re: It's Still Kiddie Porn Even if it's AI Generated - Wisconsin Arrest

quote:

---

Thus, they, like all of these sick pedos should be put down like the animals they are.

---

Agreed.

There's just an elephant in the room that needs to be addressed.

Every guy in here has likely had a girl throw on a cheerleader outfit at some point. Or wear her high school uniform. (Looking at every St Joe Ho in existence) Are we going to pretend she was pretending to be an 18 year old high school cheerleader?

Or... Let's make a wierder comparison for the younger generation. Belle delphine marketed herself as a younger teen... Even though she was likely 20 or so, it always struck me as creepy.

Both are pretend. Both are certainly odd if you think about it

Shoot the guy in the OP, but an actual discussion should occur.

[

quote:

---

I suspect there’s a spectrum, ranging from treatable to lost cause / nature to nurture, as there is with many mental illnesses. I don’t think it’s unreasonable to assume there are likely victims of sexual abuse out there who have urges but have not acted on them.

And you should be able to call it a mental illness without giving them a pass for their actions. I can understand the reasons that folks who research this kind of thing would want some level of de-stigmatism - so those people can come forward. If there is a way to treat those people, it’s a net benefit for society to do so. It’s not unlike wanting kids who shoot up a school to get help before they pass the point of no return.

---

Yeah. Almost all pedophiles were abused as children. And I tend to agree, there is probably a wayyyyy bigger number of pedophiles who are ethical and non offending. I saw a special on one a few years back. Dude flat out admitted to being a pedophile, but that he would never lay a hand on a child. I wouldn’t let him babysit my nephew or anything, but I’d have trouble finding a problem with a dude who’s laying his cards on the table and says he would rather deny his impulses no matter how strong they were, than ruin a kid’s life. He spends time online in support groups / message boards for people like himself who genuinely do not want to act on anything.

But just like alcoholics who were physically abused by their parents who were addicts/alcoholics, we don’t excuse their behavior when they get DUI’s. We don’t excuse serial killers who were neglected as kids and have a psychopathy diagnosis. It’s a shitty hand of cards to be dealt to have the one sexual orientation that could get you locked up for decades, but it’s the way of the world.

That being said, the pedophiles who would never hurt a child should absolutely be helped, but I don’t think there’s a cure anymore than there is for alcoholism or being gay. As far as the pedophiles in prison, I hope they stay there forever. I’m sure this stuff can be treated as far as acting upon impulses, but that’s about it.

Even then what makes it tougher is a lot of psychiatrists start the first meeting off with “we have doctor patient confidentiality unless you talk about harming a child” etc.. basically no normal person has a favorable opinion of a pedophile.

quote:

---

Wonder what this will lead to in terms of AI’s liability. Who creates an AI platform that allows the creation of this content? Seems there are some questions that need to be asked and answered there too.

---

So, this part is an entirely different conversation but it’s a really big deal in the world of AI right now and I think it’s worth discussing in the context of this thread. Long post ahead. If you’re not interested in the state of generative AI, just stop reading now.

Stability AI is the company behind Stable Diffusion. Stable Diffusion is an open-source* AI text-to-image model. There have been several Stable Diffusion models released over the past couple of years, but the game-changer was the 1.5 model released in late 2022.

The things that made Stable Diffusion way different from the other AI image generation models like DALL-E, Midjourney, etc. were A) the fact that it was released to the public for free and B) that it can be run on consumer hardware. So users can download the model itself and run it on their own GPUs, rather than using an online service like DALL-E. Additionally, Stable Diffusion 1.5 could be trained (fine-tuned) on consumer hardware.

Fast forward to 2024, and there is an entire ecosystem that has built up around Stable Diffusion. Since SD1.5 was by far the most popular open-source model, it became the model of choice for a lot of published AI research. There is a thriving community that is constantly implementing improvements from the published research and then releasing it to the public for their own use. Meanwhile you have people training (again, more accurately “fine-tuning”) the base Stable Diffusion models to focus them on various styles, themes, etc.

In many ways it’s a case study in everything good about open source software.

But then there’s the dark side. You see, since Stable Diffusion models can run locally on consumer hardware, Stability AI has no control over what people do with the models once released. Since they can be trained on consumer hardware, Stability can’t control what subject matter they are trained on. And honestly, you don’t need to train a generative AI on CSM for it to be capable of creating simulated CSM. It knows what a child looks like, and it knows what naked people look like. It can put 2 and 2 together, in the same way that it can create any number of other things it’s never seen.

The online services avoid this complication because they have the ability to filter/censor inputs in real time. Creating CSM is certainly a violation of the Stable Diffusion license, but there’s no way for Stability to enforce it. And CSM is only one facet (albeit a major one) of the potential concerns. You also have deepfakes/face swapping, text-to-audio generation (voice deepfakes), the controversy around where the training data comes from, etc.

So now there’s a lot of focus, both in the AI field as well as in Washington DC, around “responsible” or “safe” AI. But it’s complicated by lobbying efforts and special interests. There is a big push right now to outright ban model weights (i.e. the files that allow you to run and train a model locally) from being released to the public. Companies like OpenAI, Google, Meta, etc. are big proponents of this approach. Why? Well maybe safety plays a part, but it’s fairly obvious that they stand to benefit from gatekeeping access to/development of generative AI as well.

You have some folks who feel the public should have access to whatever anyone is willing to release open source, with no restrictions whatsoever. You have others who feel that the companies developing these models should be building in various safeguards to effectively censor them/prevent this sort of thing from occurring. And you have others who believe the general public shouldn’t have access to actual model weights in the first place.

It’s a legal and ethical minefield right now, and the regulatory / legislative decisions that are made over the next year or two will have a major impact on the landscape of generative AI models as we move forward.

*Whether Stability AI is actually “behind” Stable Diffusion, and whether Stable Diffusion 1.5 is truly “open source,” are matters of debate. It’s complicated, but not particularly germane to the larger conversation at hand.

quote:

---

Since they can be trained on consumer hardware, Stability can’t control what subject matter they are trained on.

---

It would be like blaming Adobe because someone used Photoshop to draw child porn.