RE: Suns considered harmful (was: Pluto)

From: Robert J. Bradbury (bradbury@aeiveos.com)
Date: Sun May 25 2003 - 02:35:22 MDT

  • Next message: Eliezer S. Yudkowsky: "Re: Suns considered harmful (was: Pluto)"

    On Sat, 24 May 2003, Spike wrote:

    > Robert, Eli, I recall the conversation, down to where I was sitting
    > and who else was in the room.

    Spike, thanks for confirming this. I'm going to have to see if
    my rather slow memory neurons can recall more of this. (See
    my comments to Eli's msg. regarding what I can recall.)

    > ... when I argued that we might have a long period (millions of years
    > perhaps) of sun-burning after we build a first-generation MBrain
    > and before we manage to actually stop the nuclear fusion in the sun.

    It is still an open question as to whether or not one
    optimal architecture is not based on a "stellar" approach.
    Gravity confines matter and drives fusion. I will agree
    that controlled fusion is a much better process for
    "managed" (extropic) energy production. But I haven't
    seen any arguments presented that the best architectures
    for the allocation of matter for controlled fusion energy
    production would not require the sacrifice of such matter
    from the actual computronium. I.e. You can use your
    energy sources more efficiently but you have to sacrifice
    CPU cycles to do so.

    From Eli's perspective (perhaps) -- to maximize the intelligence
    of the AI it may very well be necessary to waste energy.

    In your terms (perhaps) -- to minimize wasting energy it
    may require producing a computronium matrix with significantly
    less capacity to host AIs or uploads.

    > Eliezer's notion has considerable power: any logical growing
    > AI will naturally calculate its theoretical upper bounds,
    > then scramble like hell to achieve that level.

    This statement I agree with. But from a computer science
    and biological standpoint I would argue that it *cannot*
    know what the "upper bounds" are. The phase space of what
    "intelligence" can explore is simply too large. All you
    can do is explore some corner of it. So the best one can
    do is optimize the architecture of energy production and
    consumption for a specific variant of "intelligence".
    The best architecture to do voice recognition may not
    be the best architecture to do face recognition and
    they may have *very* different energy consumption
    requirements.

    > This would mean stopping the absurd waste of energy
    > pouring out into cold dead space from the stars.

    This is true -- but whether it means doing star-lifting
    or some other creative approach to managing star downsizing
    and therefore slowing stellar evolution is open to debate.

    It is an important question as to how many cycles can
    I get in what amount of "real" time and what is the most
    efficient way to achieve that?

    I'll completely agree with the idea that wasting energy
    radiating it into space is probably not extropic (which
    is the point behind Dyson shells and MBrains). *But*
    that does not immediately translate into the idea that
    one must dismantle stars or find some other process
    by which to perform explicitly controlled energy production
    using the resources they contain.

    The devil is in the details here and I'll need to see
    those details before I'm convinced.

    Robert



    This archive was generated by hypermail 2.1.5 : Sun May 25 2003 - 02:45:27 MDT