I really don’t want my photos, writings, etc to be used for things like StableDiffusion or ChatGPT, but some of them I still want to release under an open access license that’s free for others to use in conventional derivative works, just specifically not AI. Does such a license exist?
Or at the very least, if my work is to be used to train AI, then I think the final models and neural networks themselves need to be open source and also free for anyone to use (as in, people should be able to download and run the AI on their own computer, not have to use the company’s web app. Does CC-BY-SA protect against this since it requires that any derivative works also be released under the same license? Does it work like GPL in that regard?
In theory, a copyleft licence should work. The problem however is a) how are you going to find out and b) how are you going to enforce it?
So, CC-BY-SA, which would require AI training database to be copyleft…
I have no actual answer, but given the very messy state of AI legality right now, I imagine it could be a while before we’re even able to define everything well enough to establish a solid legal framework for this sort of thing.
That said, I’d be happy to be proven wrong - this is definitely an important idea for society moving forward.
I mean, is copyright not specifically designed (by the big corporations mind you) to default to not allowing content to be used unless permission is explicitly given by the rights holder? So shouldn’t the answer to whether any content can be used is a big NO unless the author or distributor specifically allows it to be used?
There’s this thing called fair use .
The usage is clearly limited as can be determined by size of trained materials versus size of models. I would argue the use is transformative enough, after all you got from text/image inputs to effectively a tool that can produce texts and images.
If I write a story too similar to a Disney movie I will get sued, yet this is okay? Wtf
I am not a lawyer. This is not a legal advice.
I just want to note you here that ideas (in creations) alone are not subject to copyright (see: Wikipedia). (Ideas involving a particular invention can be subject to patents but Disney movie is not an invention, so this is not the case).
This way, if you watch a movie, become inspired by it, and just take the idea of the story, and implement in your creation with different characters, then you should be fine. However, you cannot use the assets of that movie (original names, fashion, specific attributes of characters, scenery).
At the end of the day, for such things, you may want to consult your advocate (a.k.a. lawyer). Court practice makes DOs and DON’Ts much clearer.
I don’t think the legal protections will be effective to prevent AIs from being trained on your works, because the data sets used to train the models are scraped from art sharing websites and it won’t be possible to identify that your art was part of the training set, legally or not. A better way is to use a tool like Glaze, which modifies your artwork so it looks the same to a human viewer but introduces errors when fed to an AI model.
You might have more luck asking on https://opensource.stackexchange.com/. I’d certainly hope that somebody using data from an AI trained on that image should be required to give attribution or shouldn’t be allowed to use it if modification is not allowed.
I think one of the problems right now is the lack of a proper legal definition of what is AI doing with your material. A human learning how to create original work by reading your work would not be required to cite it. The question is why and how exactly is AI doing something different.
But if a human straight up copies someone else’s writing, that’s illegal. AI spits out word for word passages from training data all the time.
I worked in data collection for an AI project (in a specialized domain, so no text or picture), and pour lawyer guided us on two things: first, we needed explicit approval to use some data, and second, if someone retracted their approval, we could keep the data as long as we couldn’t “trace it back” to that person… I ended up leaving the project.
