A sex offender convicted of making more than 1,000 indecent images of children has been banned from using any “AI creating tools” for the next five years in the first known case of its kind.
Anthony Dover, 48, was ordered by a UK court “not to use, visit or access” artificial intelligence generation tools without the prior permission of police as a condition of a sexual harm prevention order imposed in February.
The ban prohibits him from using tools such as text-to-image generators, which can make lifelike pictures based on a written command, and “nudifying” websites used to make explicit “deepfakes”.
Dover, who was given a community order and £200 fine, has also been explicitly ordered not to use Stable Diffusion software, which has reportedly been exploited by paedophiles to create hyper-realistic child sexual abuse material, according to records from a sentencing hearing at Poole magistrates court.
You’re spreading misinformation and getting called out for it.
Just a note - csam has been found in model training sets: https://cyber.fsi.stanford.edu/news/investigation-finds-ai-image-generation-models-trained-child-abuse
Ok? Hundreds of images of anything isn’t going to necessarily train a model based on billions of images. Have you ever tried to get Stable Diffusion to draw a bow and arrow? Just because it has ever seen something doesn’t mean that it has learned it, nor, more importantly, does that mean that is the way it learned it, since we can see that it can infer many concepts from related concepts- pregnant old women, asian nazis, black george washingtons (NONE OF WHICH actually have ever existed or been photographed)… is unclothed children really more of a leap than any of those?
It is, yes. A black George Washington is one known visual motif (a George Washington costume) combined with another known visual motif. A naked prepubescent child isn’t just the combination of “naked adult” and “child” naked children don’t look like naked adults simply scaled down.
AI can’t tell us what something we’ve never seen looks like… a kid who knows what George Washington and a black woman looks like can imagine a black George Washington. That’s probably a helpful analogy, AI can combine simple concepts but it can’t innovate - it can dream, but it can’t know something that we haven’t told it about.
What you’re saying is based on the predicate that the system can’t draw concepts it has never seen which is simply untrue. Everything else past that is sophistry.
Edit: also not continuing a conversation with someone who is hostile to the basic rules of logic.
You have a basic misunderstanding of how AI works and are endowing it with mystical properties. Generative AI can’t accurately infer concepts or items it doesn’t understand. It has all the knowledge of the internet but if you ask it to draw a schematic for a hydrogen bomb it’ll give you back hallucinated bullshit. I’ll grant that there’s a small chance that just enough random details have been leaked that the AI may actually know how to build a hydrogen bomb - but it can’t infer how that would work from “understanding physics”.
Either way, these models were trained on csam, so my initial point is accurate and not misinformation.
It isn’t misinformation, though, generative AI needs a basis for it’s generation.
The misinformation you’re spreading is related to how it works. A generative AI system will (without prompting away from it) create people with 3 heads, 8 fingers on each hand and multiple legs connecting to each other. Do you think it was trained on that? This argument of “it can generate it, therefore it was trained on it” is ridiculous. You clearly don’t understand how it works.
deleted by creator
You’re extremely correct when it comes to combining different aspects of existing works to generate something new - but AI can’t generate something it doesn’t know about. If a generative model knows what a prepubescent naked body looks like it has been exposed to them before. The most generous way to excuse this is that medical diagrams exist and supplied the majority of inputs for any prompts about cp to work off of. A must more realistic view is that some cp made it into the training set.
I don’t disagree with any of your assessments but if you wanted a Van Gogh painting of a Glorp from Omnicron Persei 8, you’ll get out… something, but because the model has no reference for Glorps it’ll be hallucinations or guesses based on other terms it can find.
To be clear, I’m coming at this from the angle as someone who has trained and evaluated models in a company that’s used them for the better part of a decade.
I understand I’m going up against your earnestly held belief, but I’ve seen behind the curtain on a lot of this stuff and hopefully in time the way it works becomes demystified for more people.
For reference, the comment I made was improperly displayed, and I thought I replied to the wrong person. It said:
Has your model seen humans in a profile view? Has it seen armor? Has it seen Van Gogh style paintings? If yes then it can create a combo of those things.
For CSAM it needs to know what porn looks like, what a child looks like and what a naked pubescent body looks like to create it. It didn’t make your van Gogh painting from nothing it had an idea of what those things were.
Yes, that’s my point. It didn’t need to be trained on a portrait of Van Gogh in profile; it had several portraits of Van Gogh, a bunch of faces in profile, and used them to create something new. In the exact same way, a network trained on photos of people that include nude adult bodies and children in innocent situations can feasibly create facsimiles of csam without ever having been trained on it.
Yea, specifically, the model shouldn’t have had access to a significant training set on naked prepubescent bodies - that’s been my main objection in this thread.
Except you can’t know that. CSAM has been found in training data already and as long as they pull from social media they will continue to be trained with more.
https://cyber.fsi.stanford.edu/news/investigation-finds-ai-image-generation-models-trained-child-abuse
Awesome link, I’ll share it up thread where someone was asking for it. Yea, it’s something that’s hard to prove since models aren’t upfront with how they’re sourcing their data.
Are you paying attention? It didn’t need to be trained on a portrait of Van Gogh in profile; it had several portraits of Van Gogh, a bunch of faces in profile, and used them to create something new. In the exact same way, a network trained on photos of people that include nude adult bodies and children in innocent situations can feasibly create facsimiles of csam without ever having been trained on it.
The model should not have had access to naked prepubescent imagery. If it did, that’s a problem. My argument in this thread is that it did have access to csam and thus is able to regurgitate them.
I honestly think you and I are in agreement. I’m not arguing that the model is regurgitating known csam but the model ingested csam[1] and the output is derived from that csam. The fact that it can now make csam in the style of Van Gogh is a property of how these models can combine motifs… the fact that it understands how to generate csam at all is the problem.
The bodies of children are not just small versions of adult bodies.There are meaningful differences that an ai wouldn’t be able to just guess. Also do you not see any problem in using photos of real children to generate csam? Imagine someone used a picture of your child/niece/nephew to generate porn. Does that not feel wrong to you? It’s still using real photos of real children either way, even if it’s abstracted through training data.
If the system must see something to generate it, and the system can’t generate things that don’t exist, then how is it generating pregnant old women?
Because it’s a transformation that can be accurately predicted, at least as far as we can conceive. This is sort of the problem with this thread - there are plenty of examples of derivative combinations that are being presented as counter examples but naked children don’t just look like adults scaled down. This is a rather unique situation because most people have been parents or siblings and know what naked children look like but photographs of that nudity are restricted and shouldn’t be included in model training.
The other example we might have to work with would be copywrited material but we know that models did consume material they weren’t licensed to - as a result AI has been able to generate Disney characters and the like in a recognizable way.
k