this post was submitted on 11 Jul 2023
59 points (96.8% liked)
Asklemmy
43376 readers
1371 users here now
A loosely moderated place to ask open-ended questions
If your post meets the following criteria, it's welcome here!
- Open-ended question
- Not offensive: at this point, we do not have the bandwidth to moderate overtly political discussions. Assume best intent and be excellent to each other.
- Not regarding using or support for Lemmy: context, see the list of support communities and tools for finding communities below
- Not ad nauseam inducing: please make sure it is a question that would be new to most members
- An actual topic of discussion
Looking for support?
Looking for a community?
- Lemmyverse: community search
- sub.rehab: maps old subreddits to fediverse options, marks official as such
- [email protected]: a community for finding communities
~Icon~ ~by~ ~@Double_[email protected]~
founded 5 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Yes, but training is not infringement under current law.
If you learn to draw by tracing Mickey Mouse, but then professionally draw original works, you haven't infringed copyright.
If you subsequently draw Mickey Mouse, you'll hear from Disney's lawyers.
So yes, AI producing IP protected material when prompted results in infringement, just as any production would.
The thing people seem to be up in arms about are things like copying style (not protected) or using for training (not infringing).
If anything, all of this discussion over the past year around AI has revealed just how little people understand about IP laws. They complain that there needs to be laws for things already protected and prohibited, and they complain that companies are infringing for things that are not protected nor prohibited.
For example, in relation to the OP question, in the US there is no federal protection around IP rights for voices and case law to the opposite.
Tracing Mickey is already copyright infringement. It's insane but that's how it works.
I think copyright is a weird idea in this day and age where you're almost always standing on the shoulders of giants too but that doesn't make it go away. Best we can do is hack it (copyleft).
Not copyrightable but I'm not sure whether a style could be trademarked instead.
Whether that's infringing or not is hotly debated topic. Key point here is whether training falls under fair use. If it doesn't, training on copyrighted material without a license would be infringing under most jurisdictions.
Because US == world. I'd love to leave this implication behind on Reddit.
You think tracing for educational use which is then never distributed such that it could not have a negative impact on market value is infringement?
What the generative AI field needs moving forward is a copyright discriminator that identifies infringing production of new images.
But I'll be surprised if cases claiming infringement on training in and of itself end up successful under current laws.
And yeah, most of the discussion around this revolves around US laws. If we put aside any jurisdiction then there is no conversation to be had. Or we could choose arbitrary jurisdictions to support a position, for example Israel and Japan which have already said training is fair use.
That's not what I think, that's what the law says.
I said what I think in the second paragraph. Sorry if I wasn't being extra clear on that.
Good luck with that.
Depends. If the imitative AI imitates its source material too closely, that could absolutely be laid out as a distribution of copyrighted material.
Think about it like this: If I distributed a tarball of copyrighted material, that would be infringement, eventhough you'd need
tarto unpack it. Whether you need a transformer ortarto access the material should make no difference in my layman interpretation.No, it doesn't. The scenario outlined squarely falls under fair use, particularly because of the non-distribution combined with research/education use. Fair use is not infringement.
We'll see.
I mean, if we're talking about hypothetical models that only produce infringing material, you might be right.
But if we're talking about current models that have no ability to reproduce the entire training set and only limited edge case reproducibility of training images with extensive prompt effort, I stand by being surprised (and that your
tarmetaphor is a poor and misleading one).If we're going with poor metaphors, I could offer up the alternative of saying that distributing or offering a cloud based Photoshop isn't infringement even though it can be used to reproduce copyrighted material. And much like diffusion based models and unlike a tarball, Photoshop requires creative input and effort from a user in order to produce infringing material.