Title: General Bayesianism
Speaker: Dr. Xueyin Zhang (NYU and UC Berkeley)
Time: 20:00 ~ 23:00 (Oct.19th)
Zoom Meeting Link: https://us02web.zoom.us/j/86356536364?pwd=SEo4UEhibjJ3eXVWNWZldU9mZ0h3UT09
Meeting ID: 863 5653 6364 Passcode: 179052
Abstract: According to orthodox Bayesianism, rational agents should update by conditionalizing on their total evidence. One way of justifying Bayesianism is by appealing to instrumentalist considerations: Bayesian conditionalization is rational because it is optimal for promoting certain kinds of value (e.g. avoiding sure monetary loss or maximizing expected accuracy). However, recent epistemologists have pointed out that these arguments do not go through given evidence externalism, the view that it is possible for rational agents to lack evidence about what their evidence is. In other words, we have an inconsistent triad: (i) Bayesianism, (ii) instrumentalism and (iii) evidence externalism.
This talk has two parts. The first part is diagnostic: We argue that, given evidence externalism, the tension between Bayesianism and instrumentalism arises from a deeper tension between two norms of updating plans: self-stability and localism. The former says that rational agents should not change their plan given information about what they plan to do; the latter says that what a rational agent should plan to do should not depend on their counterfactual evidence -- information that they do not have but could have gotten.
The second part of the talk presents a positive proposal. We propose a new norm, general Bayesian conditionalization, and show that it is the most optimal amongst all updating rules that satisfy localism. We also prove that both Jeffrey conditionalization and Adams's conditionalization can be viewed as special instances of general Bayesian conditionalization. The talk concludes with a generalization of our results to the infinite context.
