• pearsaltchocolatebar
    link
    fedilink
    English
    arrow-up
    26
    arrow-down
    2
    ·
    2 days ago

    Hasn’t AI already been shown to be better at catching things like cancer than humans?

    There are some things that computers can be better at than humans.

    • GiveMemes@jlai.lu
      link
      fedilink
      English
      arrow-up
      2
      ·
      13 hours ago

      Yes… well, sorta. For example, AI was found to be better at identifying TB than medical doctors. The catch here is that it also falsely diagnosed st a much higher rate than doctors. When an investigation was done as to how the AI was evaluating the imaging that it was given, they found that sets of virtually indistinguishable images were given different diagnoses by the AI. In many cases where there were no visible indicators of TB, a positive diagnosis wss given. The reason for this is that the AI was not weighting their TB diagnosis based on markers that doctors would look for alone, but also the age of the machine. Older machines have a much greater chance of being located in developing countries where TB is both more common and more deadly, leading to the age of the machine being considered an important factor, whereas a human would know that the age of a machine has absolutely zero relationship with the chance of getting TB, and doctors in these areas are already aware of and watching out for TB as it’s a much more serious illness than in Germany, for example.

      Idk much about the cancer thing, but basically the machine learning for diagnosis thing is iffy at best afaik.

    • 9488fcea02a9@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      21
      arrow-down
      1
      ·
      edit-2
      1 day ago

      Machine learning for helping a radiologist analyze images is super helpful and a mature field.

      Whatever “AI” LLM nonsense tech bros are trying to add in to everything in the last 2 years is probably not all that helpful, but i could be proven wrong

    • whithom
      link
      fedilink
      English
      arrow-up
      27
      arrow-down
      1
      ·
      2 days ago

      Yes! And we should use it when it has been proven effective. But the AI shouldn’t be able to administer drugs.

      • pearsaltchocolatebar
        link
        fedilink
        English
        arrow-up
        15
        ·
        1 day ago

        For sure. There always needs to be a human in the loop. But this notion people seem to have that all AI is completely worthless just isn’t true.

        What’s scary is the hospital administration that will use AI to deny care to unprofitable patients (I’ve listened in on these conversations).

        • deranger@sh.itjust.works
          link
          fedilink
          English
          arrow-up
          9
          ·
          edit-2
          1 day ago

          Where’s anyone saying it’s worthless? That’s not in the article nor in these comments.

          The issue is how it’s being used. It’s not being used to detect cancer. It’s being used for “efficiency”, which means more patients being seen by fewer nurses. It’s furthering the goals of the business majors in hospital administration, not the nurses or doctors who are caring for the patient.

          • gravitas_deficiency@sh.itjust.works
            link
            fedilink
            English
            arrow-up
            2
            ·
            13 hours ago

            LLMs are largely worthless (in the context of improving human society).

            Neural Nets aimed at much more specific domains (recognizing and indicating metastases or other abnormalities in pathology slides for human review, for example) are EXTREMELY useful and worthwhile.

          • kryptonidas@lemmings.world
            link
            fedilink
            English
            arrow-up
            2
            ·
            1 day ago

            Ai nearly everywhere is to improve efficiency, less people become more productive so that the owners keep more money. Because a pay rise because of it is off the books. Since now you need to be “less skilled” anyway.

    • coolmojo@lemmy.world
      link
      fedilink
      English
      arrow-up
      7
      ·
      1 day ago

      Dogs are can be also better at detecting cancer than humans. And dogs tend to hallucinate less

      • conciselyverbose@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        7
        ·
        1 day ago

        Hallucinations aren’t a problem with the actually medically useful tools he’s talking about. Machine learning is being used to draw extra attention to abnormalities that humans may miss.

        It’s completely unrelated to LLM nonsense.

        • coolmojo@lemmy.world
          link
          fedilink
          English
          arrow-up
          2
          ·
          1 day ago

          Perhaps, we should consider not calling all of them as AI. Machine learning is a useful tool.

            • coolmojo@lemmy.world
              link
              fedilink
              English
              arrow-up
              3
              ·
              1 day ago

              You are right. My pet peeve is that it is now used as a marketing term without actual meeting. Used to be the word smart. Now instead of “buy this smart toaster”, “buy this AI powered toaster”. Sorry if this reply was too verbose for your liking.

      • rumba@lemmy.zip
        link
        fedilink
        English
        arrow-up
        5
        ·
        1 day ago

        They’re better at smelling cancer than humans.

        I’m not sure we can definitively say they hallucinate less.