There’s been this weird idea lately, even among people who used to recognize that copyright only empowers the largest gatekeepers, that in the AI world we have to magically flip the script on copyr…
“Let companies rip off your work, or else only Big Tech will be able to rip off your work”
Maybe we’re so far in capitalist hellholle that we simply consider everything to be for sale. What about GPL work that OpenAI steals? Or personal data? With how secretive they are with data they “scraped” we don’t even know if they have any right at all to repackage and sell it.
Making only big companies able to “rip off your work” (not an accurate representation, but whatever) Is not the solution you think it is.
The only solution is to force all models trained on public data to not be covered by copyrights by default. Any output from those models should also by default be in the commons. The solution is to avoid copyright cartels, not strengthen them.
IMO, we need to ask: What benefits the people? or What is in the public interest?
That should be the only thing of importance. That’s probably controversial. Some will call it socialism. It is pretty much how the US Constitution sees it, though.
Maybe you agree with this. But when you talk about “models trained on public data” you are basically thinking in terms of property rights, and not in terms of the public benefit.
Agreed that interim solution should be to make all “AI” work public domain since it treats everything it trains on as public domain. I’m for it because it would would immediately stop being profitable for commercial enterprises. Then check who they ripped off and settle any financial claims and damages before moving on to establish license for already created output.
Exactly. Make ALL output public domain. Force them to release their training sets. Force them to open source their models.
There will still be companies like Adobe and DeviantArt who will be able to work around this due to their ToS, but we have enough existing models to make them obsolete due to the power of FOSS.
Making all ai work public domain is a great idea… until you start trying to draft actual laws. If ai is only used to make the eyeballs of character is it public domain? If I use a stable diffusion base but then fine-tune it on my own work is it public domain? What if I use ai to make the general idea, then I use that as inspiration to make my own work? How does anyone prove that anything is or isn’t ai generated or assisted? The list goes on and on. Making laws about ai use in art simply isn’t realistic, they are just too hard to nail down, and too easy to skirt. I don’t know what the solution is, but it isn’t this unfortunately.
The other big problem with it is that it just means that few big companies who already own almost all the IP(yes, most professional artists don’t actually own their own work) just make their own models with their own work and are able to enjoy the benefits of AI while any small group just has yet another disadvantage. It will probably be these big companies pushing for anti-AI/“pro artist” laws.
Of course things are messy. I still think it’s the best option. I would say that yes, a character with AI eyes, would be public domain. Treat it like the GPL. If a small part of your code is GPL, all of your code has to be GPL.
Likewise, it isn’t easy to prove, people will get away with it doing in very small quantities and sufficiently reworking it, but extravagant examples would be caught, like serial plagiarists eventually are. The resulting loss in credibility could end careers. Of course, the best approach would be to completely remove copyrights altogether, then this wouldn’t be an issue at all.
Fair enough. What’s your stance on this - should someone be allowed to create a text prompt and a list of settings for a specific model and then sell that data that they 100% created themself?
I haven’t heard anyone saying they think people should not be allowed their sell their own text creation like this, but if they are allowed to, then it means that anyone who wants to sell AI art just needs to sell the instructions for someone else to create the art themself. This could easily be set up as a file format that the purchaser then just has to run on their own. Seems like a waste of energy for everyone to generate their own copy of the work, but I can’t imagine any laws being set up that say people are not allowed to sell their own creations because the purchaser may plug what they created into an AI.
Should this be allowed or should the law extend to people not being allowed to sell text that may be used by someone else to create art?
Is this because of your general anti-copyright stance, or is this specific to people selling things that some people think are likely to be used to make AI artwork? I mean, are you saying that anyone who makes anything should be allowed to sell what they make and anyone should be allowed to share it for free?
What I am getting at is that you said anything made by AI should be in the public domain, so should prompts that a person rights (100% on their own) be considered “AI art” because they are likely to be turned into AI art? Or do you just think there is nothing special about AI art and all of everyone’s work should be in the public domain?
It would be interesting if we end up with lawyers in court arguing over whether or not something would make a good enough AI art prompt to be it in or out of the public domain.
The models (ie the weights specifically) may not be copyrightable, anyways. There’s no copyright on the result of number crunching. Once the model is further fine-tuned, there might be copyright, but it’s still unlike anything covered by copyright in the past.
One analogy I have is a 3D engine. The engineers design the look of the typical output by setting parameters, but that does not create a specific copyright on the parameters. There’s copyright on the design documents, the code, the UI, if any and maybe other stuff. It’s not quite the same, though.
Some jurisdictions have IP on databases. I think that would cover AI models. If I am right, then that means that any license agreements that come with models are ineffective in the US.
However, to copy these models, you first need to get your hands on them. They are still trade secrets, so don’t on leaks.
I’ll lift a comment from techdirt:
Maybe we’re so far in capitalist hellholle that we simply consider everything to be for sale. What about GPL work that OpenAI steals? Or personal data? With how secretive they are with data they “scraped” we don’t even know if they have any right at all to repackage and sell it.
Making only big companies able to “rip off your work” (not an accurate representation, but whatever) Is not the solution you think it is.
The only solution is to force all models trained on public data to not be covered by copyrights by default. Any output from those models should also by default be in the commons. The solution is to avoid copyright cartels, not strengthen them.
IMO, we need to ask: What benefits the people? or What is in the public interest?
That should be the only thing of importance. That’s probably controversial. Some will call it socialism. It is pretty much how the US Constitution sees it, though.
Maybe you agree with this. But when you talk about “models trained on public data” you are basically thinking in terms of property rights, and not in terms of the public benefit.
Well, I think that removing copyrights altogether is in the public interest, so…there you go :)
Agreed that interim solution should be to make all “AI” work public domain since it treats everything it trains on as public domain. I’m for it because it would would immediately stop being profitable for commercial enterprises. Then check who they ripped off and settle any financial claims and damages before moving on to establish license for already created output.
Exactly. Make ALL output public domain. Force them to release their training sets. Force them to open source their models.
There will still be companies like Adobe and DeviantArt who will be able to work around this due to their ToS, but we have enough existing models to make them obsolete due to the power of FOSS.
Making all ai work public domain is a great idea… until you start trying to draft actual laws. If ai is only used to make the eyeballs of character is it public domain? If I use a stable diffusion base but then fine-tune it on my own work is it public domain? What if I use ai to make the general idea, then I use that as inspiration to make my own work? How does anyone prove that anything is or isn’t ai generated or assisted? The list goes on and on. Making laws about ai use in art simply isn’t realistic, they are just too hard to nail down, and too easy to skirt. I don’t know what the solution is, but it isn’t this unfortunately.
The other big problem with it is that it just means that few big companies who already own almost all the IP(yes, most professional artists don’t actually own their own work) just make their own models with their own work and are able to enjoy the benefits of AI while any small group just has yet another disadvantage. It will probably be these big companies pushing for anti-AI/“pro artist” laws.
Of course things are messy. I still think it’s the best option. I would say that yes, a character with AI eyes, would be public domain. Treat it like the GPL. If a small part of your code is GPL, all of your code has to be GPL.
Likewise, it isn’t easy to prove, people will get away with it doing in very small quantities and sufficiently reworking it, but extravagant examples would be caught, like serial plagiarists eventually are. The resulting loss in credibility could end careers. Of course, the best approach would be to completely remove copyrights altogether, then this wouldn’t be an issue at all.
Fair enough. What’s your stance on this - should someone be allowed to create a text prompt and a list of settings for a specific model and then sell that data that they 100% created themself?
I haven’t heard anyone saying they think people should not be allowed their sell their own text creation like this, but if they are allowed to, then it means that anyone who wants to sell AI art just needs to sell the instructions for someone else to create the art themself. This could easily be set up as a file format that the purchaser then just has to run on their own. Seems like a waste of energy for everyone to generate their own copy of the work, but I can’t imagine any laws being set up that say people are not allowed to sell their own creations because the purchaser may plug what they created into an AI.
Should this be allowed or should the law extend to people not being allowed to sell text that may be used by someone else to create art?
They can sell it all they want, and then the buyer should be able to share it for free. I’m OK with people selling their labor.
Is this because of your general anti-copyright stance, or is this specific to people selling things that some people think are likely to be used to make AI artwork? I mean, are you saying that anyone who makes anything should be allowed to sell what they make and anyone should be allowed to share it for free?
What I am getting at is that you said anything made by AI should be in the public domain, so should prompts that a person rights (100% on their own) be considered “AI art” because they are likely to be turned into AI art? Or do you just think there is nothing special about AI art and all of everyone’s work should be in the public domain?
It would be interesting if we end up with lawyers in court arguing over whether or not something would make a good enough AI art prompt to be it in or out of the public domain.
(I edited my comment slightly due to my scatter brain then saw you basically expanding my thought in the same way)
The models (ie the weights specifically) may not be copyrightable, anyways. There’s no copyright on the result of number crunching. Once the model is further fine-tuned, there might be copyright, but it’s still unlike anything covered by copyright in the past.
One analogy I have is a 3D engine. The engineers design the look of the typical output by setting parameters, but that does not create a specific copyright on the parameters. There’s copyright on the design documents, the code, the UI, if any and maybe other stuff. It’s not quite the same, though.
Some jurisdictions have IP on databases. I think that would cover AI models. If I am right, then that means that any license agreements that come with models are ineffective in the US.
However, to copy these models, you first need to get your hands on them. They are still trade secrets, so don’t on leaks.
That’s how it is now.
No, the models is not in the commons. Their training data is also not known.