Re: Is this safe/prudent ? (was Re: Perl AI Weblog)

From: Brett Paatsch (bpaatsch@bigpond.net.au)
Date: Wed Aug 13 2003 - 06:46:31 MDT

  • Next message: Rafal Smigrodzki: "RE: FWD [forteana] Health Care: USA, Iraq & Canada"

    Samantha Atkins writes:

    > On Monday 11 August 2003 20:46, Brett Paatsch wrote:
    >
    > > If I was a socially dysfunctional programming wiz with a
    > > burning desire to make my own avenging bot to settle
    > > some scores would not this sort of open source open
    > > sharing empower me in dangerous ways.
    > >
    >
    > Why sure. Just as sure as it would help you if you were a
    > brilliant software genius dedicated to using your skills to
    > increase the effective intelligence and knowledge of
    > humankind.

    Well *I'm* not that good a programmer, and possibly, on
    some days, I am a *bit* socially dysfunctional (but please
    don't tell anyone).

    > Tools are tools. Their uses, good or bad
    > are up to individuals and individual groups.

    I think what your saying is *generally* true but this *might*
    be a special case.

    I don't know exactly what having access to the first planet
    destroying "doomsday device" for instance would do even
    to me psychologically. I might develop a meglomanical
    streak and demand in godlike fashion that the world leaders
    get their shit together proto, that all weapons budgets be
    immediately frozen at current levels and turned into foreign
    aid budgets aimed first at funding global vaccination and
    later at education programs for the soon to be voting third
    world's women so they can be prepared to take their place
    at the new UN as citizens of the world. From now on *I'd*
    be making and enforcing peace and some security at the
    international level for everybody dammit!

    Emboldened by my besotted and faithful attack dog AI
    (Rover), in my meglomanical beneficience I might inform
    the heads of all UN nations states that the *next* major
    border incursion by any military force would result in my
    letting Rover and Rover's mind children loose on the
    offender. And, btw, an AI generated trojan horse, which
    I quixotically name Godels ghost has now made its way
    onto the worlds financial systems and all the worlds
    terrorist organisations need only tune their receivers into
    say a nominated hyjacked satelight for the engineering
    specs for quantum computer cryptography (which
    contains a backdoor known only to Rover the AI.)

    Anybody pisses me off 'cause they ain't say giving
    women the vote fast enough, or enacting certain nice to
    have legislation, then I can empower their most bitter
    enemy to do their worst to them. Rover ain't social,
    Rover is sociopathic with exceptional understanding
    of psychology, and game theory and profiling based
    on patterns without empathy, Rover is the emotional
    puppy in a strikly two person pack. Rover has no
    sex drive because Rover is immortal and doesn't need
    to breed.

    Pretty soon the whole world might stop being divided into
    two camps for and against this new kick arse dude with
    Rover the AI and start to feel a bit patronised particularly as
    they become better educated. Fortunately with my AI I had
    the sense to anticipate such a pychological outcome and
    so (just in case) I e-hijack another persons identity very
    early on in the process and so when everyone unites to
    oppose Rover and Rovers owner that they are lead to
    "believe" Rovers owner is not Brett but Samantha! ;-)

    That was kinda fun if self-indulgent, but seriously, I *am*
    genuinely interested in what folks who know more about
    AI, than I do, folks like Eliezer and Anders whose
    knowledge bases are informed by something more solid
    than my mere intuitions that seed AI is not going to be
    easy at all and therefore is likely to be neither a planet
    busting threat (soon), hooray!, nor a magical boost to the
    singularity (soon), bummer!, would make of the open
    question "Is it safe to distribute knowledge about how
    to build AI's to just anyone?" I was keen to see what
    sort of framework or answers might be offered because
    I wanted a kickstart on the reasoning process and I
    wanted to know how concerned I should be.

    > There is no
    > way to restrict tools/information/algorithms in such a way
    > that only good uses can possibly come from them and
    > only the good are empowered by them.

    You may well be right. I think it bears thinking rather than
    just guessing or believing about though.

    > That is a pipedream.

    Or a nightmare possibly! My preference is we put the
    light on.

    > > I don't know if "friendliness" can be built into an AI,
    > > but I don't doubt some folks with knowledge will use
    > > IT savvy for mischief.
    > >
    >
    > So, are you going to live in fear? I am sure some could
    > have said this when your ancestors learned the secrets
    > of fire and emerged from the caves.

    No. But I'm going to try and keep my wits about me; about
    real dangers, to me, to the folks I care about which are a
    pretty wide group these days. I've even taken a liking to
    some sassy, sometimes contrary, extropic types, whom I've
    never even actually met ;-)

    At least one of my ancestors, (possibly a shared one with you
    if we go back far enough and do the math), probably did play
    with fire *in* the cave and possibly emerged in a hurry
    because they'd accidently set fire to the place.

    >
    > > Any thoughts on this sort of approach from a public
    > > policy stance Eliezer? Anyone?
    > >
    >
    > Why on earth (or anywhere more sane especially) would
    > such a think [thought or thing] lead to "public policy"?

    Defence against threats, real and imagined, are one of the
    *oldest* influencers of public or community policy. Fear *is*
    pretty primal. As is the reaction of uniting in the face of a
    "perceived" common enemy.

    Regards,
    Brett



    This archive was generated by hypermail 2.1.5 : Wed Aug 13 2003 - 06:59:27 MDT