Ranking in Networks: Models, Leadership, and AI Preferences

Many social and behavioral data can be understood as pairwise comparisons: A team defeats another; a person leads another; or a language model prefers one response over another. The Bradley-Terry model is widely used to learn rankings from such data.

In this talk, I will show how simple extensions to the model let us develop a much more nuanced understanding of behavior from pairwise comparisons, through two examples.

The first examines the emergence of leadership among children using a rare archive of naturalistic observations from a mid-twentieth-century Han Chinese village in Taiwan. I develop a multivariate extension of the Bradley–Terry framework that connects leadership to networks of dominance, prosocial, and antisocial behavior. The second application concerns pairwise preferences expressed by large language models. Such preferences are often intransitive and are commonly modeled as noise around a single latent ordering. I will show that a noise-augmented mixture Bradley–Terry model can instead uncover multiple internally consistent preference orderings, suggesting that aggregate evaluations can hide coherent forms of preference heterogeneity in LLMs.

Together, these studies illustrate how extending simple ranking models with richer link functions, relational information, and latent mixtures can turn pairwise comparisons into more nuanced accounts of social organization and artificial intelligence.

Jean-Gabriel Young is an Assistant Professor of Mathematics and Statistics at the University of Vermont. He is also a faculty member of the Translational Global Infectious Diseases Research Center and the Vermont Complex Systems Institute. His research lies at the intersection of statistical inference, AI, and complex systems.

He recently spent a year as Founding Principal Data Scientist at Tilt.io, a seed-stage startup that raised $8M in 2025 to develop AI-driven financial indexes. Previously, he was a James S. McDonnell Foundation Fellow at the University of Michigan’s Center for the Study of Complex Systems, where he was mentored by Mark Newman. He earned his PhD in Physics from Université Laval in Canada under the supervision of Louis J. Dubé and Patrick Desrosiers.