Sociotechnical AI Lab (SAIL)

AI's hardest challenges are not only technical but also sociotechnical, arising from how capable models interact with the people who use them. Our Sociotechnical AI Lab (SAIL) studies that interaction directly, asking: What can today's AI systems do reliably, and where do they fail? How do people understand, trust, and adapt these systems in everyday settings like health and urban life? And how can we design agents that are not only capable, but transparent and useful to the communities they serve?

SAIL develops and evaluates large language models and AI agents, and applies them with human-generated data to address sociotechnical problems. Our two methodological areas are tightly coupled: Agent design shapes the questions we ask about human-AI interaction, and findings on interaction guide how we build AI agents.

We ground this framework in two application domains. In healthcare, we analyze public experiences of treatments and policies and build agents that make health information easier to find and act on. In urban settings, we use crowdsourced data and build models to study community resilience, disaster impact, and equitable access, identifying public needs that are otherwise hard to observe at scale. Ultimately, we aim to build trustworthy AI systems whose reliability and transparency make them dependable partners in addressing sociotechnical challenges.