Re: Duplicates are Selves

From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Sun Apr 06 2003 - 23:37:38 MDT

  • Next message: randy: "Re: Help with a Minimum Wage Model"

    Lee Corbin wrote:
    >
    >>Perhaps you have not sufficiently internalized this branching
    >>aspect of selfhood if you cannot understand why I would want
    >>as much of myself as possible to go into the best available
    >>branch; a nonbranching transition to Disneyland is better than
    >>a branching transition to Disneyland, for the same reason that
    >>a million dollars cash is better than a quantum lottery ticket.
    >
    > Yes, even in those million-to-one duplication events I just
    > described, I'd prefer all my duplicates to go to the preferred
    > event. Yet if it's a choice between me *not* getting a duplicate
    > made, and having it made in an inconvenient location, I'll choose
    > the latter. The ratio of outcomes is not the right way to look
    > at this---what is correct is the total run time I get, and that's
    > why probability also is not appropriate.

    "Total run time" seems to break down very fast in the presence of either
    MWI or an open cosmos - your total run time is apparently increasing at an
    exponential rate of 10^150 branchings per second, within which a mere
    twofold duplication would scarcely be noticed, or else is simply infinite.
      In both cases the non-renormalized ethics of branching breaks down,
    while renormalized ethics continue with nary a blink; "it all adds up to
    normality".

    >>I *do* anticipate being in both places. That's *why* I want a move rather
    >>than a copy. I *want* be in the grocery and I anticipate the frustration
    >>of the Eliezer who finds himself still at home. When I snap my fingers I
    >>should just go where I want, dammit!
    >
    > But the *moving* is inferior in most copy scenarios because you get
    > less run time.

    "Less run time" seems a remarkably fragile concept. The *relative* amount
    of runtime *this version* of you gets determines the *relative* weighting
    of this present in your past self's future. You should care which future
    versions of you get the most runtime. But how can you tell whether your
    present has a "little" or a "lot" of runtime? As far as I can tell, you
    cannot. This leads me to propose that "runtime" is a relative quantity
    used in renormalization of futures, rather than an absolute quantity that
    applies to the present.

    (And then he gave a quiet, mysterious smile.)

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


    This archive was generated by hypermail 2.1.5 : Sun Apr 06 2003 - 23:46:57 MDT