[sparky_]

Not a lawyer, but this sure seems to open a legal and ethical can of worms.

Image generation models capable of generating this type of content would necessarily need to be trained on the real thing, the possession of which is inarguably illegal and immoral.

So how could the model be legally or ethically trained? And if they _cannot_ be legally or ethically trained, then how can the _use_ of those models be okay?

What will be the implications of this in cases where _real_ CSAM was produced or possessed? Certainly this opens the door to a whole plethora of new "it's AI art, I swear!" defenses. After all, how can one definitely prove that CSAM is authentic or not, unless the chain of production is verified?

From the article: > ...If purely private possession of AI-CSAM is constitutionally protected under current caselaw but production is not, then using AI models (even locally-hosted ones) to generate child obscenity in one’s own home is not wholly insulated from criminal prosecution. Subsequently transmitting it to someone else, especially someone underage, is also grounds for liability...

Can of worms, ye be released!

[braiamp]

This is a common misunderstanding. The thing knows how a naked woman looks, also knows how a child looks, it puts two and two together and voila. It doesn't need to be trained on the real thing to be able to generate it.

[wavemode]

Well, maybe. Maybe not. See "Why can't ChatGPT draw a full glass of wine?"

https://youtu.be/160F8F8mXlo

Diffusion models do posses some capability to synthesize ideas, but that capability does not necessarily generalize to every possible use case. So it's impossible to say for certain that that is what is happening.

[Dylan16807]

We can get more certainty by testing combinations of those concepts with a whole bunch of other ones. Naked skateboarder. Child construction worker. It has a lot more variety for both of those concepts than with wine glasses.

We can also check models that have very highly vetted input sets.

[Calavar]

That video makes some good observatons, but it's also hilarious that he tried to "retrain" ChatGPT by asking it in the chat to remove some items from its dataset.

[delecti]

Does that necessarily follow? Wouldn't that be prone to outputting small naked adult women, and/or naked children with boobs?

[0cf8612b2e1e]

Why not? I am pretty sure there is no training data of “whales playing the guitar”, but if you ask a model to draw one, it will do a respectable job of imagining that scenario.

[userbinator]

Another somewhat famous example: https://en.wikipedia.org/wiki/File:Astronaut_Riding_a_Horse_...

[dragonwriter]

If it is trained on well-tagged images of adult men and women both clothed and unclothed, and clothed children (not that all pictures of unclothed children are CSAM to start with), understanding the relation of clothed to unclothed appearance could allow a model to reasonably generalize unclothed child bodies.

Further, models that are otherwise well trained with a mix of photographic and drawn content can often generalize specific concepts for which their training only includes examples from drawn content to photorealistic imagery involving that concept.

[LinuxBender]

I don't believe that is true. A woman and a child have distinct characteristics that are not interchangeable. A child for example can be detected by the shape of the nose and nostrils as just one data point. There are many more data-points that psycho-analysts use to determine if a person is attracted to children. AI would have to understand quite a bit of biology and understand how humans develop to get this right.

[Sharlin]

Image generation models are perfectly capable of mixing different concepts to create images of things they're incredibly unlikely to have seen during training.

[LinuxBender]

That sounds like a hit-or-miss concept to me. Without logic that teaches the AI why a child and an adult look the way they do it would maybe sometimes get it right by chance and other times not. I do not see how one could guarantee an outcome with the logic of the generation models unless it understands biology or was trained on real people which advertisements have no shortage of. Advertisements lack nudity and would leave out details like a hymen.

[charcircuit]

You can have artist draw art with the differences. And then you can get it via style transfer. But there are already many images of children with noses.

[jliptzin]

This is not a matter of opinion. Go to any AI image generator and tell it to generate whatever you want

[LinuxBender]

Opinions aside the question is how it learned to draw details about a child that do not exist on an adult and vice versa.

[anigbrowl]

Statistically, you'd expect this to result in depictions of children with public hair - some adults opt to get rid of theirs, but most have it. Are you sure you're not projecting your prior knowledge about human biology onto an image transformer model?

[roenxi]

> Image generation models capable of generating this type of content would necessarily need to be trained on the real thing,

I doubt that is so. In practice they might be trained on the real thing, but models generalise pretty well. It is going to be technically possible to train a model on other material (children, nudity and non-CSAM abuse scenes or maybe not even that) and have it generate CSAM.

But even if it was true, that would only make training the model illegal and ethically dubious. We use a tonne of technologies where the creator was legally and morally dubious. It's never been an ongoing issue before. So once the model is created there isn't a good reason to encumber it by how it was created.

[davisp]

I’m gonna give this a very charitable read by saying that while I find the ways that the treatment of burn victims was advanced by abhorrent means, we as a society have still benefited from those means.

> So once the model is created there isn't a good reason to encumber it by how it was created.

I am trying to be very specific here. I assume no untoward motivations from the parent commenter. I am not intending to cast aspersions. Whoever wrote this, I feel no ill will for you and this is not meant as a personal slight.

And I will be very clear, this statement as written could probably be defended because of the “by how it was created” clause.

However, “So once the model is created there isn’t a good reason to encumber it” is so… fucking I don’t even know, because what the actual fuck?

I apologize for the profanity, I really do. But, really? Are you fucking kidding me?

These models should not exist. Ever. By any means. Do not pass Go. Go directly to jail.

I understand the engineering brain enough to contemplate abstract concepts with detachment. That’s all I think happened here. But holy fuck, please pause and consider things a bit.

[throwaway7313]

Outlawing the use of existing material is vital market protection for producers. Denouncing these models may not actually be a good way to reduce harm.

[throwaway749372]

> These models should not exist. Ever. By any means. Do not pass Go. Go directly to jail.

If it's possible to produce CSAM that doesn't involve actual children and have a measurable impact in profitablity and demand of the real thing, leading to a net reduction in the harm done to children wouldn't you be on the wrong side of the argument you think you're making?

> I understand the engineering brain enough to contemplate abstract concepts with detachment.

I would argue it's a rational take.

Can we agree the goal to reduce harm of children is good? Or only if the solution is comfortable to you?

[rdtsc]

> These models should not exist. Ever. By any means. Do not pass Go. Go directly to jail.

Exactly. It's disturbing that this needs to be explained to people.

[roenxi]

All the alternatives are likely to result in more children being harmed. These models probably should exist. Ideally they'll destroy the commercial incentives for people to hurt children.

[RajT88]

> Certainly this opens the door to a whole plethora of new "it's AI art, I swear!" defenses

You are probably right, given what we saw with all the porn popup adware back in the 90's and 2000's. A friend of mine was a malware analyst for the FBI for a while.

All CSAM possession cases she heard about, the defense was "malware did it". Nearly all cases the jury convicted them. 100% of her cases for sure.

At some point using the defense everyone else uses and fails with is probably going to become a liability. Shit I am sure people are already trying to use this defense and failing!

[grepfru_it]

It only went to court because they had enough evidence to prove it is not malware. You have excluded all of the possible cases that used the malware defense and plea’d out or never went to trial.

Similarly, I think using the AI art excuse may be an uphill battle but not one that is impossible to defend

[anigbrowl]

This isn't how the legal system works. Most CP cases get prosecuted because the defendant solicits or shares CP with a minor of some other CP collectors, but let's imagine a situation where someone gets busted from something else and then investigators discover they have CP (real-world non-AI CP in this example).

Having it at all is a strict liability crime. If the defendant says malware put it on their computer and they don't know how or when it got there, that's called an affirmative defense - it's admissible as a claim, but the burden of proof for the claim is on the defendant. Otherwise you could just claim it was planted on your computer by ghosts or demons or space aliens. If the machine was infested with malware to the point of the browser being nearly inoperable and all sort so fother junk being present, a jury might buy it. But the defendant has to make some sort of showing to back up the claim. The whole thing about 'reasonable doubt' in criminal cases is not that something sounds possible or even plausible, but that you can support the claim with some mix of logic and empirical evidence like any other reasoned argument.

https://www.law.cornell.edu/wex/affirmative_defense

[JKCalhoun]

While disgusting, I'm thinking that if the courts insist on allowing this, I can try and comfort myself with this thought: I don't doubt that if you find one "AI CSAM" image on someone's drive, keep digging — you'll find illegal stuff too.

Sick people will still go to jail.

[themaninthedark]

Usually when someone is arrested for this kind of thing it's not one image but a couple thousand so you are right.

Edit add:

I have mixed feeling about this...Let me preface by saying that child abuse is abhorrent.

One of my former coworkers was arrested for possession of CSAM...he was never charged or accused of any abuse. I wasn't close to him; he was quiet and went out of his way to help. He (probably)killed himself(young, died suddenly) last week ahead of his trial.

So I have to wonder would having something like this help him and protect kids as well? Or does possession lead to abuse?

I don't know...people have problems and are sick. At what point do we write them off as irredeemable?

[JKCalhoun]

I know. I am aware of a man who abused some of my friends — also committed suicide before trial. By all accounts, most of the boys (who were from broken homes, FWIW) thought of him as a father figure and were not eager to see him prosecuted — shocked when they learned of the suicide.

A friend of the family tried to molest me when I was young. A quick search on the internet shows he kept at it (arrest records across decades abound). So, yeah, this kind of sick person kind of stays sick.

[themaninthedark]

Man that sucks. Sorry.

Not sure what else to say....

I hate the idea of permanently broken people but I also know that not everything is fixable.

[dragonwriter]

> Image generation models capable of generating this type of content would necessarily need to be trained on the real thing

This is absolutely not true. Generalization is a key capability of image generation models.

> Certainly this opens the door to a whole plethora of new "it's AI art, I swear!" defenses.

The worst justification for a criminal prohibition that I can think of is that it is provides a convenient out for the difficulty of proving another, more clearly warranted, crime.

> After all, how can one definitely prove that CSAM is authentic or not, unless the chain of production is verified?

"Beyond a reasonable doubt" is not, and never had been, "definite".