Groups of people are disposed to divide into subgroups that polarize on a variety of topics: individuals in the same subgroup tend to converge in opinions, while individuals in different subgroups tend to diverge in opinions. This widely-confirmed empirical tendency is standardly taken to be a hallmark of human irrationality. It need not be. I’ll first show that rational, predictable polarization is possible: whenever you face ambiguous evidence—evidence that you should be unsure how to react to—predictable polarization can be fully epistemically rational. This claim can be proven in a general Bayesian framework, as well as illustrated with a simple demonstration. I’ll then argue, further, that this abstract possibility may play a role in the actual polarization we observe. One core contributor to predictable polarization is confirmation bias: roughly, the tendency for people to seek and interpret evidence in a way that is partial to their prior beliefs. And I’ll argue that—given common structures of evidential ambiguity—rational agents who care only about the truth should sometimes exhibit confirmation bias.
After the talks all are invited to socialise and continue discussion over drinks in the Ryle Room. If you would like to join for dinner after the drinks reception, please email daniel.kodsi@balliol.ox.ac.uk.
