"A race against time to protect the children of our country from the dangers of AI."
AI-generated child sex imagery has every US attorney general calling for action::"A race against time to protect the children of our country from the dangers of AI."
They're not pictures of real people, proceeding against it on that basis undermines the point and makes them look like idiots. It should be banned on principle but ffs there's got to be a way that doesn't look reactionary and foolish.
But aren't these models built from source material? I imagine if you want CP AI, you need actual CP to train it, no? That definitely would be a problem.
Not necessarily. You could mix PG pics of kids with a Lora trained on pubescent looking but otherwise legal naughty bits. At least that would be the sane way to do it. But hey, world's full of insane people.
No, you can use a genetic algorithm. You have your audience rate a legal, acceptable work. You present the same work to an AI and ask it to manipulate traits, and provide a panel of works to your audience. Any derivative that the audience rates better than the original is then given back to the AI for more mutations.
Feed all your mutation and rating data to an AI, and it can begin to learn what the audience wants to see.
Have a bunch of pedophiles doing the training, and you end up with "BeyondCP".
My question is where did they get the training data for a sufficiently advance CP image generator. Unless it's just ai porn with kids faces? Which is still creepy, but I guess there are tons of pictures that people post of their own kids?
Manga, manwha(?) CG sets etc of shota/loli. Sprinkle in some general child statistics for height, weight etc . And I'm sure social media helped as well, with people making accounts for their babies for God sake.
It's obviously very distasteful but those needs don't just go away. If people with that inclination can't satisfy their sexual urges at home just looking at porn, it seems more likely they're going to go out into the world and try to find some other way to do it.
Also, controlling what people do at home that isn't affecting anyone else, even in a case like this isn't likely to target exactly just those people and it's also very likely not to stop there either. I'd personally be very hesitant to ban/persecute stuff like that unless there was actual evidence that it was harmful and that the cure wasn't going to be worse than the disease.
Humans have been raping kids since our inception. Childhood is a relatively modern concept that young adults are now apart of. It's an ugly and pervasive subject that needs further study to reduce child harm.
No one chooses to be a pedophile - as far as we know it's just the unluckiest possible sexual attraction.
Stigmatizing it won't help anyone - those people need help and everything that doesn't hurt real children until they get themselves that psychological help is good in my book
Isn't AI generated better than content sourced from real life? It could actually drive a reduction in child sexual abuse instances due to offenders leveraging alternative sources.
Based on my understanding of how current diffusion models work, you actually don't need to train it on CP. As long as it knows how humans look like without clothes and how children look like even if fully clothed with abayas and stuff, it can make the relation and generate CP when asked to.
Just to be clear, I'm totally against any form of CP and CSAM. Just explaining how the tech works.
I see what you're saying, but ai has yet to offset regular porn production at all. There's no reason I see to think accepting ai cp would do anything but normalize it and make it more accessible, possibly increasing demand for the real stuff.
Also, the ai models need to be trained on something...
One big problem is that it makes enforcement of real abuse impossible. If there is an explosion of that kind of ai generated content and it gets good enough to be confused for the real thing, then real abuse will slip under the radar. It would be impossible to sift through all that content trying to differentiate between ai generated and real if ai generated were ever allowed.
Majority of pedofiles never offend and most of the people in jail for child abuse are plain old rapists with no particular interest in kids per se, they're just an easy target.
This is the same old "violent games makes people violent" -argument all over again.
On Wednesday, American attorneys general from all 50 states and four territories sent a letter to Congress urging lawmakers to establish an expert commission to study how generative AI can be used to exploit children through child sexual abuse material (CSAM).
In particular, open source image synthesis technologies such as Stable Diffusion allow the creation of AI-generated pornography with ease, and a large community has formed around tools and add-ons that enhance this ability.
Since these AI models are openly available and often run locally, there are sometimes no guardrails preventing someone from creating sexualized images of children, and that has rung alarm bells among the nation's top prosecutors.
Establishing a proper balance between the necessity of protecting children from exploitation and not unduly hamstringing a rapidly unfolding tech field (or impinging on individual rights) may be difficult in practice, which is likely why the attorneys general recommend the creation of a commission to study any potential regulation.
In the past, some well-intentioned battles against CSAM in technology have included controversial side effects, opening doors for potential overreach that could affect the privacy and rights of law-abiding people.
Similarly, the letter's authors use a dramatic call to action to convey the depth of their concern: "We are engaged in a race against time to protect the children of our country from the dangers of AI.
The original article contains 960 words, the summary contains 225 words. Saved 77%. I'm a bot and I'm open source!
I'm sure it is entirely coincidental that the call to action is to restrict/control the free open-source software, and leaves Google and Microsoft safely in control with their curated models.
This is just like the time the US made websites responsible for their users' content, and coincidentally made it much more legally dangerous to start your own social media platform.
But sure, I mean, just think of the (imaginary) children! We need to stop this theoretical abuse of imaginary children by passing laws that make it harder for any AI not created by a tech giant to operate.
An AI generated image is essentially the solution to a math problem. Say the images are/become illegal. Is it then also illegal to possess the input to that equation? The input can be used to perfectly replicate the illegal image after all. What if I change a word in the prompt such that the subject of the generated image becomes clothed? Is that then suddenly legal?
I understand the concern, but it's just incredibly messy to legislate what amounts to thought crimes.
Maybe we could do something to discourage distribution, but the law would have to be very carefully worded to prevent abuse.