• Lojcs
    link
    fedilink
    English
    arrow-up
    139
    arrow-down
    1
    ·
    8 months ago

    It doesn’t help that all the humans have beauty filters on

    • Tar_Alcaran@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      32
      arrow-down
      1
      ·
      8 months ago

      Human male 40 absolutely does not have a filter, that’s probably why he rated as the most human looking human.

    • FishFace@lemmy.world
      link
      fedilink
      English
      arrow-up
      9
      arrow-down
      1
      ·
      8 months ago

      Not only that but several of them are a bit weird looking (sorry to those people…) as in, 37 and 47 have obvious asymmetries, 31 is a bit bug-eyed, 18 seems to have been taken with a super telephoto lens or have a really flat face.

    • gullible@kbin.social
      link
      fedilink
      arrow-up
      3
      arrow-down
      1
      ·
      8 months ago

      This reminds me of an argument I saw here last week about AI and its use as a grammar checker. You can definitely do it, but you’re going to have all the markers of using AI to cheat.

      • wischi@programming.dev
        link
        fedilink
        English
        arrow-up
        6
        ·
        8 months ago

        Not really, if you write the text first and only apply minor changes to fix the grammer (and not rewrite entire sentences) no AI detector will detect that because the sentence structure and pattern wouldn’t match typical AI output.

        • gullible@kbin.social
          link
          fedilink
          arrow-up
          4
          ·
          8 months ago

          Be sure to remember that, at best, AI takes prompts as interpretable guidelines and a request of “grammar checking” can involve some additional, unwarranted, restructuring. Points to whoever notices both AIisms that I noticed that chatgpt added to my grammar checked critique on grammar checking.

  • FaceDeer@kbin.social
    link
    fedilink
    arrow-up
    88
    arrow-down
    4
    ·
    edit-2
    8 months ago

    I feel like that “corporate wants you to find the differences between these two photos” meme. Isn’t everyone in those photos, in both the top and bottom rows, white?

    Edit: Ah, I see, OP has given this a highly misleading title. The “whiteness” of the faces is not actually particularly relevant. In another thread someone summarized what the article is actually about:

    For anyone who doesn’t want to read the paper, they basically took an 60 white men and 60 white women, and showed them a whole bunch of white faces, half of which were generated by AI. It turns out that AI faces were rated as more human-like than actual humans, and they had some hypothesis why. Principally that AI, by its nature, generates images close to “average”, while real people tend to have features that are not “average”. The reason the study focused on white people is that most AI have been trained on white faces, so AI tends to do better with white faces.

    • Two@lemmy.world
      link
      fedilink
      English
      arrow-up
      29
      arrow-down
      6
      ·
      8 months ago

      An op? Making a misleading title? On Lemmy?

      Man, it’s as if the severe lack of moderation and rules that so many people wanted when moving from Reddit is hurting the quality of posts on here.

        • Two@lemmy.world
          link
          fedilink
          English
          arrow-up
          4
          arrow-down
          5
          ·
          8 months ago

          You know there is a significant difference between it being a thing you might see occasionally versus an ongoing issue?

          • StorminNorman@lemmy.world
            link
            fedilink
            English
            arrow-up
            9
            arrow-down
            1
            ·
            8 months ago

            Yeah, and it was an ongoing issue with Reddit. Probably still is. Sure, small, niche subreddits that had active moderation were generally free of the scourge. But it was quite a common phenomenon across the site, in my experience.

          • silly goose meekah@lemmy.world
            link
            fedilink
            English
            arrow-up
            3
            arrow-down
            1
            ·
            8 months ago

            how is it only a thing you might see occasionally on reddit? I saw misleading titles daily when I still used that site.

    • SeaJ
      link
      fedilink
      English
      arrow-up
      1
      ·
      8 months ago

      It turns out that AI faces were rated as more human-like than actual humans

      I tried guessing from the ArsTechnica article and got a whopping 1 out of 8 correct.

      • Sanyanov@lemmy.world
        link
        fedilink
        English
        arrow-up
        5
        ·
        8 months ago

        I feel like real faces were heavily edited, which made them lose a lot of realism. I do get uncanny valley looking at some of them.

    • barsoap
      link
      fedilink
      English
      arrow-up
      5
      arrow-down
      1
      ·
      edit-2
      8 months ago

      Not really much of an issue with SXDL any more, and even SD1.5 got quite good at the end. Admittedly haven’t stress-tested either, though (things like clasping hands etc). It’s also not a hand-specific thing, it just happens more commonly with fingers because they’re small features:

      The thing that happens is that diffusion-type interference first nails down gross structure (which limb is where) and then fills in details. Sometimes steps somewhere in the middle decide that a limb should be somewhere else, though, and suddenly you have two, and if steps immediately after don’t think “that old limb doesn’t look like it should be there” and erase it, later steps will happily refine both to photorealism because they don’t even look at the overall composition. That is, it’s not an issue with anatomical knowledge, or not having seen enough hands, but the model changing its mind but not backtracking. It’s actually astonishing how good it can get at not making that mistake without being able to tell that it has two competing goals in mind.

  • Fredselfish@lemmy.world
    link
    fedilink
    English
    arrow-up
    9
    ·
    8 months ago

    Creppy to think that none of those faces are real but at the same time do look just like real people.

  • deafboy@lemmy.world
    link
    fedilink
    English
    arrow-up
    8
    ·
    8 months ago

    Remember when you could’ve just look them in the eyes, and if they were in the center it was not a real human?

  • scarabic@lemmy.world
    link
    fedilink
    English
    arrow-up
    8
    ·
    8 months ago

    So… sidebar. Hang on.

    When human simulacra start to approach realism, they go into the “uncanny valley” once they’re pretty good but still obviously off. What’s after the uncanny valley once they’re totally convincing? Is there even a name for that?

    • Mubelotix@jlai.lu
      link
      fedilink
      English
      arrow-up
      5
      ·
      8 months ago

      It’s a valley. Meaning you be either before, in, or after. We are after

    • paris@lemmy.blahaj.zone
      link
      fedilink
      English
      arrow-up
      9
      ·
      8 months ago

      We used the 100 AI and 100 human White faces (half male, half female) from Nightingale and Farid. The AI faces were generated using StyleGAN2. The human faces were selected from the Flickr-Faces-HQ Dataset to match each of the AI faces as closely as possible (e.g., same gender, posture, and expression). All stimuli had blurred or mostly plain backgrounds, and AI faces were screened to ensure they had no obvious rendering artifacts (e.g., no extra faces in background). Screening for artifacts mimics how real-world users screen AI faces, either as scientists or for public use, and therefore captures the type and range of stimuli that appear online. Participants were asked to resize their screen so that stimuli had a visual angle of 12° wide × 12° high at ~50 cm viewing distance.

      • bitsplease@lemmy.ml
        link
        fedilink
        English
        arrow-up
        5
        arrow-down
        2
        ·
        8 months ago

        I don’t know why people (not saying you, more directed at the top commenter) keep acting like cherry picking AI images in these studies invalidate the results - cherry picking is how you use AI image generation tools, that’s why most will (or can) generate several at once so you can pick the best one. If a malicious actor was trying to fool people, of course they’d use the most “real” looking ones, instead of just the first to generate

        Frankly the studies would be useless if they didn’t cherry pick, because it wouldn’t line up with real world usage

        • kase@lemmy.world
          link
          fedilink
          English
          arrow-up
          3
          ·
          8 months ago

          Tbh I’m more concerned about how they chose the human faces. I can’t explain it, but it feels like they were biased toward choosing ‘fake-looking’ faces, lol

          • bitsplease@lemmy.ml
            link
            fedilink
            English
            arrow-up
            0
            ·
            8 months ago

            The way it sounds right now is “AI generated faces don’t have all these artifacts 99% of the time” (I’m paraphrasing A LOT, but you get what I mean.)

            The only way it sounds like that is if you don’t read the article at all and draw all your conclusions from just reading the title.

            Don’t get me wrong, I’m sure many do just that, but that’s not the fault of the study. They clearly state their method for selecting (or “cherry picking”) images

  • Nepenthe@kbin.social
    link
    fedilink
    arrow-up
    5
    arrow-down
    1
    ·
    8 months ago

    I am uncomfortable to say that I failed 3 of the human ones. In my defense, the guy on the bottom right has pointed teeth like Sweet Tooth

    • db2@sopuli.xyz
      link
      fedilink
      English
      arrow-up
      2
      arrow-down
      5
      ·
      8 months ago

      I don’t remember where I first saw it, but it’s become a favorite saying: Isn’t it funny how it only takes a pretty face to make you want to put someones genitals in your mouth?

  • kakes@sh.itjust.works
    link
    fedilink
    English
    arrow-up
    5
    arrow-down
    4
    ·
    edit-2
    8 months ago

    Kinda makes sense, right?

    The AI images are a representation of what an AI thinks a human “should” look like, so when another AI (likely trained on a similar dataset) tries to classify them, the AI images will more closely fit what it expects a human to look like.

    • RecallMadness@lemmy.nz
      link
      fedilink
      English
      arrow-up
      1
      ·
      8 months ago

      Exactly. The AIs job is to generate humanness. The things that don’t look human get discarded, the things that have strong human indicators get kept. Oh look, the AI did its job. Shocked pikachu.

      The white thing is probably just a case of biased training data. Which is going to be a problem across all AIs. I wouldn’t be surprised if in 5-10 years (if the fad lasts longer than NFTs lmao) we find out the ‘AIs’ have all been fed biased data as yet another means of large corporations controlling the narrative of the population.

  • tygerprints@kbin.social
    link
    fedilink
    arrow-up
    2
    arrow-down
    18
    ·
    8 months ago

    It’s very odd, because “white” is not the color of any actual human being’s face. Look at the white background against this text. Have you ever actually seen anybody that color? Unless they’re coated in whitewash, you have not. Human skin is a complex blend of many different colors from pink to orange to brown to beige and many others also. Every human person is a composition. Nobody is actually white, black, red, or yellow. We’re all colors, blended together. Some are lucky enough to have dark complexions that shine like the finest of earth’s woods and minerals.

    • epicsninja@lemmy.world
      link
      fedilink
      English
      arrow-up
      10
      arrow-down
      1
      ·
      8 months ago

      Ah, you seem confused. “White” is colloquially used to refer people descended from Europeans, particularly the Caucuasus region, due to them typically having much lighter skin tones then people from other regions.

      • Mahlzeit@feddit.de
        link
        fedilink
        English
        arrow-up
        3
        arrow-down
        3
        ·
        8 months ago

        There’s no rational, scientific connection between the Caucasus and the American race term. I’m not even sure if Americans would consider Caucasians all that white, given that many of them are muslim; like the Chechen ethnicity.

        • barsoap
          link
          fedilink
          English
          arrow-up
          4
          arrow-down
          1
          ·
          edit-2
          8 months ago

          The term comes from an old theory that said that humanity started out in the Caucasus and spread from there, people becoming darker as they were exposed to more sun. The guy who made that theory wasn’t racist, but his work was used by racists (and he railed against that, saying things like “there’s villages in Africa with greater artistic and philosophical output than [European region where one of his racist “admirers” was from]”). He was a scientist and interpreted archaeological evidence – which we now understand to be the evidence for the Urheimat, and spread of, Indo-European people. Who came to the Caucasus, just like everyone else, from Africa, but that evidence hadn’t been unearthed yet. (Technically the Urheimat is probably the Ukrainian plains, not mountains, but close enough).

          All that is 2000-3000 years before the Pyramids.

          There’s actually multiple different mutations which contribute to whiteness, all caused by the double-whammy of not getting as much vitamin d from the sun, and not getting as much vitamin d from meat with the advent of agriculture. Pre-agricultural Europeans (hunter-gatherers, pastoralists) were actually quite a bit tanner than we are now, the original Proto-Indo-Europeans were (very probably) nomadic cattle herders and thus probably also tanner, agriculture and another set of whiteness genes came from the Euphrat/Tigris region.

          • Mahlzeit@feddit.de
            link
            fedilink
            English
            arrow-up
            3
            ·
            8 months ago

            The term comes from an old theory that said that humanity started out in the Caucasus and spread from there, people becoming darker as they were exposed to more sun.

            Not quite. The guy who coined the term, Blumenbach, believed that the Caucasians (in particular the Georgians) were the most beautiful and therefore must have been the original humans. Maybe “old theory” means the biblical belief that “Noah’s Ark” stranded in the Caucasus Mountains. I don’t know that Blumenbach used that as a justification. Biblical race doctrines defined races as descent from different sons of Noah.

            The Caucasians are certainly far from the palest people on the planet. The south of the region is part of Turkey and Iran. Those are maybe the most well-known countries and the region and I’m sure that no one pictures very pale people. I remember an article about the considerable diplomatic and PR efforts that Turkey undertook in the early 20th century to be made a white country under US law. I wish I could recall the details.

    • LWD
      link
      fedilink
      English
      arrow-up
      4
      arrow-down
      1
      ·
      edit-2
      7 months ago

      deleted by creator

      • tygerprints@kbin.social
        link
        fedilink
        arrow-up
        1
        arrow-down
        1
        ·
        8 months ago

        It’s true, some people on earth have beautiful dark skin, there’s even a magazine targeted toward such people called Mahogany. Just like the finest wood.