Yik Yak logo
Download TikTok Instagram X
Yik Yak icon
Join communities on Yik Yak Download
US Politics icon
US Politics 2w
I don’t know how you’d change the Constitution in a way that wouldn’t potentially affect other legal speech but I feel like the concept of AI-generated CSAM would be so inconceivable to the writers and feels like it should warrant an exception
comment

7
upvote 21 downvote
default user profile icon
Anonymous 2w

Like dude imagine being at the Constitutional Convention and trying to explain that their rule to allow free speech and religion and stuff, inspired by their distaste for authoritarianism and mixed religious/political backgrounds, would one day potentially allow for a machine to perform a computation so complicated and sophisticated that it can basically turn words into a realistic video of child pornography. They’d be wildly confused but probably also terrified

upvote 15 downvote
default user profile icon
Anonymous 1w

It’s actually *not* constitutionally protected. The judges clearly don’t know anything about AI. They said it’s constitutionally protected because it has no victims and isn’t real. But AI uses actual photos and videos from across the internet to make new media. That means that AI generated CSAM is made using actual children’s faces, bodies and yes, genitals. So per the Constitution and previous SCOTUS ruling, AI CSAM is not constitutionally protected.

upvote 1 downvote
default user profile icon
Anonymous replying to -> #1 1w

Well listen, I’m not trying to defend the ruling or anything obviously, but I think these are two separate concerns. Is it a concern that CSAM might appear in the training data of a lot of these frontier labs? Absolutely. Does them eliminating any CSAM from the training data mean you can’t use the model to generate CSAM? Not quite. Models these days generalize well enough to make generating CSAM very possible without it necessarily being present in the training data. In fact, I’d say banning

upvote 1 downvote
default user profile icon
Anonymous replying to -> OP 1w

the use of models trained on CSAM to make CSAM is a lot weaker a restriction than just banning people from generating CSAM altogether

upvote 1 downvote
default user profile icon
Anonymous replying to -> OP 1w

Like for example I can very easily generate a picture of a dog face morphed onto the front of a jumbo jet but that doesn’t mean it’s in the training data

upvote 1 downvote
default user profile icon
Anonymous replying to -> OP 1w

Correct. So AI-generated CSAM would be produced either using pre-existing CSAM or media of actual children combined with regular porn. Either way, children are victims.

upvote 1 downvote
default user profile icon
Anonymous replying to -> #1 1w

I think we’re kind of just arguing over semantics here but I think this is generally a crude oversimplification. It’s not like image generation models are stitching together pictures of children with pictures of porn to generate CSAM. It’s learning a statistical distribution of what those concepts look like. You could feasibly make a model that doesn’t use real images of children at all that’s still capable of generating child pornography based on illustrations, cartoons, etc. of children.

upvote 1 downvote