I'm really uncomfortable with both thought crime and implied free speech issues in this and the mechanistic aspects of how an llm and gpt system works makes me wonder who else but the hosting company can be legally at risk here.
I think it would be a mistake to say this is victimless crime, even if I can't point to a victim.
The hentai/domai and related cartoon industry probably has a higher legal risk if this gets up.
Today, seeking out images like that puts you on the definite wrong side of a line. A child was harmed to produce that image. You’ll face significant legal and societal consequences if you get caught with it in your possession. Given that you’ve already crossed the line, maybe the next line is not as hard to cross.
Whereas if images like these are decriminalised, folks who look at them don’t have to directly correlate their actions with any specific harm to any specific child, and so that line remains uncrossed.
I think there may be complications around, for example, using inpainting on an image of a specific child, but overall, my approach is more one of harm reduction. Whichever approach leads to fewer real world children being abused is the one to take, and moral outrage should take a back seat to pragmatism if it turns out that decriminalising these images reduces real world offending.
Not sure there’s a good way to test and find out which of us is right though.
But when the difference between the difference between the digital representation of a child and that of an adult just becomes a variable in a parameter space?
It'll be impossible to stop.
Child pornography is a byproduct of human trafficking, and that's where the focus needs to be.
But real children _were_ harmed in order to train the AI.
A more accurate and less click-baity title might be 'EU proposes criminalizing AI-generated and deepfaked child sexual abuse content'
On the other hand, any AI-generated content, no matter what it shows, is substantially different from CSAM, as creating the next image did not require any more abuse. I'd also argue that "artistic" depictions should not be labeled CSAM and instead the older CP term is more appropriate, and the difference is meaningful with respect to whether it should be criminalized; someone paying money for CSAM funds a rape of a kid, but the same person paying same money for an "artistic depiction" just funds someone clicking around on a computer - there's a world of difference in the harmfulness of these activities, and the legal penalties should reflect that.
that distinction becauses dodgier when you consider the checkpoint training methods are suspected of using csam imagery within the training corpus[0], but I agree with your points otherwise.
In principal an AI generated image doesn't exploit someone elses' existence if trained on safe and sane data.
[0]: https://www.forbes.com/sites/alexandralevine/2023/12/20/stab...
Moreover even AI generated CSAM has to trained somewhere. It is not certain at all that an increase in AI generated CSAM market would necessarily mean less child abuse.
E.g. I have a local example in mind after a local pedophile was arrested in a bust of an international CSAM ring, the story goes that he was consuming CSAM, and joining that "better" ring of CSAM content required providing "original content" (that's one way they protected against law enforcement infiltration) and allegedly that need for that was the reason he went from watching CSAM to actually molesting several kids to film it. If we can remove this kind of motivation, that's IMHO a good thing for society.
I worry that left to their own devices, some may slip deeper into the psychosexual sadism side of things.