These companies should be criminally liable for training their models on child porn.

you are viewing a single comment's thread
view the rest of the comments
[–] 1 point 16 hours ago* (2 children)

Indeed, I agree with the last part. We should tread carefully with policy here, for two main reasons:

  1. We still have exceptionally little evidence one way or another, which is honestly insane given the topic's importance
  2. "Protect the children" narrative is commonly abused for tyrannical laws that trample otherwise unrelated freedoms. I can easily see how this narrative can transform in broader AI oversight in a way that tramples democracy and fair access to information.

I merely addressed the "why is this different" issue. You're unlikely to even want to watch any form of CSAM unless you're into this, i.e. unless you're a pedophile. You don't have to be a sadistic psychopath to enjoy wreaking havoc in GTA or beating the hell out of your opponent in Mortal Kombat.

CSAM attracts a very specific demographic, and we need to make sure we understand how AI-generated CSAM affects this demographic and whether it induces real life abusive behaviors, or maybe represses them. And yes, it is also important to understand how it affects children, how likely they are to find such materials, and how it affects their sexual development. But this part is exceptionally hard to study, because you can't just take children and expose them to CSAM to see how it's going.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 1 point 6 hours ago (1 child)

    You don't have to be a sadistic psychopath to enjoy wreaking havoc in GTA or beating the hell out of your opponent in Mortal Kombat.

    I mean, I personally don't find those things appealing and wonder how you can enjoy those things, but it's been shown to reduce aggression. Crime went down as violent video games went up too - I know, correlation isn't causation - but I could see for example how a cousin of mine would have been more likely to join the local gang if he couldn't instead be home playing GTA as a teen.

    You're unlikely to even want to watch any form of CSAM unless you're into this, i.e. unless you're a pedophile.

    CSAM yes, but there's an important distinction that's been blurred in recent years and I think it's no coincidence - the distinction between CSAM and pedophilic content. CSAM involves actual human children - it's in the name. Pedophilic content however can include content that doesn't include actual children, like drawings.

    When you have an unevenly enforced justice system, who does it benefit to square the two? If the ruling class is full of pedophiles who actually rape children, then putting CSAM in the same rank as a drawing benefits them. It would be the same as saying a snuff film is as bad as a movie with props. And that benefits them, because it makes CSAM seem less bad than it is, while giving them expanded powers against those they don't like. Calling something AI-generated CSAM (assuming it's from a model that wasn't trained on any specifically) makes it seem that what people like Trump do is actually just as bad as a basement dweller geberating pedo content. The reality is one is FAR more horrific than the other.

    On everything else you said, I mostly agree. I just want to point out for point 1, I was alive during the right when there was little research or evidence about violent video games too. Yeah, movies and books like that existed, but that wasn't as immersive because those mediums aren't interactve, and that was the important distinction back then.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 2 points 5 hours ago*

    Violent games have been shown to reduce aggression

    Yes, but we don't know if it translates the same way to CSAM. That's why we desperately need more studies. Before then, any stance is just a wild guess.

    The distinction between CSAM and pedophilic content

    Indeed, I use CSAM as an accepted broad term, even though, if AI was not trained on real CSAM, sexual abuse was not involved. You raise an interesting political point on why we shouldn't.

    Other than that, we need to know several things when considering legality of such options:

    • Is real CSAM involved in a given model training data
    • Does consumption of different forms of AI-generated content increase or decrease real world sexual abuse rates
    • Is there any distinction between hyperrealistic AI-generated content and drawing-like content in terms of real world behaviors and training data
    • How can we protect children from being exposed to such content online - and how incidents of it happening anyway affect sexual and mental development of children

    Powerful people may intentionally put equality between real-life CSA and AI-generated content

    That's an interesting angle to look at. So, we also need to know how normalization or criminalization of AI-generated materials affect public perception of real world CSA. Makes sense.

  • source
  • parent