First of all, copyleft and uncopyrightable are entirely different things.
Second, if something is a derived work of a copyleft work, then either it must also be copyleft, or it’s simply infringement and entirely unusable. You’re suggesting that AI remixing can effectively “remove” the copyleft, but it would be entirely unjust (and more to the point, contrary to established legal precedent) for it to work that way.
At firstglance, if AI art is copyleft, there’s no reason to buy/license the original from anyone; just include their stuff in the model and tweak the prompts until it’s close enough. Voila, free art! As long as tweaking the model is cheaper than buying art, the AI industry wins.
It’s not that there’d be no reason to buy/license it for commercial use, it’s that it would be impossible to do so. Downstream users simply couldn’t legally use it at all – no matter how much or little they wanted to pay – unless they were willing to release their work as copyleft, too.
In other words, making* AI output copyleft maximizes freedom, but it’s hardly “free.” And that impossibly-high cost to those who would leech is why it would be a good thing.
(* Or rather, affirming it as such in court, since it’s already rightfully copyleft by virtue of having already used copyleft input. It wouldn’t be a change in status, but rather a recognition of what the status always was.)
AI art would be used in products that can be copyrighted in the first place, and not things like marketing/political campaigns or decor.
depending on the exact license agreement, you could use copylefted things in commercial products. The actual art can be free to reuse/share, but the rest of product may not be; things like illustrations in a book say (an analogy I drew up based on how Android works, commercial products based on a copylefted component).
Simple solution: all AI output is copyleft.
That’s already the case. Copyright is only possible for creative works of human authorship. By definition AI generation is uncopyrightable.
First of all, copyleft and uncopyrightable are entirely different things.
Second, if something is a derived work of a copyleft work, then either it must also be copyleft, or it’s simply infringement and entirely unusable. You’re suggesting that AI remixing can effectively “remove” the copyleft, but it would be entirely unjust (and more to the point, contrary to established legal precedent) for it to work that way.
At firstglance, if AI art is copyleft, there’s no reason to buy/license the original from anyone; just include their stuff in the model and tweak the prompts until it’s close enough. Voila, free art! As long as tweaking the model is cheaper than buying art, the AI industry wins.
It’s not that there’d be no reason to buy/license it for commercial use, it’s that it would be impossible to do so. Downstream users simply couldn’t legally use it at all – no matter how much or little they wanted to pay – unless they were willing to release their work as copyleft, too.
In other words, making* AI output copyleft maximizes freedom, but it’s hardly “free.” And that impossibly-high cost to those who would leech is why it would be a good thing.
(* Or rather, affirming it as such in court, since it’s already rightfully copyleft by virtue of having already used copyleft input. It wouldn’t be a change in status, but rather a recognition of what the status always was.)
I feel this assumes two things.
AI art would be used in products that can be copyrighted in the first place, and not things like marketing/political campaigns or decor.
depending on the exact license agreement, you could use copylefted things in commercial products. The actual art can be free to reuse/share, but the rest of product may not be; things like illustrations in a book say (an analogy I drew up based on how Android works, commercial products based on a copylefted component).