Suno and Udio could face damages of up to $150,000 per song allegedly infringed.
See full article...
See full article...
No, everyone will, as creative expression of all kinds will be excluded from the list of activities that is able to earn you a living. That will he outsourced to AI, so humans can concentrate better on the mindnumbing jobs that can't be automated yet.Can both sides lose?
What about the scenario that a capable opensourced version starts to make the rounds? Even though no corporations would profit from ot, it could lead to a situation where everyone is able to generate their own personal playlist of music in the styles they love. That is something that sounds like a great desirable thing to some, but I really wonder how that would play out. Loss of a collective shared culture, loss of the concept of scarcity in a creative context, loss of ANY kind of curation.The record companies are exploitative weasels, but so are the AI companies. In this case both sides are bad, but at least the record companies give a few pennies out of every dollar to the artists.
The AI companies will take the artists work, give them nothing, then flood the streaming services with cloned music that takes away what little the artists get now.
Maybe try injecting it, like that orange idiot suggested.Do I actually agree with the music industry lawyers on something?
God, I feel so dirty. Time to go do tequila shots and boil myself in bleach.
Fair use really needs to be established better. Until then as Fenixgoon pointed out, the lawyers win.
I fully agree. I do however wonder what the license would allow AI companies to do. I don’t think “use” is a description that works for anyone.Definitely on team They Both Suck
But that said, I think it's only ethical that if you profit off someone's content then you should pay for it. We'll see whether the law says training is copying but either way it's certainly use.
Will rightsholders license their content for training though? …
Compulsory licenses for AI training data could make AI model development economically impractical for small startups like Udio and Suno—and judging by the aforementioned open letter, many musical artists may applaud that potential outcome. But such a development would not preclude major labels from eventually developing their own AI music generators themselves, allowing only large corporations with deep pockets to control generative music tools for the foreseeable future.
See, that's easy. A verbatim copy is a copy, and thus a copyright violation. What's hard is when the output is not a verbatim copy. There's also the argument that the dataset itself is a copy, but there's previous case law about Google that it's not, or at least is fair use.I fully agree. I do however wonder what the license would allow AI companies to do. I don’t think “use” is a description that works for anyone.
Would they allow AI’s te be trained on the music? And what if that AI generates an almost verbatim copy of an existing work? Who would own that?
So it’s not only the question of whether they would license - for enough money they would - but what that license will allow.
Something tells me that the wording of the license will be such that rightsholders don’t need to pay the artists under contract though.
They are going to have to show that the LLMs have reproduced, verbatim, a couple bars of a melody line of an existing copyrighted song.
Because putting existing songs in a blender and coming up with your own version is what musicians have been doing for thousands of years, and there have already been many many lawsuits predating LLMs that allege that some musician 'appropriated' some other musician's song. And at that point it's come down to 'oh this one note makes it totally different, maaaaan' in jury trials. See https://en.wikipedia.org/wiki/List_of_songs_subject_to_plagiarism_disputes for an incomplete list.
But! Since LLMs are happy to reproduce, verbatim, a couple bars of an existing copyrighted song if you ask it to, they probably won't have far to look?
Here's the thing, Musicians who release their music publicly consented to other humans doing this, they did not consent to machines doing this. Your ability to understand the difference that musicians do actually see is irrelevant to the reality that they see it as different and they did not consent to this difference.what musicians have been doing for thousands of years
See, that's easy. A verbatim copy is a copy, and thus a copyright violation. What's hard is when the output is not a verbatim copy. There's also the argument that the dataset itself is a copy, but there's previous case law about Google that it's not, or at least is fair use.
I'm starting to develop something of a growing aversion against techheads who insist on treating machine learning processes the same as fellow human beings. It's thrown around so casually, and with such arrogant confidence and a patina of unassailable logic. But this is not about some law of nature or mathematics, or inevitable chain of events that we have to understand or accept. This is more akin to going against the robber barons by calling forunions etc. Recognizing that we're badly in need of constructing defensive walls against developments that are going to set back the human condition for huge swats of people. Maybe OTHER people right now,, but this is just the start.Human songwriters, composers, and musicians also train by learning existing (often copyrighted) music. Assuming the music was purchased legally, they are not infringing, unless they also produce music that is similar enough to be considered a derivative work (or a straight copy).
The same should be true for AIs. Training on (legally purchased) copyrighted works should not be a copyright violation. Copyright suits should only be entertained if the AI actually spits out something that's a copy, or at least highly derived from another work.
I don't trust current AIs enough to believe they don't or won't do that, but trying to limit training is a vast expansion of copyright, that will ultimately end up being used against human creators as well. Let the rights holders sue for AIs creating infringing works, not just because a computer "listened" to copyrighted works and then remembered some metadata about it.
We'll call it Weird AI.The solution is obvious. They need to retrain their models so that they'll only produce satire. Problem solved.
I believe the flaw in your ointment is equating human-derived works and AI-derived works.Human songwriters, composers, and musicians also train by learning existing (often copyrighted) music. Assuming the music was purchased legally, they are not infringing, unless they also produce music that is similar enough to be considered a derivative work (or a straight copy).
The same should be true for AIs. Training on (legally purchased) copyrighted works should not be a copyright violation. Copyright suits should only be entertained if the AI actually spits out something that's a copy, or at least highly derived from another work.
I don't trust current AIs enough to believe they don't or won't do that, but trying to limit training is a vast expansion of copyright, that will ultimately end up being used against human creators as well. Let the rights holders sue for AIs creating infringing works, not just because a computer "listened" to copyrighted works and then remembered some metadata about it.
This!They are going to have to show that the LLMs have reproduced, verbatim, a couple bars of a melody line of an existing copyrighted song.
Because putting existing songs in a blender and coming up with your own version is what musicians have been doing for thousands of years, and there have already been many many lawsuits predating LLMs that allege that some musician 'appropriated' some other musician's song. And at that point it's come down to 'oh this one note makes it totally different, maaaaan' in jury trials. See https://en.wikipedia.org/wiki/List_of_songs_subject_to_plagiarism_disputes for an incomplete list.
But! Since LLMs are happy to reproduce, verbatim, a couple bars of an existing copyrighted song if you ask it to, they probably won't have far to look?
I think we are essentially saying the same thing: it is the output that could be problematic. But that is separate from the training of the model. Copyright law is already stretched enough and shouldn't be deformed significantly to cover training. If the record labels can show infringing output that should be the basis of the suit, not the training element.I believe the flaw in your ointment is equating human-derived works and AI-derived works.
A human has the ability to create original works based on what came before. An AI can only create derivative works, because it only knows what came before.
There's some cross-over between the human and the AI because humans can create derivative works quite easily, of course. But that doesn't mean AI's can create original works, too. Regardless of what it is, an AI will always have elements of what came before. A human can come up with new things the AI can't do.
I also tend to give humans more credit because of the effort they put into their creations. Call it "soul" if you will. AI-derived works are soulless imitations.
Objectively, one might argue there's no difference. And that's the problem.
Human art and endeavors are not soulless. An AI will never feel the consequences of their creation. As a kind of artist, and certainly someone with a lot of experience making things, that makes a huge difference in my respect for what people do, versus my complete disdain for what an AI does.
Even leaving aside the question of whether AI is essentially doing that, it's a delicious irony that these music rights companies are profiting from people who have done exactly this.Humor me here; What makes this any different from me listening to the radio for my entire life, and writing a rock song that would no doubt be inspired by all of the songs I've listened to over the years regardless of whether or not it was intentional?