Abstract
Belief polarization occurs when the beliefs of agents diverge upon updating on certain types of evidence. Recent research indicates that belief can arise even amongst rational agents \cite{Jern_Polarization, Kelly_2008, O_Connor_Polarization}. Although the specific mechanisms differ, I distinguish two general origins of belief polarization. First is agent network-driven polarization \cite{Axelrod1997, HegselmannKrause, Macy2003, Deffluant2006, BaldassarriBearman, O_Connor_Polarization}, which arises due to the relationships between agents. With this form of polarization, epistemic influence between agents is determined by factors as the similarity in prior beliefs. The second origin is belief-network driven polarization \cite{Jern_Polarization, Kelly_2008}, which arises due to the relations between different beliefs held by agents. I argue that a formalism involving epistemic networks of agents, each with Bayesian belief networks allows us to represent both kinds of polarization in a unifying framework. I set out certain conditions under which each type of polarization can arise, in terms of the structure and relationships of the agents' epistemic and belief networks.
