Nonlinear Social Learning and Cultural Evolution on Networks: A Perturbative Analysis
How nonlinear learning rules and network structure shape cultural evolution
Completed work, with Raja Marjieh and Thomas L. Griffiths.
Overview
How do individual learning biases and social connections shape the evolution of language and other cultural traits? In Nonlinear Social Learning and Cultural Evolution on Networks: A Perturbative Analysis, we study Bayesian learners who repeatedly observe and learn from their neighbors.
Our starting point is a model in which random social learning reproduces the learners’ prior at the population level. We examine how this changes when learners favor the local majority (conformity bias) or prefer one cultural variant over another (utility bias).
Methods and findings
Using perturbative expansions around random social learning, we show how nonlinear learning rules couple local network correlations to population-wide outcomes. This allows network structure to affect cultural evolution and produces departures from the baseline distribution determined by learners’ priors.
Simulations on random-regular, grid, and modular networks support the analysis. By holding population size and node degree fixed, we isolate the effects of network topology and show how conformity and utility biases produce different patterns of cultural change.