Speaker: Jakub Bijak

Institution: University of Oxford

Date: 14:00 BST, 11th June, 2026

Talk Title: Models as Theory-Building Machines? Insights from Agent-Based Simulations of Human Migration

Abstract: Abstract (150 words): Demography has been famously strong on data and weak on theory. Yet, early 21st century witnessed renewed interest in enhancing inference about observed population patterns (what?) with formal explanations of underlying mechanisms (how?) and causal links (why?). One possible class of models well suited for building such “theories of the middle range” (RK Merton) for specific demographic processes involves agent-based simulations, whereby simulated ‘agents’ interact according to a mix of decision rules and empirical data. In this talk, I will present an example of applying the agent-based approach to modelling migrant route formation, integrating interdisciplinary insights from demography, statistics, cognitive psychology and computer science. Bayesian inference will be used as an integrating perspective, focusing on the role of uncertainty in shaping model outputs. I will sketch the promises and limitations of this approach, and will invite a discussion of prospects and constraints of using large language models for future model-building.

Bio: Jakub works as Professor of Demographic Science at the Leverhulme Centre of Demographic Science and Reuben College, University of Oxford, having previously spent 17 years at the University of Southampton. He holds PhD from Warsaw School of Economics, obtained for pioneering research on Bayesian migration forecasting. A statistical demographer, Jakub works on demographic uncertainty, migration and population modelling and forecasting, and demography of armed conflict, occasionally venturing into other areas, such as palaeodemography. He led an ERC project Bayesian Agent-Based Population Studies (BAPS, 2017-22, www.baps-project.eu), and a Horizon 2020 collaborative project Quantifying migration scenarios for better policy (QuantMig, 2020-23, www.quantmig.eu).