Re: Status of Superrationality

From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Wed May 28 2003 - 17:34:02 MDT

  • Next message: Emlyn O'regan: "RE: my analysis of the Matrix Reloaded [SPOILERS!!!!!!!!!!!!!!!!! !!!!]"

    Hal Finney wrote:
    >
    > A couple of comments on this. First, I think it was Wei himself who
    > pointed out to me a few years ago that the "averagist" philosophy is
    > internally inconsistent. This is the view that we should maximize
    > the average happiness of human beings, i.e. the sum of all happiness
    > divided by the number of people. The problem is that half the people
    > are of below average happiness, hence they are dragging down the curve.
    > By killing off the less happy half of the human race, we can greatly
    > increase average happiness, hence this is an acceptable goal for the
    > averagist. However it does not stop there, for having done so, the
    > remaining half can once again be divided in to the more and less happy,
    > and we can justify killing off the less happy half of the remainder.
    > This is repeated until only one person is left, the single happiest
    > individual of the entire human race, who probably has something wrong
    > with him.

    The main reason for being an averagist is that you live in an infinite
    universe and hence cannot be a totalist. 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. To have any influence
    on this, you need to instantiate people. Killing people just makes death
    the most likely future of their past observer-instants, which is unhelpful
    (it makes the universe a sadder place). You didn't actually make any
    observer instants stop existing; you just established that a bunch of past
    observer instants would have subjective conditional probabilities that
    lead very strongly to death. 10^10^something meters away would be a
    person just like the one you killed, a very rare survivor, somewhere you
    couldn't reach. To help an observer-instant, to thread it into a happy
    future, you must first instantiate it. Once a person exists, killing them
    just gives the observer-instants in their past history a sad future.

    This is different enough from the "averagist" viewpoint that it deserves a
    different name. Call it the "threaded" viewpoint. You want
    observer-instants to have happy futures. You can't make a given
    observer-instant "exist" or "not exist"; *any* observer-instant will exist
    *somewhere*. But once a person lives *here*, you can try and keep the
    thread of their future from falling into a Death instant and terminating;
    you can try and make as many observer-instants as possible have strong
    conditional probabilities of leading to happy future observer-instants.

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


    This archive was generated by hypermail 2.1.5 : Wed May 28 2003 - 17:46:59 MDT