Totally not a an AI asking this question.

  • Greyscale@lemmy.sdf.org
    link
    fedilink
    English
    arrow-up
    19
    ·
    1 year ago

    the problem with the current model for building AI is training it based on existing policy and thought. Which means it’d just be what we have now but somehow hallucinate more contradictory policy.

    • pjhenry1216@kbin.social
      link
      fedilink
      arrow-up
      8
      arrow-down
      2
      ·
      1 year ago

      There are other forms of machine learning that could be utilized. Some work more toward being given a set of circumstances to reach and then it just keeps trying to new things and as it gets closer, it just keeps building on those.

      • Greyscale@lemmy.sdf.org
        link
        fedilink
        English
        arrow-up
        4
        ·
        1 year ago

        That would require the humans controlling the experiment to both be willing to input altruistic goals AND accept the consequences that get us there.

        We can’t even surrender a drop of individualism and accept that trains are the way we should travel non-trivial distances.

        • pjhenry1216@kbin.social
          link
          fedilink
          arrow-up
          2
          ·
          1 year ago

          In a dictatorship with an AI being in control, I don’t think there’s a question of accepting consequences at they very least.

          There is no such thing as best case scenario objectively, so it’s always going to be a question of what goals the AI has, whether it’s given them or arrives at them on its own.

    • StijnVVL@lemmy.world
      link
      fedilink
      arrow-up
      1
      ·
      1 year ago

      That’s where it would start. I imagine it would be capable to see the flaws in the system and rectify them. This most probably means we as humans won’t come out on top however.

      A sentient ai would probably be the most dangerous thing to the human species as a whole.

      • Greyscale@lemmy.sdf.org
        link
        fedilink
        English
        arrow-up
        2
        ·
        1 year ago

        If the humans can’t see the flaws and correct them now, what do you think the AI would learn from the training data?

        • StijnVVL@lemmy.world
          link
          fedilink
          arrow-up
          1
          ·
          1 year ago

          First of all, a lot of humans do see the flaws but are indeed unable to correct them. This would also show in the training data. The AI OP is talking about would be much more powerful to actually act and change something.

          Don’t confuse Artificial Narrow Intelligence (ANI) with Artificial General Intelligence (AGI) or even Artificial Superintelligence (ASI). Your statement suggest you understand ANI, which is all the AI that we know today. However powerful they seem, they can only reproduce what they have learned from the training data.

          AGI (or human level AI) will be more what OP means here. Sentient, in a way that it can make its own decisions, think on a human level, feel on a human level and act on those feelings. If it feels humans are not important or harmful to what it values, it will decide to remove humanity as a whole. Give it the power to govern the world and it most certainly will act not in our favour.

          • Greyscale@lemmy.sdf.org
            link
            fedilink
            English
            arrow-up
            2
            ·
            1 year ago

            Until computers can be genuinely creative, and not emulate creativity, its not gonna happen. And when that happens, we’re either getting the startrek luxury space communism, or a boot smashing our head into the kerb for eternity. No middle ground.

            • howrar@lemmy.ca
              link
              fedilink
              arrow-up
              2
              ·
              1 year ago

              The entire premise of the OP is a hypothetical.

              In any case, there’s plenty of work on making agents that are “genuinely creative”. Might happen sooner than you think.