A Bayesian response to the irrelevancy challenge

Abstract

Many of our beliefs are caused by something irrelevant to their truth. For example, an agent holds her religious or political beliefs because of her upbringing, but where she grew up has nothing to do with whether her religious or political beliefs are true. In such cases, an epistemic adversary can say: ”You only believe that because...” Call it the Irrelevancy Challenge. This thesis explores how the agent should rationally respond to the Irrelevancy Challenge.

After formulating the challenge in an argumentative form, this thesis reviews the existing responses that respond to the challenge by evaluating whether the challenge is a valid argument. Then it points out that each response has its problem and thus motivates an alternative way for the agent in question to respond. That is, to respond by updating her beliefs on new evidence provided by the challenge. The Bayesian framework stands out to be a promising model of how a rational agent updates her belief on new evidence. This thesis then explores how a Bayesian agent should rationally respond to the Irrelevancy Challenge.

As soon as an agent takes the Irrelevancy Challenge as a piece of new evidence and processes it with the standard Bayesian machinery, especially the principle of conditionalization, she will find that this evidence is two-folded. On the one hand, the Irrelevancy Challenge provides mere factual information that can be processed simply by conditionalization. On the other hand, the mere factual information sometimes implies that the agent should revise her initial priors or refine her algebra, which the standard Bayesian model cannot accommodate. This thesis respectively addresses these two aspects of the evidential import from the Irrelevancy Challenge. It concludes that a Bayesian response to the Irrelevancy Challenge should include two parts: (1) the agent updates her beliefs by conditionalizing on the mere factual information about how her beliefs are caused, and (2) a modified Bayesian model allows the agent to update her beliefs by processing the implications by procedures other than conditionalization.