Re: Doomsday Example

Nick Bostrom (
Tue, 25 Aug 1998 18:50:10 +0000

Robin Hanson writes:

> Nick Bostrom writes:
> >> Doomsday argument folks also seem to want talk about the possibility
> >> that I might have been some other human at some other place in
> >> space-time.
> >
> >That is only shorthand. You can replace this phraseology if
> >you find it problematic. Instead you can talk about the possibility
> >that Robin Hanson, while remaining Robin Hanson, could forget certain
> >facts, and about what probabilities RH would then assign to the
> >hypothesis that RH is living in such-and-such a place and time.
> >That's actually quite simple.
> It's not at all simple for me.

Well, I shouldn't have said simple. After all, it leads to the problem of the reference class.

> I could sortof accept the idea that I
> might have been someone else instead. But I find it very hard to believe
> that you could make Robin hanson forget not so much that he was no longer
> Robin Hanson, *and* enough so he couldn't tell he wasn't a Martian living on
> Hermes in 2200.

Let me put it like this. The amnesia heuristic sets a lower bound for what should be included in the reference class.

For example, it clearly makes sense to say that you might not have known (indeed you may not know) the exact hour of your birth. If you didn't know that, then you would use some probability distribution over possible birth hours compatible with what else you know. If you conditionalize this distribution on your exact birth hour, you should get back the distribution you held before you forgot the birth hour. If you don't get back the original distribution then that indicates that you had forgot to take account of some effect. The doomsday argument claims that there is such an effect that you have neglected to take into account.

When we're talking about forgetting about whether you are a Martian living on Hermes in 2200, then things get problematic. The amnesia heuristic seems to gradually break down. I agree with that. That's why there is the reference class problem.

> I'm not sure the SIA is sufficient to deal with all the cases of
> interest. I'd rather instead accept an approach to defining states
> and priors.

It would be nice if it were that simple, but we're still waiting for a coherently specified state space. Do you have any specific cases in mind where the SIA would be inadequate?

Nick Bostrom
Department of Philosophy, Logic and Scientific Method London School of Economics