Nick B. writes:
>> let me explain what I mean by formalize.
>Ok, I'll make a stab at it. ...
>
>Suppose there are two possible worlds, W1 and W2. In W1 there are two
>consequtive states, S1 and S2; in W2 there is only one state, S1,
>whereafter the world ends.
>There is one agent, a, in each of these possible worlds. a can be in
>state a(S1) or in a(S2).
>The "possible states" for the a's states are: P(S1)={(S1, W1), (S1,
>W2)} and P(S2)={(S2, W1)}.
>We can assume that the prior probabilities are equal for W1 and W2.
>Suppose RH-now is a being in state S1. In S1 you have information I.
>Question: What is Prob(W1|I)?
>Some would say 1/2. But the doomsdayer claim that that is to overlook
>the fact that you should conside RHnow as a random sample from all
>actual observer-moments. The doomsdayer defends this claim ...
Once you've defined states, information partitions, and a prior, you're done; there's no more place for argument. So this must be an argument for a certain prior. And I don't see why one should take "I'm a random sample from the true reference class" as a fundamental axiom in choosing a prior. As you note, in your reply to Doug,
>if we try to apply the DA to other classes of objects we have to be
>careful. ... to make sure that we are not neglecting some underlying
>correlation.
I see no obvious reason why we shouldn't expect such correlations for whatever reference class you choose. The only reasonable way I can see to see if there are such correlations is to define what seems a neutral prior and then work on the math. Why should we work to make sure our prior ensures that we are a random sample from "the" reference class, especially when you aren't even sure what class that is?
My next post on this thread will give an example of a "neutral" prior calculation.
Robin Hanson
hanson@econ.berkeley.edu http://hanson.berkeley.edu/
RWJF Health Policy Scholar, Sch. of Public Health 510-643-1884
140 Warren Hall, UC Berkeley, CA 94720-7360 FAX: 510-643-2627