Скачать или смотреть A Unified Theory of Rational and Biased Beliefs - Andrew Little (PIBBSS Speaker Series) 2023

The second speaker for the 2022/2023 PIBBSS Speaker Series. This was recorded in February 2023.

Biography

Dr. Little is Associate Professor of Political Science at UC Berkeley. His research uses formal models to explore the causes and political consequences of biased beliefs. This work provides some ways to model motivated reasoning (wanting to believe certain things are true) and selection neglect (not understanding the biases in observed information). Both in this line of work and more generally he studies topics like authoritarian politics, communication, information manipulation, and conflict.

Abstract
There are two widely used formalizations of what it means to have rational beliefs. The dominant approach in the social sciences centers around updating in response to new information using Bayes' rule. An alternative approach originating from statistical mechanics focuses on maximizing the entropy of beliefs subject to the information available. Several scholars from disparate disciplines have noted a way to capture Bayesian updating using the entropy-based approach, though this idea has received less attention than it deserves. Andrew discusses how treating rational beliefs as the solution to an entropy-based maximization problem also provides a helpful way to think about incorrect beliefs as solving the "wrong" maximization problem. This can capture many widely documented biases such as motivated reasoning, partition dependence, anchoring, overconfidence, confirmation bias, base-rate neglect, and conservatism. I'll go into some detail about the application to motivated reasoning, which highlights how hard it is to detect this bias in lab settings, let alone real world situations. Fortunately, the formalization also provides a path forward to detecting motivated reasoning.