Content area
Abstract
Epistemic polarization arises when the statistical dispersion of a population's beliefs increases, especially when all agents update on exactly the same evidence. Factionalization arises when not just one belief, but many different beliefs become correlated across a population. Polarization and factionalization have generally been viewed as examples of human irrationality.
In this dissertation I study the phenomena of epistemic convergence, polarization and factionalization for ideally rational agents, with multiple, probabilistically related beliefs. I demonstrate that rational belief polarization arises very naturally for such agents, even when they update on identical evidence. I demonstrate that probabilistic relations between beliefs can drive various kinds of belief convergence, polarization, and factionalization. Importantly, polarization and factionalization arise generically, without needing specific initial conditions. Under certain circumstances, polarization can even be rationally anticipated in advance. Furthermore, I show that a population of rational agents, should always expect their beliefs to either converge or factionalize under certain conditions.
