I fucked with the title a bit. What i linked to was actually a mastodon post linking to an actual thing. but in my defense, i found it because cory doctorow boosted it, so, in a way, i am providing the original source here.
please argue. please do not remove.
I think we should have a rule that says if a LLM company invokes fair use on the training inputs then the outputs are public domain.
That’s already been ruled on once.
A recent lawsuit challenged the human-authorship requirement in the context of works purportedly “authored” by AI. In June 2022, Stephen Thaler sued the Copyright Office for denying his application to register a visual artwork that he claims was authored “autonomously” by an AI program called the Creativity Machine. Dr. Thaler argued that human authorship is not required by the Copyright Act. On August 18, 2023, a federal district court granted summary judgment in favor of the Copyright Office. The court held that “human authorship is an essential part of a valid copyright claim,” reasoning that only human authors need copyright as an incentive to create works. Dr. Thaler has stated that he plans to appeal the decision.
Why would companies care about copyright of the output? The value is in the tool to create it. The whole issue to me revolves around the AI company profiting on it’s service. A service built on a massive library of copyrighted works. It seems clear to me, a large portion of their revenue should go equally to the owners of the works in their database.
You can still copyright AI works, you just can’t name an AI as the author.
That’s just saying you can claim copyright if you lie about authorship. The problem then is, you may step into the realm of fraud.
You don’t have to lie about authorship. You should read the guidance.
Well, what you initially said sounded like fraud, but the incredibly long page indeed doesn’t talk about fraud. However, it also seems a bit vague. What counts as your contributions to the work? Is it part of the input the model was trained on, “I wrote the prompt”, or making additionally changes based on the result?
The vagueness surrounding contributions is particularly troubling. Without clearer guidelines, this seems like a recipe for lawsuits.
The outputs are not copyrightable.
But something not being copyrightable doesn’t necessarily mean openly distributed.
It does mean OpenAI can’t really restrict or go after other companies training off of GPT-4 outputs though, which is occurring broadly.
Not just the outputs but the models as well
Google scanned millions of books and made them available online. Courts ruled that was fair use because the purpose and interface didn’t lend itself to actually reading the books in Google books, but just searching them for information. If that is fair use, then I don’t see how training an LLM (which doesn’t retain the exact copy of the training data at least in the vast majority of cases) isn’t fair use. You aren’t going to get an argument from me.
I think most people who will disagree are reflexively anti AI, and that’s fine. But I just haven’t heard a good argument that AI training isn’t fair use.
here’s a sidechannel attack on your position: every use, even infringing uses, are fair use until adjudicated, because what fair use means is that a court has agreed that your infringing use is allowed. so of course ai training (broadly) is always fair use. but particular instances of ai training may be found to not be fair use, and so we can’t be sure that you are always going to be right (for the specific ai models that may come into question legally).
“Its perfectly legal unless you get caught!”
Considering most copyright cases come down to the individual judge’s decision, essentially yes
I am no lawyer, but I suspect what will be considered either fair use or infringing will probably depend on how the programmed AI model is used.
For example, if you train it on a book of poetry, asking it questions about the poetry will probably be considered fair use. If you ask the AI to write poetry in the style of the book’s poems and you publish the AI’s poetry, I suspect it might be considered laundering copyright and infringing. Especially if it is substantially similar to specific poems in the book.
If you ask the AI to write poetry in the style of the book’s poems and you publish the AI’s poetry, I suspect it might be considered laundering copyright and infringing.
is the image of a cabin in a snowy landscape copyrighted by Thomas kinkade? fuck no. That’s an idea. ideas can’t be copyrighted. a style isn’t a discreet work. it is an idea. it can’t be copyrighted. if I produce something in the style of Keats or Stephen King or Rowling, they can’t sue me for copyright unless I make a substantially infringing use of their work. The style isn’t sufficient, because the style can’t be copyrighted.
What constitutes fair use?
17 U.S.C. § 107
Notwithstanding the provisions of sections 17 U.S.C. § 106 and 17 U.S.C. § 106A, the fair use of a copyrighted work, including such use by reproduction in copies or phonorecords or by any other means specified by that section, for purposes such as criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research, is not an infringement of copyright.
GenAI training, at least regarding art, is neither criticism, comment, news reporting scholarship, nor research.
AI training is not done by scientists but engineers of a corporative entity with a long term profit goal.
So, by elimination, we can conclude that none of the purposes covered by the fair use doctrine apply to Generative AI training.
Q.E.D.
“Such as” means that these are examples and not an exhaustive list.
Can you explain how the 3 factors you listed rule out scholarship or research purpose? Regarding the first factor, how do you determine that AI developers are all engineers and never computer scientists?
I’d argue that the community benefit aspect of the “scholarship or research purposes”language preclude for-profit AI companies from falling under fair use. These aren’t education programs. They’re not research for the greater good. They are private entities trying to create a machine that can copy until it creates. For their own needs, not the greater good. Education has a net positive effect on society, and those stipulations in the law are meant to better serve the whole.
If these generative AI machines were being built by students, it would fall under these specifications of fair use. But the profit motive changes everything.
I’d say “fair use” pretty much covers educational and community benefit. Private companies do neither. They are stealing and reproducing for themselves, not society.
I’d argue that the community benefit aspect of the “scholarship or research purposes”language
How do you get the “community benefit aspect” out of that? Also, why do feel that a profit motive is at odds with the greater good? That seems to run counter to the whole conception of US copyright. The other examples are mainly produced with a profit motive.
Okay, first of all, that was my interpretation. Because “teaching” has always been tied to education. I was extrapolating the point to argue that fair use laws are there for the sake of education and cultural growth. You can use copyrighted works for use that benefits society as a whole, I.e. education. See what I’m saying? Fair use laws were written with the benefit of all in mind, using established works to broaden education and knowledge in the community and for purposes of culture. That’s my interpretation of their entire purpose.
Oh, I actually was just talking of my own interpretation of the point of he laws, but this is from copyright.gov:
Under the fair use doctrine of the U.S. copyright statute, it is permissible to use limited portions of a work including quotes, for purposes such as commentary, criticism, news reporting, and scholarly reports.
They’re saying right there the purpose if for news, discussion, and education. Cultural benefits. That proves my point, I think.
But onto this:
why do you feel that a profit motive is at odds with the greater good?
Because…it is. Profit is extremely limited to the entity at the top of the capitalist structure of business (on a case by case basis, I mean. Not the top of capitalism period.) “Profit” is what a business rakes in for itself. The entire concept of profit has exploitation written right into itself. All a company’s payroll is a cost and does not factor into profit. So literally if I pay my workers less, I profit more. If they’re starving? Even more profit. If I eliminate their jobs and outsource them so I can have less expenditure and more profit by making even poorer people work for even less? Boom. Fuck these workers, I can exploit and squeeze some other poor saps even harder for more profit.
“Profit” and “greater good” are diametrically opposed concepts. Profit is limited. Greater good is collective. It’s literally the entire problem with capitalism. Profit needs exploitation. The more you exploit down the line, the more profit (read: the more people I can hurt and cut out of the money, the more profit I have). Capitalism is built on the profit motive and look where that’s led us. To a time with for-profit healthcare, sweatshops, slave labor…profit necessitates exploitation. The more you can take from the greater population—whether in price paid to you or cost cut at the expense of everyone possible—the more you profit.
Like I said, profit motive is almost the exact opposite of doing something for the greater good.
They’re saying right there the purpose if for news, discussion, and education. Cultural benefits. That proves my point, I think.
News media is usually for-profit, though. Commentary and criticism is also a staple of for-profit media. That ordinary people can and do publish their own takes via the internet is much more recent than section 107.
Much of medical research is for-profit. Biontech is a for-profit company, but their covid vaccine benefits the public.
I agree that the public benefit aspect is there, but I’d go higher to find it, right to the constitution. Congress is empowered To promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries. It’s a fascinating turn of phrase. Congress is not quite empowered to make copyright law (limiting the freedom of the press). It is empowered to promote progress through certain means.
The whole idea is that one can serve the greater good by introducing a profit motive to the production of, among other things, creative works. Without copyright, everything would be public domain.
Come to think of it, it is kind of weird how you apply your moral views on profit to this fair use issue. You’re saying that copyright owners should make a profit. If it’s not fair use, then the copyright owners have to be paid, right? That means that, EG, newspapers, like the NYT, can demand money for training on its archive. That’s all paid off. The production cost has been recouped (or not). Any licensing payments now are pure profit. AI developers still have to put in the work to actually develop the AI.
Some good points here.
And I’d argue that the for-profit aspect of every single one of those institutions has corrupted and degraded the purpose and quality of each. For-profit news turned what was once a public service into what we have today: agenda-driven corporations tarnishing information for their own ends. Universities driving kids into lifelong debt. And in the case of the Covid vaccine, they took public funds and then privatized the medicine for profit.
Profit corrupted every single one of these fields.
Im not saying that a profit motive absolutely negates any positive outcome. But eliminating the profit motive eliminates selfishness. Profit is the end goal. And think about any example you can in which something good came out of a company’s desire for profit. Any example has immediate diminishing returns because while putting a new vaccine, say, onto the market that was driven by a company’s profit motive immediately loses the benefit for the greater good because it’s not the end goal. The end goal is profit. So access for the poor is immediately out of reach. Because of profit.
The motivation for development might have been driven by profit, and new discoveries come about from a company’s r&d. Great. But immediately a problem occurs when access is limited to funds. So I see what you’re saying, capitalists love to say “competition spurs innovation,” but that only goes so far, if it’s even true in the first place.
And think about public development of anything—it’s immediately sold to the highest bidder and paywalled. How about Volvo and the three point seatbelt. Did profit motive drive the discovery of that feature? Presumably, to some degree. But they immediately made it accessible to all by eliminating the profit motive for the greater good. If hey had decided to patent it and only sell it for profit to other manufacturers, it’s a detriment to he greater good.
So again, I’m not saying that nothing good has ever been discovered or created via a profit motive, but I am saying that it corrupts the reader good by exploiting need for profit. See what I’m saying? So you’re not entirely wrong, but it’s a ethical philosophy question. When your motives are selfish/corrupt, your deeds aren’t good, even if good may come about. The motives are corrupt, so any good is nullified by said profit motive.
We can talk about what the world would look like today if humanity was always cooperative instead of implementing capitalism, what would the Industrial Revolution have looked like, etc. And maybe capitalism was, at some point, the best thing for humanity to progress. But it always should’ve been a stepping stone TO a system for the greater good. Instead, the profit motive has corrupted humanity and made a system that exploits everyone possible. Exploitation is rewarded under a system that places profit above everything else. They say you have to break a few eggs to make an omelette, and I think that maybe applies to capitalism’s place in human history. Maybe it was necessary to bring about progress in the early 20th century (although the robber barons/gilded age would suggest it was too great a price to pay), but I’d argue that, as it exists today, the profit motive is harmful and needs to be done away with. Because it runs contrary to the greater good. They are diametrically opposed.
You skipped right over “teaching”.
Why is that?
Show me an application of Generative AI for teaching right now. As in, already existing.
Not teaching with AI
Teaching AI
I think their point is the law is written to benefit people. Not private companies or machines.
If this wide definition of “teaching” were acceptable, then the entire concept would cease to exist.
“You stole my paper and reproduced it for profit!”
“NOO, I’m just teaching my employees to write better. It’ll happen eventually, but we’re at the stage where reproducing something incredibly similar to your paper is necessary!”
I agree that all this needs to be examined, and some new laws and regulations should be developed. But, for good or ill, teaching is a covered use as written in the section of the law quoted above, and teaching is part of the process of training.
If anything, laws will likely have to be rewritten to adress changing technologies, but it seems disingenuous to quote a section of the law and then ignore the most relevant word in the entire text
I definitely get your point. But you don’t “teach” a machine. You program a machine. In the case of AI, technically the machine is building its own database and sort of growing and adapting as it gets more advanced.
I get your point, but I just don’t think “teaching” is even what is happening here. Like I said, if the definition were that broad, it would be rendered meaningless. Not to mention, there are so, so, so many examples of the generative AI just reproducing something specifically in the style of a known artist. Writing in the style of a specific author. It does that because we ask it to, but the point is the program is a machine for reproduction. You don’t teach something without sentience. You teach living things, you write code and make a program act in a specific way. And right now, the programs are blatantly reproducing signature pieces of work.
Now, OP mentioned we are “teaching” the machines to do things on its own. But my point is that’s not teaching. It’s reproducing and stealing. It’s not creating anything, it’s spitting out elements of what it’s absorbed. And because these machines can’t think, can’t add their own style—because what’s super fucked up is we are pretty much just discussing the machines replacing artists at the moment—these things are about experience and personality. Neither of which AI has. They ingest everything and spit back out what we ask for. And they’re spitting out elements of this or that—and in these cases, it’s intellectual property of artists and writers. And the most depressing aspect of this whole thing is that we have pretty much moved beyond the “wait, out of everything, we are teaching machines to take human creativity and expression away from…humans?” stage and just moved on to talking about whether it’s technically legal.
I agree, laws will definitely have to be rewritten. But for the sake of argument, I don’t think the letter of the law can be as broad as you’re suggesting. Interesting thought experiment for us, though. Because…no one gives a shit about our takes on the matter lol
Or the take of artists and writers. But that’s a whole different problem.
it is pretty obviously scholarship and research
It is pretty obviously Research and Development of a commercial product in many cases. Not fair use.
So fair use if it’s an opensource model?
there is no stipulation that the research must be non-profit.
woosh
Selling an AI model (or usage of that model) that allows for producing works that are clearly based upon those copyrighted works and would be considered copyright infringement if a person did the same thing is not fair use.
If a person creating the same thing as generative AI would be infringing, then it isn’t magically not infringing because it is on the internet or done by a program. Basically, AI needs to follow the same rules and restrictions as a person would. That does mean that the AI also needs to be trained to not create copyright infringing works if the use of the AI is being sold.
As a downloadable model that anyone can use at no cost? Sure, whatever is fine. Then it is on the person who uses it and tries to infringe. But if someone pays a company to use their AI to create infringing work, that is on the company and they are just as at fault as if they sold T shirts that infringed on copyright.
If a person creating the same thing as generative AI would be infringing, then it isn’t magically not infringing because it is on the internet or done by a program
no one is arguing otherwise.
That does mean that the AI also needs to be trained to not create copyright infringing works if the use of the AI is being sold.
no it doesn’t.
Selling an AI model (or usage of that model) that allows for producing works that are clearly based upon those copyrighted works and would be considered copyright infringement if a person did the same thing is not fair use
it is.
I think you might want to elaborate
instead of making 4 replies in 3 minutes
each averaging
2.75 wordsI don’t see how selling a model or the use of a model infringes on a specific copyright. whose copyright has been infringed? how can you prove that? take AI out of the question. if you wanted to prove that some other author has infringed the copyright on your novel, how would you do that? if you want to prove that some quote unquote artist has infringed on your copyright, how would you do that? if any of your methods for proving that a person has infringed on your copyright is applicable to an AI, then that’s what that is. but if you can’t prove it, if the AI just learned about how style works, if an AI just saw your work but never actually copied it, then it’s not infringing.
instead of making 4 replies in 3 minutes
each averaging
2.75 words
this is irrelevant to the truth of my claim.
Yes, but at the time I wrote my reply there was no truth replied by you, only what can be summarized as “no”.
I presented exactly as much justification for my claims as the people to whom I was responding.
Again, right now, yes, but when I wrote that, no.
wrong.
I’m sorry, are you saying that selling a book that has the same characters as a recently released book doing the same things but with wording differences is somehow fair use? Like a book called Harry Potter and the Something Rock with the exact same plot points but worded slightly different is fair use?
Do you even understand what copyright is?
are you saying that selling a book that has the same characters as a recently released book doing the same things but with wording differences is somehow fair use? Like a book called Harry Potter and the Something Rock with the exact same plot points but worded slightly different is fair use?
no. I was saying selling an AI model or access to it that is capable of producing that work is not, itself, copyright infringement.
in fact, do you know what a clean room is? if I provided to a writing team every English language work except those written by JK Rowling and it produced a work exactly like you’re describing, The resultant work would not be infringing copyright. it should not be any different for AI where you cannot prove what materials it was provided.
Copyright doesn’t care if the writer is unaware of the source material because intent doesn’t matter.
intent does matter for fair use claims, and knowledge matters for bare infringement.
if someone pays a company to use their AI to create infringing work, that is on the company and they are just as at fault as if they sold T shirts that infringed on copyright.
wrong.
It isn’t fair use, See most of faq @ fairuse faq.
“Fair Use” is often the subject of discussion when talking about online copyright with regards to online video content or music sampling, but it’s notably a flawed defense as it generally has no legal definition for how much of certain content can be used or referenced. The very first line of that faq has the following note:
How do I get permission to use somebody else’s work?
You can ask for it. If you know who the copyright owner is, you may contact the owner directly. If you are not certain about the ownership or have other related questions, you may wish to request that the Copyright Office conduct a search of its records or you may search yourself. See the next question for more details.All artists / writers and others are asking LLM model producers to do is a) Ask for permission or B) Attribute the artists work in some kind of ledger, respecting the copyright of their work. Every work you make (write/play/draw/whatever) has a copyright that should be respected by companies and are not waived by EULA or TOS (ever) and must be respected in order for author attribution as a concept to work at all. There is plenty of free, permissive copyrighted content on the internet that can be used instead to train an LLM, but simply asking for permission or giving attribution would at least be a step in the right direction for these companies and for the industry as a whole.
Defenders of AI will note that the “use” of art in LLM is limited and thus protected by fair use, but that is debatable based on the content of the above listed FAQ.
How much of someone else’s work can I use without getting permission?
Under the fair use doctrine of the U.S. copyright statute, it is permissible to use limited portions of a work including quotes, for purposes such as commentary, criticism, news reporting, and scholarly reports. There are no legal rules permitting the use of a specific number of words, a certain number of musical notes, or percentage of a work. Whether a particular use qualifies as fair use depends on all the circumstances. See, Fair Use Index, and Circular 21, Reproductions of Copyrighted Works by Educators and Librarians.You can see that the use cases above (commentary, criticism, news reporting and scholarly reports) does not qualify LLM companies to use or train their models with copyrighted data for privatized industry. Additionally, you’ll note that “market disruptive” uses cannot be protected by fair use in it’s definition, meaning that displacing artists with AI automatically makes LLM use of copyrighted material an infraction of copyright that is not protected by the fair use clause.
Regardless, this will need to be proved in court and even if it passes certain criteria, it will not apply to all infractions. Fair use is a defense, not a protection, and thus LLM producers will have to spend time in court in order to defend individual infractions. There’s no way for them to catch all copyright infringement with one ruling, it needs to be proved on a case-by-case basis.
IANAL but this is my 2 cents on the matter.
this will need to be proved in court
this is true of all fair use. this is almost the definition of fair use. Fair use can only exist after a judge has adjudicated it. before it is questionable.
You can see that the use cases above (commentary, criticism, news reporting and scholarly reports) does not qualify LLM companies to use or train their models
Seems quite obvious that the text you quoted refers exclusively to plagiarism. This does not include things like being inspired by it, referencing it, parodying it and of course not training AI either, because what matters is whether the result is protected content.
You can argue that memorizing and sharing training data is a copyright violation, and that’s a fair point, but it’s also worth noting that this is very much a minority, accidental and is being addressed.
Here’s another good one: https://www.eff.org/deeplinks/2023/04/how-we-think-about-copyright-and-ai-art-0
deleted by creator
Ignore the petapixel article. Even the headline is false.
Fair Use in the US comes directly from the constitution. I don’t think other countries have anything quite like it. Also, in the English Law system much of the law is made by judges (case law). Japan’s system is a Civil Law system which seeks to keep law-making with the actual legislative bodies. Law-makers in these countries have to be much more on the ball.
About 15 years ago, law-makers in these countries began explicitly allowing AI training (actually, data-mining in general). In Japan this first happened in 2009 and was expanded in 2018.
It is an advantage for these countries that the matter is already cleared up by their functioning legislatures, but if there is something they didn’t think of (or lobbyists messed things up) then tough luck. The US still has its constitutional protections which means that any necessary corrections can still be made at “runtime”. I think the internet as we know it might not have been possible in any other country.
I agree with your statement as stated. (EDIT: title has changed since posting)
The part I have an issue with is the output that tends to fall into 4 categories:
-
output that is creative and unique is acceptable
-
output tricked into revealing source is infringement
-
output not creative, as in biometric identification, utilizes the entirety of copyright work in a non-transformative way and needs to be licensed.
-
output of creative work, while not copyrightable, can still infringe trademark.
-
Agreed. I would also argue that trained model weights are not copyrightable.
They aren’t.
Courts have already ruled that copyright requires human creation, and weights are not decided by humans but by the training algorithms.
I didn’t know it was already settled law. But in that case, why are models like llama still released under licenses? If they are non-copyrightable, licenses should be unenforceable and therefore irrelevant.
The license is related to access.
Basically it’s gated and not publicly available, and the only way to open the gate is to say “I promise not to do anything outside what you are limiting me to do.”
A second person that gets access without agreeing to that can use the weights however they want (what copyright would relate to), but the person who gave them access to the weights would have been in breach of their agreement.
So separate things with different scopes.
True. They are instead can be patented.
Do you mean “fair use” in the legal sense or the ethical sense?
in the ethical sense, everything is fair use. period.
in the legal sense, everything is fair use until it’s proven in court not to be.
I totally agree.
Copyright and patent laws need to die.
If and only if the trained model is accessible without licence.
E.g. I don’t want Amazon rolling out a Ilm for $100 a month based on freely accessible tutorials written by small developers.
But yeah duck copyright
in the ethical sense, everything is fair use. period.
So if someone spends decades of their life and millions of their own dollars to creating or researching something, in your opinion, you and everyone else is entitled to the fruits of their labor?
in the legal sense, everything is fair use until it’s proven in court not to be.
That makes zero sense. Just because a court hasn’t yet deemed that specific action illegal doesn’t mean it’s not illegal when you do it. Doesn’t matter if the crime is theft, rape, murder, etc.
if anybody gets a copy of it, they have no ethical obligation not to share it, and every ethical justification for sharing it.
Wow.
this reads like an appeal to ridicule. if you have an objection to what I said please state it.
Every web request costs someone money. If you aren’t paying them you are being provided a service. They’ve given you knowledge/ material in their possession free of charge. You are taking advantage of that good will by using the content for purposes not intended. That is a moral failing.
To be clear the ownership of the material is not important, just the access is immoral, as the harm is already done.
Ill add the caveat that it can be moral if they’ve specifically told you you can via the websites robot.txt file which websites of consequence all have. But the assumption has to be they don’t intend this because that is how consent works.
They’ve given you knowledge/ material in their possession free of charge.
this is a very common human activity
the assumption has to be they don’t intend this
why? if someone publishes something on port 80, why should I ever assume they mean anything but for me to have and use that data?
You are taking advantage of that good will by using the content for purposes not intended. That is a moral failing.
only if there were so e sort of agreement about what the acceptable uses are and what is not acceptable.
If you aren’t paying them you are being provided a service.
if you ARE paying them, you’re being provided a service, too
this reads like an appeal to ridicule.
I don’t know what this is supposed to mean. Speak plainly.
if you have an objection to what I said please state it.
I don’t even know where or how to begin arguing against a position that’s flawed on such a basic level.
an appeal to ridicule is also called a horse laugh fallacy. it’s like writing lol instead of actually explaining what’s wrong with the position to which your objecting. this response also reads like an appeal to ridicule. if you can’t explain what’s wrong with my position, maybe you shouldn’t be speaking about my position.
Just because a court hasn’t yet deemed that specific action illegal doesn’t mean it’s not illegal when you do it. Doesn’t matter if the crime is theft, rape, murder, etc.
theft rape and murder are criminal matters. copyright is civil, and, yes, the courts can adjudicate every individual case.
theft rape and murder are criminal matters. copyright is civil
It’s…a civil crime. Not sure what your point is.
the courts can adjudicate every individual case.
Just like theft, rape and murder…
Just like theft, rape and murder…
except that sometimes those are statutory. fair use claims cannot be statutory.
no such thing as a civil crime. you are thinking of a tort.
Removed by mod
please cite that for me, if you have 3 seconds.
Sure, that can be fair use, but only if using them can also be fair use
