The incident in northern California marked the latest mishap blamed on the electric vehicle company’s Autopilot tech

  • scratchee@feddit.uk
    link
    fedilink
    English
    arrow-up
    2
    ·
    3 months ago

    They don’t have to be any good, they just have to be significantly better than humans. Right now they’re… probably about average, there’s plenty of drunk or stupid humans bringing the average down.

    It’s true that isn’t good enough, unlike humans, self driving cars are will be judged together, so people will focus on their dumbest antics, but once their average is significantly better than human average, that will start to overrule the individual examples.

    • rottingleaf@lemmy.world
      link
      fedilink
      English
      arrow-up
      1
      ·
      3 months ago

      Right now they are not that at all.

      When people say neural nets are unable to reason, they don’t mean something fuzzy-cloudy like normies do, which can be rebutted by some other fuzzy-cloudy stuff. They literally mean that neural nets are unable to reason. They are not capable of logic.

      • scratchee@feddit.uk
        link
        fedilink
        English
        arrow-up
        1
        ·
        3 months ago

        Reasoning is obviously useful, not convinced it’s required to be a good driver. In fact most driving decisions must be done rapidly, I doubt humans can be described as “reasoning” when we’re just reacting to events. Decisions that take long enough could be handed to a human (“should we rush for the ferry, or divert for the bridge?”). It’s only the middling bit between where we will maintain this big advantage (“that truck ahead is bouncing around, I don’t like how the load is secured so I’m going to back off”). that’s a big advantage, but how much of our time is spent with our minds fully focused and engaged anyway? Once we’re on autopilot, is there much reasoning going on?

        Not that I think this will be quick, I expect at least another couple of decades before self driving cars can even start to compete with us outside of specific curated situations. And once they do they’ll continue to fuck up royally whenever the situation is weird and outside their training, causing big news stories. The key question will be whether they can compete with humans on average by outperforming us in quick responses and in consistently not getting distracted/tired/drunk.