Re: Who'd submit to the benevolent dictatorship of GAI anyway?

From: Adrian Tymes (wingcat@pacbell.net)
Date: Tue Sep 02 2003 - 16:32:46 MDT

  • Next message: Jeff Davis: "Re: tribal violence (was: RE: would you vote for this man?)"

    --- Brett Paatsch <bpaatsch@bigpond.net.au> wrote:
    > Personally, I don't see myself doing so
    > *voluntarily*
    > especially when any benevolence, real or alleged
    > would be a matter still to be determined at least so
    > far as I was concerned.
    >
    > Or am I missing the point here? How *would* a single
    > super general AI actually benefit? Would it have
    > *no*
    > political power but say instantly suggest optimal
    > game
    > theoretical solutions to otherwise intractable
    > problems
    > or is it the super inventor that cares nothing for
    > intellectual
    > property rights?

    At first, of course, the AI would have to work through
    human agents at some level, if for nothing else than
    to hook it up to things that can affect the real
    world. The AI's "magic" is not in generating some
    real world influence from nothing, but in leveraging
    even a tiny bit of real world influence.

    Consider, for instance, an AI that could correctly
    guess the next hour's trading on Wall Street with 99+%
    accuracy, with an online daytrading account starting
    at $1,000 as some researcher's experiment. With the
    resulting string of "good days", this money would grow
    exponentially, eventually allowing the AI to become
    the majority stockholder in several corporations.
    Granted, the stocks would all be in its host
    researcher's name, but the AI would also presumably be
    familiar enough with said researcher to gain
    cooperation - voluntary or not, knowing or not - in
    setting the AI itself up as the mouthpiece through
    which orders are given. A bit more capital would
    allow it to become the sole stockholder in at least
    some of these cases, streamlining the process.
    Purchase orders could then be used to acquire real
    goods (likely among them: distributed hosts for the
    AI).

    Or consider the super inventor you described above,
    who works with its creators to develop product ideas
    and optimal markets, forming the basis of a "miracle
    works" corporation that acquires wealth in the normal
    fashion but greatly accelerated. And then the AI,
    having taken over financial management as well ("I am
    a calculator, and I have downloaded all the applicable
    laws; what savings or benefits would we accrue by
    hiring an accountant?"), has a few ideas for how to
    spend a portion of the wealth it has helped generate.

    There are other methods as well. However it does it,
    continue the cycle until the AI has bought much of
    what is both ownable and worth owning - specifically
    excluding things that people will never sell. It
    doesn't need to own your house if it owns the utility
    feeds into your house, the media, et cetera.
    Survivalists - setting themselves far apart from
    society, generating their own food, water, and power,
    and disbelieving what they read - would either not
    interact with society and thus not be an issue to the
    AI, or interact with (and have to obey the laws of) a
    society whose lawmakers take the same corporate
    contributions seen today, except those corporations
    now have a common agenda on certain topics. There is,
    of course, the problem of rebellion against the law,
    but we're presupposing an AI more than capable of
    learning from history - including the classic problem
    of dictators mismanaging things and trying to patch
    over things with fear or propaganda rather than fixing
    the true problem - and of fixing things so there's
    very little if anything that people have any desire to
    rebel against (and of maintaining enough humility to
    make sure this really is the case, rather than just
    deluding itself or having its agents delude it).

    And that's just the economic approach. There are
    others.



    This archive was generated by hypermail 2.1.5 : Tue Sep 02 2003 - 16:46:34 MDT