusted election information and examines its effects on voters' turnout intentions, candidate preferences, and partisan attitudes.
With colleagues, we develop a public interest roadmap for applying generative AI to improve social service delivery, from resource navigation to SNAP and TANF processing.
We evaluate potential positive impacts of AI chatbots on emotional and psychological well-being through brief, structured interactions grounded in prior research.
