• FaceDeer@kbin.social
    link
    fedilink
    arrow-up
    4
    arrow-down
    1
    ·
    1 year ago

    This seems like something that could be searched for by an automated process. I bet lots of old papers have edits that the people who made them at the time figured would be “good enough” to not be spotted, but now we’ve got better automation.

    • JoBoOP
      link
      fedilink
      arrow-up
      3
      ·
      1 year ago

      That technology is still in development, as far as I know. Certainly, when Bik started there wasn’t any software that could do the job anywhere near as well as she could. I know she’s been testing out some more recent software iterations, and no doubt they’re learning from the masses of images she has detected. But it is not an easy problem to solve. Bits of images are not just duplicated (which might be ‘easy’ to detect if you could do enough comparisons between all possible areas of duplication), they are also rotated, stretched, squished… Automating that kind of pattern recognition would be amazingly hard.

      For some kinds of images, detecting the telltale signs of manipulation might be fruitful but that would likely require a whole new set of requirements for images submitted for publication.

      No doubt it will happen, and it does work for some kinds of manipulation. But I don’t think anyone is close to covering all the bases yet.

    • JoBoOP
      link
      fedilink
      arrow-up
      4
      ·
      1 year ago

      Yes, it was a good day when she appeared there.

      It’s been interesting to see who moved from science Twitter (and other Twitters). Some predictable movers, some predictable stayers, and some very disappointing stayers.