Events

Social science increasingly builds on foundations that move beneath it. Large language models, foundation models trained on life-course data and automated analysis pipelines change faster than the institutions that govern access to sensitive data. In this talk I reflect on what it means to build durable infrastructure under these conditions. Drawing on many years of work with population registers, linked survey data and trusted research environments in the Netherlands, I discuss approaches that bring models to data rather than data to models. These include secure environments in which researchers analyse data they cannot see, policy-aware workflows that encode governance rules into the analysis itself, and shared compute that lets AI methods run on population-scale records without loss of control. I then look ahead to designs that combine small, targeted measurement with register data to infer outcomes across whole populations with explicit uncertainty. Throughout, I ask how infrastructure can remain stable, trustworthy and open while its methods keep shifting and what this means for the future role of social scientists themselves.

The Sociology Seminar Series for Hilary Term is convened by Zachary Parolin and Carlo Perrotta. For more information about this or any of the seminars in the series, please contact sociology.secretary@nuffield.ox.ac.uk.