There has been a ton of CSAM and CP arrests in the US lately, especially from cops and teachers, along with at least one female teacher seducing boys as young as 12. I cannot understand the attraction to kids. Even teens. Do these people think they are having a relationship, or it is somehow okay to take away another human beings’ innocence? Is it about sex, power, or WTH is it? Should AI generated CSAM and CP be treated the same as a real person since it promotes the same issues? I am a grandfather, and I am worried about how far this will go with the new AI being able to put anyone’s face into a Porno movie too.

It seems to me that a whole new set of worldwide guidelines and laws need to be put into effect asap.

How difficult would it be for AI photo apps to filter out words, so someone cannot make anyone naked?

  • Apepollo11@lemmy.world
    link
    fedilink
    arrow-up
    12
    ·
    edit-2
    1 year ago

    I’m only going to tackle the tech side of this…

    How difficult would it be for AI photo apps to filter out words, so someone cannot make anyone naked?

    Easy. The most popular apps all filter for keywords, and I know that at least some then check the output against certain blacklisted criteria to make sure it hasn’t let something slip through.

    But…

    Anyone can host their own version and disable these features, allowing them to generate whatever they want, in the exactly same way that anyone can write their own story containing whatever they want. All you need is the determination to do it, and some modicum of ability.

    People have been been creating dodgy doctored photos long before computers. When Photoshop came out, it became easier, and with AI it’s easier still. The current laws about creating and distributing indecent images still apply to these new images though.

    • Adalast@lemmy.world
      link
      fedilink
      arrow-up
      2
      ·
      1 year ago

      Technically the diffusers all have the ability to filter any material from the actual outputs using a secondary CLIP analysis to see if it kicks out any keywords which indicate that a topic is in the image. From what I have seen, most AI generation sites use this method as it is more reliable for picking up on naughty outputs than prompt analysis. AI’s are horny, I play with it a lot. All you have to do is generate a woman on the beach and about 20% of them will be at least topless. Now, “woman on the beach” should not he flagged as inappropriate, and I don’t believe the outputs should either because our demonization of the female nipple is an asinine holdover from a bunch of religious outcasts from Europe who were chased our for being TOO restrictive and prudish, but alas, we are stuck with it.

      • That’s putting a lot of faith into CLIP, though. The thing is, to get CLIP to detect things like child porn reliably, you do need to train it to make the distinction. In my experience, CLIP tends to make up keywords, or at least misunderstand the situation surprisingly common.

        If it weren’t super illegal and super unethical, AI could easily distinguish normal porn from illegal porn if you feed it enough tagged data of both. Categorisation is something these models are very good at, after all. That’s never ever going to happen (imagine the poor shmuck being hired to tag child rape for a dollar a day, horrific) but it’s the only way I trust AI to come up with something like this.

        I think we need more research into this field. I’m also at least a little mad that AI companies release these models into the wild before the science is ready to prevent it from becoming a child rape image generator for the mentally ill. Companies just seem to throw their hands in the air and go “well we didn’t program it to do that, not our fault!” and deny any responsibility for what they’ve created.

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

          You are correct, CLIP can misinterpret things, which is where human intelligence comes in. Having CLIP process the probabilities for the terminology that you describe what you are looking for then utilizing a bit of heuristics can go a long way. You don’t need to train it to recognize a nude child because it has been trained to recognize a child, and it has been trained to recognize nudity, so if an image scores high in “nude” and “child” just throw it out. Granted, it might be a picture of a woman breastfeeding while a toddler looks on, which is inherently not child pornography, but unless that is the specific image that is being prompted for, it is not that big of a deal to just toss it. We understand the conceptual linking so we can set the threshold parameters and adjust as needed.

          As for the companies, it is a tough world surrounding it. The argument of a company that produced a piece of software being culpable for the misuse of said software is a very tenuous one. There have been attempts to make gun manufacturers liable for gun deaths (especially handguns since they really only have the purpose of killing humans). This one I can see, as the firearm killing a person is not a “misuse”, indeed, it is the express purpose for it’s creation. But what this would be would be more akin to wanting to hold Adobe liable for the child pornography that is edited in Lightroom, or Dropbox liable for someone using Dropbox API to set up a private distribution network for illicit materials. In reality, as long as the company did not design a product with the illegal activity expressly in mind, then they really shouldn’t be culpable for how people use it once it is in the wild.

          I do feel like more needs to be done to make public the training data for public inspection, as well as forensic methods for interrogating the end products to figure out if they are lying and hiding materials that were used for training. That is just a general issue though that covers many of the ethical and legal issues surrounding AI training.