• CanadaPlus@lemmy.sdf.org
    link
    fedilink
    arrow-up
    3
    ·
    edit-2
    5 months ago

    I’m worried about this one, especially from an AI safety perspective as LLMs become capable of preforming simple white-collar jobs, like those of managers and investors.

    Right now a rogue AI would have trouble getting going because human contact is expected in most important business transactions. However, it’s easy to imagine a world where most people are employed by opaque apps, which are run through proprietary servers. Then, all it would take is for some server on Wall Street to calculate that it could make more money if it does buybacks until it has a majority stake in itself, and contract out whatever it needs in meatspace to apps.

    I know, I know, it sounds like sci-fi, but it always does at first.

    • andrew_bidlaw
      link
      fedilink
      arrow-up
      3
      ·
      5 months ago

      My favorite episode of X-Files predicted it at a lesser scale: https://en.m.wikipedia.org/wiki/Kill_Switch_(The_X-Files)

      Having an internet connection, a proper AI can easily order contractors around and reproduce, secure and empower itself.

      Godbless the hype is about stupid t9 on steroids, but we don’t really have any safeguards against what we assume is a proper autonomous AI.

      • CanadaPlus@lemmy.sdf.org
        link
        fedilink
        arrow-up
        3
        ·
        edit-2
        5 months ago

        Having an internet connection, a proper AI can easily order contractors around and reproduce, secure and empower itself.

        I mean, that’s the standard idea guys like Yudkowsky talk about. Having poked around a bit, it seems a couple decades of petty hackers have made that pretty impossible to do without either leaving a meatspace paper trail, or having meatspace human accomplices. Conquering the world instantly by Wifi, unless you can break encryption, is probably overblown - for now.

        Otherwise, I just agree.