Ugo Bolletta at the RITM Economics Seminar

Abstract: Post-truth pertains to the interaction between subjective beliefs and factual information. We study this interaction through a behavioral application of Bayesian technology. Agents do not learn but wish to validate their subjective beliefs around a topic of interest. When exposed to data that conflicts with their subjective prior beliefs, agents experience cognitive dissonance that can be reduced by providing (intrinsically biased) information which validates their subjective beliefs. In a multi-agent strategic setting we show that there exists a positive mass of players with moderate ideology who will never acquire information, and that there is an increasing benefit in acquiring information the more a player has an extreme ideology. Hence, the ex-post distribution of information exhibits over-representation of extreme views. Said phenomenon is particularly pronounced when the cost of information acquisition is low, which arguably mimics debating in online social media. Notably, our results are built around individual non-cooperative incentives, suggesting that the current patterns of observed polarization may not be merely a consequence of the online platform architectures and social influence.