[josephg]

> Law seem pretty complete and well defined to me.

I don't think lawyers would agree with you at all.

> Just because an ai (or a human) comes out with a copyrighted character in a novel pose, it doesn't mean it's not copyrighted

My understanding is that there isn't a legal consensus on whether or not thats true. Copyright law wasn't written with AI generated art in mind. For a work to be copyrightable, my understanding is that it requires that you can make a "sweat off my brow" argument. Ie, you had to work to create something.

Does an AI count? We don't know. Or to put it in other terms, we haven't (collectively) decided as a society whether it should be copyrightable.

On the other side, "fair use" arguments are also at play here. If I train a 1bn parameter model on 5bn images, I could probably make an argument that 1/5th of a f16 probably constitutes fair use of copyrighted work. If that argument holds, I can train my model with impunity on any amount of copyrighted work so long as my model is small and the number of training examples is large.

Will that argument sway a court room? I have no idea. Is that fair? I don't know!

The law is decided by people. And we haven't had AIs like stable diffusion and ChatGPT before, so the laws haven't been written with this stuff in mind. There isn't even a legal precedent yet for how the current laws should apply to AI art. Speculating is fun, but speculating on how a judge will apply old laws to a totally novel problem is a fool's errand.

If you want to read about fun edge cases to copyright law, look up the history of copyright law for maps (can facts be copyrighted?) and how that interacts with trap streets.

This stuff is hairy and complicated even for lawyers. As an outsider, boldly claiming that copyright law is simple only demonstrates that you don't understand law. It'd be like a lawyer boldly arguing (with no knowledge) that compilers are simple.

[sam345]

This doesn't seem too complicated to me. An AI trained on images that produces a new image even if in same style as the source database is not violating copyright. It sucks for artists but that's not an argument that current law bans it. If current law banned it, every artist who was inspired in their work by other styles or image would be violating copyright. E.g, all comic book artists study and are inspired by other comic book artists. I don't doubt there will be attempts by courts and/or legislatures hostile to AI to make up new law to impose some sort of penalty/license fee on AI generated images. One approach would be make an artists style a trademark (for all I know that is already established law but I doubt it). I doubt the effort will be successful as there will be a gazillion ways around it or it will result in a relatively small number of monopolies of protected styles which would seem even a worse outcome than a flood of AI generated art. Ultimately I think artists will need to be even more careful about branding and probably will insist on prominent displays of their signatures and promoting the idea that a premium should be placed on human generated art probably through some sort of "certified human art" label. It definitely will increase the supply of art and likely decrease the number of artists that can survive financially off their work.

[josephg]

> This doesn't seem too complicated to me. An AI trained on images that produces a new image even if in same style as the source database is not violating copyright

Yeah, software "doesn't seem too complicated" to non-programmers too.

No offense, but if you aren't a lawyer then your opinion on legal matters has about as much veracity as a dentist explaining how software is made.

I have a certain amount of scorn for non-engineers telling me how their app idea is a weekend job and I should do it for free. I'm sure lawyers feel the same way about us when we claim there are easy answers around stable diffusion and copyright law.

The AI was trained without permission on copyrighted data. If it was as cut and dry as you claim then why do both parties think its worth going to court?

I don't know much about the law, but I know enough to recognise when its a job for the lawyers to figure out.

[avereveard]

Training method and recall are of no consequences because copyright law doesn't deal with technology for the most part. There is a test to wether the image is stored or not which is interesting here for the topic at hand, but it's always in the context of distribution.

The worst is that weight may be considered storage, but the point is... Law already covers that. Because it doesn't concern with technology, but with results of actions.

That is the whole problem with the argumentation, law is technology agnostic, just adding a layer of redirection doesn't matter, because tm it cares about the input and the output not what happens in between.

[josephg]

> law is technology agnostic, just adding a layer of redirection doesn't matter, because tm it cares about the input and the output not what happens in between.

What makes you think that?

The law cares about whatever lawyers decide to care about. There was a case a few years ago where (if memory serves) a black woman sued an insurance company for discrimination after the insurance company refused to provide her cover. The company was using a neural net to decide whether to cover someone. The court demanded they explain the neural networks' decision - and of course, they couldn't. The insurance company lost the case.

In the aftermath they moved from a neural net to a decision tree based ML system. The decision tree made slightly worse decisions, but they figured if it lowered their legal exposure, it was worth it. With a decision tree, they can print out the decision tree if they were ever sued again and hand it to a judge.

> law is technology agnostic

Clearly not in this case.

There's plenty of other examples if you go looking. In criminal law, they care a great deal about the technology used in forensic analysis - both in its strengths and weaknesses.

If you don't know much about law, being humble and wrong will serve you better than being confident and wrong.

[avereveard]

Insurance is not copyright and the case is not even the same subject matter.

And again that case is technology agnostic, discrimination law requires you to be able to provide proof that results are non discriminatory, law itself doesn't care that it was specifically a neural network, it only cares about the end result, the firm lost because it failed to provide required data about their decision process, not because it was using neural networks, that they used a neural network was irrelevant on its own, and it could have been fine if they baked explainability in it.

[couchand]

It's worth noting that "worse decisions" is from the point of view of the insurance company, which would prefer to act racist if only that pesky law didn't stop them, and will continue to do so to the extent they can get away with it.

[josephg]

Curious that you assume the insurance company was racist / malicious in this case. There's nothing to suggest that reading in the story I gave.

[couchand]

As you said, they were sued for discrimination and lost.

Also, differential insurance coverage tends to be one of the worst ways that systemic racism is perpetuated.