Re: reference class of SSA

From: Gale <wmgale.domain.name.hidden>
Date: Fri, 28 May 1999 17:12:12 -0400

Nick Bostrom wrote:

> Gale wrote:
>
> >The example given is commonly called the "Doomsday Argument" and is
> >an incomplete application of SSA. The principle needs to be applied
> >not only to birth order conditional on some particular number of
> >humans ever living, but also to the number of humans ever living.
> >When SSA is fully applied, there is no longer an implication of some
> >sort of population crash in the near future.
>
> What Gale is referring to is the Self-Indication Assumption, stating:
>
> (SIA) The fact that one is an observer gives one some reason to
> believe that the world contains many observers.
>
> It is true that this cancels the DA, but the SIA is very problematic.
> There is a discussion of the SIA in my "Investigations into the DA"
> (about half way down in the document).

I think there is a basic flaw in Bostrom’s presentation of the argument
given by Kopf, Krtous, and Page. Bostrom wrote ("Investigations
into the DA"):

>The motivation for defining this axiom is that if we accept it then that
>buys us a very neat "solution" to the DA problem, i.e. a way of
>giving all the arguments advanced by Carter and Leslie their due
>without changing our minds one bit about the future of our species
>and its robotic descendants. The idea is that we could grant that a
>shift must be made ... while insisting that this
>shift is counterbalanced by the greater a priori likelihood of there
>being many observers.

The error here is the same error made by many who try to argue
against the DA without realizing that the DA argues for a *shift*
in priors which are otherwise given and not for an a priori likelihood
of any particular form. Likewise, as I read KKP, they argue that
in total parallelism with the DA, *another* shift should be made to any
prior probabilities that the DA is applied to, and they do not argue
that there is a "greater a priori likelihood of there being many
observers."

In particular, the SI *Argument* is no more an Assumption (or Axiom
as the paper has it) than is the Doomsday *Argument*. They are both
just applications of Bayes Theorem for which one must specify a prior
likelihood.

In other words, KKP are arguing that IF the DA shift is made THEN
another shift (the SIA shift??) should also be made. I do not see that
the arguments presented by Bostrom about the SIA address this cleanly.

Actually, Bostrom may not differ greatly from this view since he also
wrote:
>but it deserves serious
>consideration as it might be the only consistent way of resisting the
>conclusion of the DA

The two shifts are indeed "consistent" and, so far as I can see,
the only way of resisting the conclusion of the DA (or Leslie’s
conclusion anyhow...).

Bostrom also wrote:
>Moreover, if we
>reject an inference of the DA type, as many of us are inclined to do,
>then we must also abstain from applying the SIA: we should either
>apply both considerations or none.
>...
>There is nothing deep or very subtle
>about this. Either you cross the street or you stay on the pavement,
>and in both cases you will be safe; but if you stop midway you’ll get
>run over by a bus.

This appears to be quite consistent with the argument presented in KKP.

Thus if Leslie’s *conclusion* is to be accepted, then the DArg shift
must be made
while the SIArg shift is not made. And this is just what Bostrom has
called stopping midway, and which he would, from the previous quote,
seem to reject. But he has also written in bold letters "The Doomsday
Argument is Alive and Kicking", and he stimulated this post by
commenting
that the SIAssumption was problematic, so I would not hazard to guess
from
his writings how he might stand on Leslie’s conclusion.

Gale
Received on Fri May 28 1999 - 14:17:06 PDT