Hacker News Reader

Comment by goles | Hacker News Reader

I think to be fair to the study, it’s building off a previous study that demonstrated Diffusion models can produce CG-CSAM. Which seems like it could have some pretty serious consequences, for victims, law enforcement, AI Ethics, and the Justice system.

They reached that conclusion then naturally asked, if the model can produce that, how much of it does it need to be trained on to produce it? And they discovered the answer to that is very little.

Which could be for any number of reasons that are far beyond how I understand ML.

View on HN

> Which seems like it could have some pretty serious consequences, for victims, law enforcement, AI Ethics, and the Justice system.

What actual difference is there between a machine generating this material or a human being drawing it by hand? It seems like we're living with the "pretty serious consequences" already, and yet, it seems like we've found a way to manage those effectively already.

> if the model can produce that, how much of it does it need to be trained on to produce it? And they discovered the answer to that is very little.

The answer may very well be "none at all," particularly if these systems can create image fragments by inference and a different system or even a human can assemble them.

> are far beyond how I understand ML.

If you have a crappy product, get it regulated, it grants it the imprimatur of credibility and it hamstrings your competitors and startups in the space. An understanding of ML may not be required at all to understand this regulatory situation.

Obligatory IANAL, just lifting from the study.

> What actual difference is there between a machine generating this material or a human being drawing it by hand?

Page 8 Section 4 Legal concerns ———

U.S. Code[…]uses a standard prohibiting any visual depiction of CSA that is “virtually indistinguishable” from a minor engaging in sexual conduct. The definition it uses specifically references computer-generated material as being in scope.U.S. Code[…]states that any depiction of a minor that both contains sexually explicit conduct and is obscene can be prosecuted with the same penalties[...]

As such, the current status of CG-CSAM appears to be that prosecution is possible for any representation deemed both graphic and obscene,[…]if that material is indistinguishable. Now that CG-CSAM has reached the point that it may[…] test the application of some of these laws may be tested.

Realistic CG-CSAM also presents obvious difficulties when it comes to prosecutions for CSAM possession[…] in general, the appearance of a child being abused has been sufficient for prosecution[…]alternatives may need to be tested that do not require positive identification of a real-world victim (or worse, that would require such a victim or their family to testify)

> The answer may very well be "none at all,"

You may be correct but they do cite a number of counter-measures that will need to be tested and studied.

What is the rationale for prosecuting someone who has AI produced CSAM that isn't based off a real person?

The rationale used so far has been that it creates a market for production and revictimized the children. If this is not the case, what are we prosecuting people for exactly?

Given how many kids (teens) produce this stuff for each other and have been prosecuted as criminals for it, I think our approach to enforcement is completely dogmatic and ineffective. While SWATting is still a thing, there is no way in hell this won't become weaponized.

If the goal is to stop actual human trafficking, then target movement. Possession shouldn't be illegal (stops your pissing-cupid garden statue from being declared CSAM tomorrow, and settles making criminals of shutterbug parents at bathtime), but broadcasting, exhibiting or selling anything to do with it (or tutorials for producing it) should be illegal. I'd even suggest uploading it to cloud/external storage should be illegal.

Distribution allowable only between owner and any direct recipient(s), while both parties are under or over 18 (stops the adult-child grooming issue). Solves the underage-revenge-porn issue as well; Bob15m is on the hook for any forwarding or "losing" of the nude selfie he solicited from Alice14f, lest he face CSAM distribution charges. Either of them possessing it wouldn't be illegal, but it would be a liability if lost or stolen, so both are best served deleting it as soon as possible.

They say every advance in tech starts with pornography. This is how you normalize a culture of responsible data stewardship-- by getting people used to the idea of personal risk associated with keeping data longer than you should.

I absolutely agree with you that possession should not be illegal.

It doesn't need to be trained on CSAM at all. You can make a 2-pass workflow using Stable Diffusion: start generating normal porn for a few steps, and for the rest generate children. I don't think this can be stopped.

I don't think that will satisfy "the market" for this material. I've spent time as a forum moderator before and have seen perhaps a dozen pieces of genuine CSAM. There's something about the proportions that makes it instantly stomach-turning. The attempts I've seen by AI porn "artists" just look like small women with somewhat large heads and don't have any of that soul-searing effect. If that's what the "customers" are after I think the training set would have to have a significant amount of the real stuff to replicate it accurately.

Reply on news.ycombinator.com