Re: Status of Superrationality

From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Thu May 29 2003 - 14:05:38 MDT

  • Next message: Technotranscendence: "Re: test"

    Wei Dai wrote:
    > On Wed, May 28, 2003 at 07:34:02PM -0400, Eliezer S. Yudkowsky wrote:
    >
    >>The main reason for being an averagist is that you live in an infinite
    >>universe and hence cannot be a totalist.
    >
    > If the universe is infinite, you can't be a naive averagist either,
    > because otherwise you'd be trying to maximize something that is infinity
    > divided by infinity. Obviously both positions need to be more
    > sophisticated in an infinite universe. For example you can have a measure
    > over all objects and try to maximize a weighted average happiness,
    > weighted by measure. Then the difference between the averagist and the
    > totalist would be that the former normalizes the measure so that the sum
    > of measures of all living (or otherwise qualified) beings sums to one,
    > whereas the latter normalizes the measure so that the sum of measures of
    > all objects sums to one.

    If you're a threadist, then the point of "measure" is that it determines
    the weighting of subjective conditional probabilities in the thread of an
    observer. I.e., if observer-instant A has four times as much measure as
    observer-instant B, then observer-instant X, which lies in the past of
    both A and B, is four times as likely to lead to A than B. Similarly, the
    greater the measure of an observer-instant, the more futures it lies in,
    and hence the greater the importance of making its future happier.

    >>If so, then observer-instants
    >>are not things that either exist or do not exist; the question is where
    >>they predominantly exist. What you want to do is arrange for as many
    >>observer-instants as possible to lead to the happiest possible future
    >>instants with the strongest possible probabilities.
    >
    > How is this different from the totalist position? Can you give a concrete
    > scenario where your brand of altruism would lead to a different course of
    > action from totalism?

    It's better to "move" than to "copy", because the left-behind branch isn't
    as happy - you want the past observer-instant T0 to have all of its future
    in T1-A, rather than splitting to T1-A and T1-B.

    And it differs from the averagist position in that Hal Finney's reductio
    ad murderum does not apply.

    -- 
    Eliezer S. Yudkowsky                          http://singinst.org/
    Research Fellow, Singularity Institute for Artificial Intelligence
    


    This archive was generated by hypermail 2.1.5 : Thu May 29 2003 - 14:16:56 MDT