Bayesian Reflection Principles and Ignorance of the Future

2SarahSrinivasan

1crickets

2SarahSrinivasan

New Answer

New Comment

1 Answers sorted by

To what future self should my 2024 self defer, then? The one with

E,E*, orE**?

To each with your current probability that that will be your future self. Take an expectation.

which is likeliest [...] defer to the likeliest

Any time you find yourself taking a point estimate and then doing further calculations with it, rather than multiplying out over all the possibilities, ask whether you should be doing the latter.

cr2024 = P2024(E) * 0.5 + P2024(E*) * 0.3 + P2024(E**) * 0.7

Hi! Thank you for the answer. Your answer is a little bit cryptic, though. Could you maybe provide some more detail or elaborate a bit on what you're saying?

Are you in essence saying something like "take the weighted average of the three possibilities and my probabilities that they'll occur" and use this to determine my expected future credence?

2

Yes, exactly that.

Hi everyone,

New user here. I am learning Bayesian epistemology by working through Mike Titelbaum’s recent

Fundamentals of Bayesian Epistemology.I’ve been learning about deference principles — principles concerning deferring to experts, where experts may include experts in the colloquial sense, the objective chances, and even one’s future self (provided one is sure that one’s opinions will update via conditionalization).I have a question about the Reflection Principle. Roughly, Reflection tells us to adopt the credence now that we expect we would adopt at some future time (on the assumption that our opinions are updated by conditionalizing). More carefully, Reflection says:

Obviously, I do not have perfect knowledge of what my future evidence will be, however, so I cannot be sure what my future credences will be,

de re.(Of course, I could know themde dicto, as in: “my future credences will be whatever credences result from conditionalizing on the evidence I acquire.”)Suppose that currently my evidence is

Eand cr_{2024}(p) = 0.5. In 2050, I might have evidenceE(i.e. my evidence won't have changed),E*, orE**. If In 2050, I have evidenceE, then cr_{2050}(p) = 0.5. But suppose that I instead 2050, I have evidenceE*, and by conditionalizing onE*I have cr_{2050}(p) = 0.3. Or suppose that in 2050, I have evidenceE**, and by conditionalizing onE**I have cr_{2050}(p) = 0.7. (Suppose these are the only three possibilities, just for the sake of ease.)I don't in 2024 know what evidence I'll have as to whether

pin 2050. To what future self should my 2024 self defer, then? The one withE,E*, orE**? Is there a way to adjudicate this? The intuitive answer is that I should figure out which is likeliest: (i) that my future self hasE, (ii) that my future self hasE*, or (iii) that my future self hasE**. I can then defer to the likeliest future self. But how can I determine this?I apologize if this is a n00b question. Titelbaum does not discuss this issue in the book, and I'm not sure where to look.

Thanks.