Re: PLUTO, Our Future Home

From: Anders Sandberg (asa@nada.kth.se)
Date: Sat May 24 2003 - 17:24:07 MDT

  • Next message: Anders Sandberg: "Re: The mistake of agriculture (was: evolution and diet)"

    On Sat, May 24, 2003 at 01:46:23PM -0700, Robert J. Bradbury wrote:
    >
    > On Sat, 24 May 2003, Spike wrote:
    > > It would be nice of course, but uploading of a meat brain is
    > > a far more difficult problem than building an MBrain.
    ...
    > The process of "uploading" of a meat brain is at this time mostly
    > a mystery to us. In large part because I do not think we have a
    > robust understanding of things like motivations that are entirely
    > coded at the hormone level, self-awareness, self-direction, etc.
    >
    > But that does *not* imply that we might make very rapid progress
    > in this area over the next 5, 10 or 20 years (Anders might comment);
    > after all we *do* have the human genome in our hands at this time.

    Hormones certainly play a role but are comparatively simple. The deep
    philosophical questions are probably best probed by making partial or
    full uploads. The genome knowledge is likely pretty irrelevant,
    although it is helpful in figuring out the tricky biochemical pathways
    inside the neurons that play a part in function. The real issue will be
    whether we get a good understanding of 1) low-level neural information
    processing, especially in the synapses, 2) the basic cortical
    microcircuitry (lots of stuff happening here; I expect a breakthrough or
    something within the next 5-20 years) and especially 3) a good scanning
    method. We don't need to understand the mind, just emulate the brain
    (and bits of the body).

    I think I have a feeling of the uploading problem (famous last words),
    and it is hardly trivial. But it is a different kind of problem from
    building an MBrain. The first is deductive to a large extent, we need to
    understand a system well enough to emulate it robustly. The second is
    constructive, we need to understand physics and engineering well enough
    to construct a very complex system.

    In principle we could start building a bad MBrain today by sending out
    solar collectors and Beowulf clusters into orbit. On the other hand we
    cannot even upload a nematode yet. The MBrain could be extended
    gradually, building on top of simpler models. The uploading requires a
    number of key technologies and understandings; once we have the
    technology to do a reasonable simulation of a nematode from a scan, we
    can just scale it up to a brain.

    So my answer would be that the upload field will not appear to get
    anywhere from an outside perspective for a long time, and then
    relatively rapidly move to a full upload state. MBrains on the other
    hand will show a slow and steady progress, but take a long time to get
    to Robert's final design.

    > I think Anders begins to consider this issue with his
    > development of a "Chronos" neutronium brain -- but I
    > have significant problems with whether I/O from any
    > such entity is feasible. (Or whether the bandwidth
    > of any external I/O can deal with the amount of internal
    > information being produced.) [Side note to Anders
    > and others -- I'm not talking theories here -- I
    > want real no-magic-physics, yes-I-can-build-this-device
    > examples.]

    In one of my roleplaying scenarios people invented quantum super-AI and
    found it utterly useless, since the little ball of computronium had an
    infinity of superposed AIs attempting to communicate with the outside
    world - the end result was pure noise :-)

    Chronos would have a serious bandwith bottleneck, but an even worse lag
    both internally and externally. It could blast information across lots
    of channels, but an information production of 10^53 bits/s is heavy. The
    Bekenstein bound (eq 29 in my paper) produces a bandwidth bound of 1e35
    bit/(Js), so it would need 1e18 W just to send the new information. Of
    course, I assume quasar-like energy dissipation so this is puny (it is
    just 1e-8 of the sun's luminosity).

    I think the answer is to have hierarchical intelligence. Parts are fast
    and act as subsystems for larger, slower and deeper parts. While a
    subsystem acts as a "reflex arc" of (say) human intelligence, it
    participates in a larger system with maybe ten times slower reaction
    speed but a thousand times larger intelligence that sets overall goals,
    evaluates and distributes current mental tasks etc. This system in turn
    is part of an even larger and slower cluster, and so on. The top level
    may be millions of times slower than the local levels, but have an
    effective intelligence trillions of times the simplest parts.

    -- 
    -----------------------------------------------------------------------
    Anders Sandberg                                      Towards Ascension!
    asa@nada.kth.se                            http://www.nada.kth.se/~asa/
    GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y
    


    This archive was generated by hypermail 2.1.5 : Sat May 24 2003 - 17:30:25 MDT