The "no, actually, a human did it" is an interesting defense. How are you to prove one way or another short of finding system logs of an AI outputting the material?
Further, does copyright law really care what generate the material? Either the material is copyrighted or it isn't, and people aren't signing licenses about reproducing what they hear/interpret/etc (which, imo, would be a flagrant misapplication of copyright law).
That being said, I haven't researched how likeness laws extend beyond the person's life.
Copyright law cares about what generated the material, because it protects certain rights of the creator of the work, and if you can't point to a human creator whose rights were violated (a machine or program has no legal rights) then no one has any exclusive rights to that work, it's outside the scope of copyright law, it's effectively "uncopyrightable". These rights can be transferred from the author to a corporation, for example, in employment contract or law; but a corporation can't simply claim them if they never existed in the first place.
This is kind of an interesting twist on a Turing test. If you can't tell by examining the work whether it was created by a human or an AI, then how can we say that a human is creative an an AI is not?
This would be the belief that the text/words are imbued with a metaphysical attribute while the text is created that depends on the properties of the author.
This is sort of the same umbrella belief that underpins the "words in the Bible are written by God" conclusion that some people have.
I'm guessing that you and I both aren't authorial essentialists, but as for a way people could believe this, it's not a stretch of the imagination.
As it relates to computers though this seems incredibly messy and a bad interpretation. Is my python script copyrighted, but the bytecode (generated by a computer) not? How do compiled programs have any copyright protection? Am I only violating a computer's copyright (so no copyright) if I crack DRM on a generated executable file[1]? Does the DMCA anti-circumvention protection only apply to hand-applied protection? When is a computer doing too much of the work? What about a Blender render? I think it would take me a hot minute to do those raytracing computations by hand. Do I only get the copyright to the output if I do that?
The most straightforward way to look at it is that computers and their programs are a tool, and their output as used by a person or person on behalf of a company. I think this current definition as I (NAL) understand it is unworkable and probably won't last long-term.
[1] Say for the sake of argument that there's no images or models or things that are obviously copyrighted, it's just a CLI calculator
The bytecode is not a novel creative work and does not have a separate copyright, this affects, for example, copyright terms (not that they are already relevant for code with the terms being so long) - the effective copyright date is when you wrote the source code, not when it was compiled; and of course the authorship - it does not matter who owns the machine or who initiated the compiling, only the writing of the original source code. But it is protected effectively as a "copy in a different format" of the original work - just as for a book, perhaps the author originally wrote as a manuscript with a quill on parchment or on a computer, and for the physical printed book it matters when and how the book was originally written (first 'fixed on a durable medium'), not what machines or tools were used in typesetting and printing the copy that you have - and the printed copy is still protected as a copy of that original creative work fixed in a manuscript.
In essence, as far as copyright law is concerned, the bytecode isn't generated by a tool, it's transformed by a tool.
Processes that transform creative work do not add creativity, but also do not erase it. So both your Python files and anything in the __pycache__ dir are copyrighted. More specifically, the __pycache__ is the same work in a different form, so it's the same copyright. You couldn't, say, dedicate the source to the public domain and then claim copyright over the binaries.
DMCA anti-circumention has to do with computer programs that enforce copyright. The work being protected has to be under copyright, but not the protection itself.
The only reason why AI art is completely uncopyrightable is because the only human contribution is a text prompt which is too short and functional to be copyrighted. Even if you're getting creative with your prompt writing, none of that creativity translates over into the final art - you're just keyword stuffing a small language model.
This argument is backwards. George Carlin is a human, and he's the one whose rights were presumably violated. It's the creator of the fake George Carlin stand-up special's job to prove their behavior was legal.
Saying that the AI George Carlin was human-written only means that, if creating your own George Carlin special is fair-use for some reason, then the human gets to claim copyright over that special. If it's AI then there's no copyright - at least, over the parts created by AI.
Just to make things even more confusing, this also has nothing to do with the copyright on the training set data. If the AI you used had a 100% licensed training set, you still have to worry about how the AI's legally uncreative[0] contribution to the work dilutes your ownership of it.
[0] As in, whether or not AI is creative, as a matter of law, it's not creative.
> How are you to prove one way or another short of finding system logs of an AI outputting the material?
I imagine the defense is the same as any other defense of "no, it was not me, it was someone other than me", you must establish your presence or activity at the time, or if they're the defendant, establish enough doubt to the prosecution's position.
In this case I don't think it is terribly pertinent whether or not the content was AI generated. The issue here is someone is trying to capitalize on the legacy of George Carlin. Admitting that it is human-written does not change that aspect.
I'm not an expert in the entire IP landscape, but I suspect this might not be a copyright issue at all. The problem is the brand (how George would hate that phrasing) is being appropriated. If you paid an artist to draw a picture of Keanu eating donuts for an advertising campaign there would be the same issue.
If anything this ends up being a partial defense of AI, given that the issues raised are neither specific to AI nor require additional laws to seek redress. It is hard to determine how much of the article text
>That training would, by definition, involve making "unauthorized copies" of "Carlin's original, copyrighted routines"
...is a quote from the lawsuit. It is certainly not clear in a moral or legal sense whether training constitutes making an unauthorized copy. That is the big issue to be decided. Having an opinion and declaring it does not make it true. People are doing that on both sides already.
It should be noted that in addition to the legal issues, comedians have their own culture which is separate from the legal definitions of acceptable behaviour. The tolerance for reusing ideas is considerably lower within that community. Many things that are perfectly legal might render you a pariah to that community.
> The issue here is someone is trying to capitalize on the legacy of George Carlin.
Personally, if someone creates an entertaining video that successfully captures the spirit of George Carlin, or any deceased speaker or artist, that sounds awesome. I'm all for it. I'd also love to have another Led Zeppelin album.
How would you suggest someone creates this ethically?
That was the whole gimmick here, though. They pitched it as "AI-generated George Carlin". They used his likeness for the thumbnail. They faked his voice.
Nothing is stopping you from writing a novel in the style of Harry Potter, but if you put "Harry Potter" in the title and use the same typeface, you're probably gonna get a call from a lawyer, because you're clearly trying to capitalize on existing IP instead of letting your work stand on its own merits. Is that wrong?
There's plenty of people bashing the concept of intellectual property here because Disney or whatever, but most artists aren't fat cats, and copyright protects their livelihoods without forcing them to rely on state sponsorship or private patronage - both of which are pretty problematic for artistic expression.
Honestly I don't think so. This is one of the things that's always bothered me about copyright, that for the entire history of humanity storytelling has been a place where practitioners build on one another and add their own takes and ideas. Stories are shared and become part of people's cultural and personal identities and I don't think they should be able to be owned to the degree they are today. Copying your Phantom Menace DVD I think is totally fine to be illegal but making your own take on it, making art based on it, or telling other stories the same universe I think should be allowed.
I would require compulsory licensing and percentage fee remittance for "covers" or "remixes" of someone else's work. Like if you remade Phantom Menace, made Muppets Phantom Menace or novelized
it. But what's called fanfic today should be able to be published above board.
Sure, but you're not giving enough attention to the fact that this is all happening against the backdrop of capitalism. It's really NICE when stories can be just freely shared and remixed and built from, and we're all richer for that, culturally, sure.
But artists and authors and comedians have to earn a living! If we want their output to be free, we can't expect them to also have to pay all for their inputs and, you know, keep making art!
If we judge art to have value under capitalism, there NEEDS to be a way of protecting that value for those who create it. Alternatively, we could explore alternatives to this whole capitalism thing, but you can't have your cake and eat it, too.
Right, which is why I'm saying that artists should retain exclusive control of the actual works that they produce and receive mandated royalties from anything that "covers" it (i.e. is substantively the same). Movie adaptations would fall under the category of covers.
Where I think it gets murky is when you try to claim ownership over the characters and universe. Because that prevents anyone who's inspired from sharing and selling their work. They either have to come up with their entirely unique OC or do what a lot of authors do which is pick a universe and cast that's in the public domain which is why we get so many greek god adaptations, Camelot, vampires, zombies. The creativity that would flourish if "Harry Potter" could be used with the freedom that people use Count Dracula would be incredible.
You're correct in this particular case, but I want to point out that copyright only protects fat-cat artists. OpenAI and StabilityAI thought training on web text and images was OK, but restricted themselves to public domain & licensable audio, because the music industry is perfectly willing to bleed millions keeping you from kidnapping what they've rightfully stolen. Visual artists and writers generally own their own works, unlike musicians, which means they're on the hook for individual enforcement to the tune of 5-6 figures per lawsuit. There's no mechanism that would allow for someone - even the government - to just step in and enforce copyright over the entire training set.
If, say, someone started writing Ruffle PRs with ChatGPT and filing them as made by "AI kmeisthax", that'd probably be a trademark or publicity rights violation. I'd be furious - but I wouldn't be able to stop them. Not without paying a potentially life-ruining amount of money to pursue a marginal claim in court against some anonymous J.Doe.
Likewise, there's shittons of people who steal art and writing, reselling it on T-shirt stores and Kindle without much in the way of legal recourse. Those are often legitimately lost sales[0], too! And yet, despite having some of the strictest copyright laws in history, artists don't really have the means to shut them down beyond "dox yourself with a DMCA notice". They are starving despite copyright.
Private property is a thing that makes it harder to overconsume scarce resources. It doesn't protect against exploitation. What you want is collective bargaining and enforcement - i.e. a law specifically prohibiting misleading use of AI with rights of action for all parties.
[0] As opposed to, say, arguing that a BitTorrent download is a lost sale
Ask permission from whom? George Carlin has been dead for 15 years.
From the copyright inheritors? Personally, not a fan of copyright inheritance, so I'll pass, unless I'm legally obligated. But I think the law says parody is fair use, does it not? I guess we'll find out with this lawsuit.
> "It is certainly not clear in a moral or legal sense whether training constitutes making an unauthorized copy."
Another complicating factor here is to what extent the mainstream definition of "training" even applies to things like LLMs.
The mismatch between what "training" typically means vs. what it means very specifically in a ML context is the source of many problems in this discussion.
We have for example heard many arguments that a human listening to every Beatles album, and then writing a new song based heavily in the style of the Beatles, is very clearly legal. But it's not clear that the ML model "learns" or "trains" in any way that is analogous to the human mind.
Another problem that I think we're not really contending with is to what extent LLMs generalize and to what extent they memorize. The more memorization that occurs the more compelling the argument will be that this is to some extent copyright-violating.
It certainly doesn't help the generalization argument that many LLMs can be compelled fairly easily to reproduce parts of their training data verbatim.
> In this case I don't think it is terribly pertinent whether or not the content was AI generated. The issue here is someone is trying to capitalize on the legacy of George Carlin. Admitting that it is human-written does not change that aspect.
> It is hard to determine how much of the article text ...is a quote from the lawsuit. It is certainly not clear in a moral or legal sense whether training constitutes making an unauthorized copy. That is the big issue to be decided. Having an opinion and declaring it does not make it true. People are doing that on both sides already.
> Without authorization from Plaintiffs, or any right under law, Defendants have unlawfully used Plaintiffs’ copyrighted works for building and training a dataset for purposes of generating an output intended to mimic Plaintiffs’ copyrighted work (i.e., Carlin’s stand-up comedy).
So regardless of whether the court will agree with this (I don't think the details of how training of AI models works in relation to copyright law have been settled in case law yet), it is in fact highly pertinent to the lawsuit because there will be no copyright claim if an AI was not trained on Carlin's work.
That would still leave the two other claims though (violation of rights of publicity and deprivation of rights of publicity).
I saw that video on YouTube, and I thought it was also written in the description of the video that it was human written, and even human voiced, but then made sound like Carlin with AI.
Yeah, at the time it was pretty widely known that this was the case (I don't recall whether it was in the description specifically). There's no revelation or admission that suddenly occurred this week.
I guess I figured it was written by Kultgen with the help from ChatGPT? It would have been a noteworthy feat for a built-from-scratch-by-training-on-MADtv-episodes foundation model was capable of generating the Carlin video at the click of a button. That’s still obvious to the average person, isn’t it? Maybe not? (And is that officially the world we live in now?)
Anyhoo. Still fair use, and even more so. A joke doesn’t have to rise to a certain level of hilarity to deserve 1st A. protection.
> Anticipating potential free speech defenses, the lawsuit argues that the special "has no comedic or creative value absent its self-proclaimed connection with George Carlin” and that it doesn't “satirize him as a performer or offer an independent critique of society.”
How is commenting on current events not "an independent critique of society"?
George Carlin always credited his older brother Patrick for being the genius of the family, so if I were them I would have presented "George Carlin's Genius Brother For the First Time!" or something like that, but of course that would be wrong and disrespectful as well. Maybe more legally defensible.
29 comments
[ 4.4 ms ] story [ 69.4 ms ] threadFurther, does copyright law really care what generate the material? Either the material is copyrighted or it isn't, and people aren't signing licenses about reproducing what they hear/interpret/etc (which, imo, would be a flagrant misapplication of copyright law).
That being said, I haven't researched how likeness laws extend beyond the person's life.
This would be the belief that the text/words are imbued with a metaphysical attribute while the text is created that depends on the properties of the author.
This is sort of the same umbrella belief that underpins the "words in the Bible are written by God" conclusion that some people have.
I'm guessing that you and I both aren't authorial essentialists, but as for a way people could believe this, it's not a stretch of the imagination.
The most straightforward way to look at it is that computers and their programs are a tool, and their output as used by a person or person on behalf of a company. I think this current definition as I (NAL) understand it is unworkable and probably won't last long-term.
[1] Say for the sake of argument that there's no images or models or things that are obviously copyrighted, it's just a CLI calculator
In essence, as far as copyright law is concerned, the bytecode isn't generated by a tool, it's transformed by a tool.
DMCA anti-circumention has to do with computer programs that enforce copyright. The work being protected has to be under copyright, but not the protection itself.
The only reason why AI art is completely uncopyrightable is because the only human contribution is a text prompt which is too short and functional to be copyrighted. Even if you're getting creative with your prompt writing, none of that creativity translates over into the final art - you're just keyword stuffing a small language model.
Saying that the AI George Carlin was human-written only means that, if creating your own George Carlin special is fair-use for some reason, then the human gets to claim copyright over that special. If it's AI then there's no copyright - at least, over the parts created by AI.
Just to make things even more confusing, this also has nothing to do with the copyright on the training set data. If the AI you used had a 100% licensed training set, you still have to worry about how the AI's legally uncreative[0] contribution to the work dilutes your ownership of it.
[0] As in, whether or not AI is creative, as a matter of law, it's not creative.
I imagine the defense is the same as any other defense of "no, it was not me, it was someone other than me", you must establish your presence or activity at the time, or if they're the defendant, establish enough doubt to the prosecution's position.
Not a lawyer at all though.
"I did not shoot the sheriff, Your Honour. The Colt did it."
I'm not an expert in the entire IP landscape, but I suspect this might not be a copyright issue at all. The problem is the brand (how George would hate that phrasing) is being appropriated. If you paid an artist to draw a picture of Keanu eating donuts for an advertising campaign there would be the same issue.
If anything this ends up being a partial defense of AI, given that the issues raised are neither specific to AI nor require additional laws to seek redress. It is hard to determine how much of the article text
>That training would, by definition, involve making "unauthorized copies" of "Carlin's original, copyrighted routines"
...is a quote from the lawsuit. It is certainly not clear in a moral or legal sense whether training constitutes making an unauthorized copy. That is the big issue to be decided. Having an opinion and declaring it does not make it true. People are doing that on both sides already.
It should be noted that in addition to the legal issues, comedians have their own culture which is separate from the legal definitions of acceptable behaviour. The tolerance for reusing ideas is considerably lower within that community. Many things that are perfectly legal might render you a pariah to that community.
Personally, if someone creates an entertaining video that successfully captures the spirit of George Carlin, or any deceased speaker or artist, that sounds awesome. I'm all for it. I'd also love to have another Led Zeppelin album.
How would you suggest someone creates this ethically?
Nothing is stopping you from writing a novel in the style of Harry Potter, but if you put "Harry Potter" in the title and use the same typeface, you're probably gonna get a call from a lawyer, because you're clearly trying to capitalize on existing IP instead of letting your work stand on its own merits. Is that wrong?
There's plenty of people bashing the concept of intellectual property here because Disney or whatever, but most artists aren't fat cats, and copyright protects their livelihoods without forcing them to rely on state sponsorship or private patronage - both of which are pretty problematic for artistic expression.
Honestly I don't think so. This is one of the things that's always bothered me about copyright, that for the entire history of humanity storytelling has been a place where practitioners build on one another and add their own takes and ideas. Stories are shared and become part of people's cultural and personal identities and I don't think they should be able to be owned to the degree they are today. Copying your Phantom Menace DVD I think is totally fine to be illegal but making your own take on it, making art based on it, or telling other stories the same universe I think should be allowed.
I would require compulsory licensing and percentage fee remittance for "covers" or "remixes" of someone else's work. Like if you remade Phantom Menace, made Muppets Phantom Menace or novelized it. But what's called fanfic today should be able to be published above board.
But artists and authors and comedians have to earn a living! If we want their output to be free, we can't expect them to also have to pay all for their inputs and, you know, keep making art!
If we judge art to have value under capitalism, there NEEDS to be a way of protecting that value for those who create it. Alternatively, we could explore alternatives to this whole capitalism thing, but you can't have your cake and eat it, too.
Where I think it gets murky is when you try to claim ownership over the characters and universe. Because that prevents anyone who's inspired from sharing and selling their work. They either have to come up with their entirely unique OC or do what a lot of authors do which is pick a universe and cast that's in the public domain which is why we get so many greek god adaptations, Camelot, vampires, zombies. The creativity that would flourish if "Harry Potter" could be used with the freedom that people use Count Dracula would be incredible.
If, say, someone started writing Ruffle PRs with ChatGPT and filing them as made by "AI kmeisthax", that'd probably be a trademark or publicity rights violation. I'd be furious - but I wouldn't be able to stop them. Not without paying a potentially life-ruining amount of money to pursue a marginal claim in court against some anonymous J.Doe.
Likewise, there's shittons of people who steal art and writing, reselling it on T-shirt stores and Kindle without much in the way of legal recourse. Those are often legitimately lost sales[0], too! And yet, despite having some of the strictest copyright laws in history, artists don't really have the means to shut them down beyond "dox yourself with a DMCA notice". They are starving despite copyright.
Private property is a thing that makes it harder to overconsume scarce resources. It doesn't protect against exploitation. What you want is collective bargaining and enforcement - i.e. a law specifically prohibiting misleading use of AI with rights of action for all parties.
[0] As opposed to, say, arguing that a BitTorrent download is a lost sale
Step 1: Ask for permission.
Example: the hologram of 2pac at Coachella 2012. It was pitched to the rapper's estate, they reviewed the proposal and approved.
From the copyright inheritors? Personally, not a fan of copyright inheritance, so I'll pass, unless I'm legally obligated. But I think the law says parody is fair use, does it not? I guess we'll find out with this lawsuit.
his estate.
> Personally, not a fan of copyright inheritance, so I'll pass, unless I'm legally obligated
you are.
> But I think the law says parody is fair use, does it not?
this case isn't about parody. it's about using someone's likeness without permission.
Another complicating factor here is to what extent the mainstream definition of "training" even applies to things like LLMs.
The mismatch between what "training" typically means vs. what it means very specifically in a ML context is the source of many problems in this discussion.
We have for example heard many arguments that a human listening to every Beatles album, and then writing a new song based heavily in the style of the Beatles, is very clearly legal. But it's not clear that the ML model "learns" or "trains" in any way that is analogous to the human mind.
Another problem that I think we're not really contending with is to what extent LLMs generalize and to what extent they memorize. The more memorization that occurs the more compelling the argument will be that this is to some extent copyright-violating.
It certainly doesn't help the generalization argument that many LLMs can be compelled fairly easily to reproduce parts of their training data verbatim.
> It is hard to determine how much of the article text ...is a quote from the lawsuit. It is certainly not clear in a moral or legal sense whether training constitutes making an unauthorized copy. That is the big issue to be decided. Having an opinion and declaring it does not make it true. People are doing that on both sides already.
The article is paraphrasing one of the three claims asserted in the complaint filed by George Carlin's estate (https://embed.documentcloud.org/documents/24377438-govuscour...):
> Without authorization from Plaintiffs, or any right under law, Defendants have unlawfully used Plaintiffs’ copyrighted works for building and training a dataset for purposes of generating an output intended to mimic Plaintiffs’ copyrighted work (i.e., Carlin’s stand-up comedy).
So regardless of whether the court will agree with this (I don't think the details of how training of AI models works in relation to copyright law have been settled in case law yet), it is in fact highly pertinent to the lawsuit because there will be no copyright claim if an AI was not trained on Carlin's work.
That would still leave the two other claims though (violation of rights of publicity and deprivation of rights of publicity).
Anyhoo. Still fair use, and even more so. A joke doesn’t have to rise to a certain level of hilarity to deserve 1st A. protection.
Even when, especially when, the joke is on us.^0
0: https://www.npr.org/2022/10/04/1126773469/onion-supreme-cour...
How is commenting on current events not "an independent critique of society"?
https://variety.com/2022/film/news/patrick-carlin-dead-90-ge...