From: Wei Dai (weidai@weidai.com)
Date: Tue Jun 03 2003 - 17:16:53 MDT
On Tue, Jun 03, 2003 at 04:29:54PM -0400, Eliezer S. Yudkowsky wrote:
> Each time you recalibrate the pointer state from scratch, you should
> arrive at an answer consistent with applying Bayes' Theorem to past actual
> pointer states
How exactly do you "recalibrate the pointer state from scratch"?
Here is what I think you mean. There may be a way to compute subjective
probabilities without reference to Bayes's rule, but just looking at
relative frequencies (or measures) of mind states in possible worlds,
setting the probability of a possible world being actual higher if your
current mind state has a higher measure in that possible world.
However there is no general agreement on exactly how one should do that.
The way I look at the debate on DA and SIA is that it's really an argument
about which method to "recalibrate the pointer state from scratch" is the
correct one. People who reject the SIA would normalize the measure so
that in each possible world, the measure of all sentient creatures (or
whatever the reference class is) sums to 1. People who accept the SIA
would normalize the measure so that in each possible world,
the measure of all objects sums to 1 (or equivalently so that the measure
of all sentient creatures sums to the fraction of the world that is
sentient).
Strangely, this argument looks a lot like the one between averagists
(people who want to maximize average happiness of everyone) and
totalists (people who want to maximize total happiness of everyone).
(Refer to thread on "superrationality" if you don't understand what I'm
talking about.) Notice that the problems with infinity in both the
totalist position and in SIA disappear when you phrase them in terms of a
measure on mind states.
This archive was generated by hypermail 2.1.5 : Tue Jun 03 2003 - 17:28:25 MDT