GenAI tools ‘could not exist’ if firms are made to pay copyright::undefined
So they’re admitting that their entire business model requires them to break the law. Sounds like they shouldn’t exist.
It likely doesn’t break the law. You should check out this article by Kit Walsh, a senior staff attorney at the EFF, and this one by Katherine Klosek, the director of information policy and federal relations at the Association of Research Libraries.
Headlines like these let people assume that it’s illegal, rather than educate people on their rights.
The Kit Walsh article purposefully handwaves around a couple of issues that could present larger issues as law suits in this arena continue.
-
He says that due to the size of training data and the model, only a byte of data per image could be stored in any compressed format, but this assumes all training data is treated equally. It’s very possible certain image artifacts are compressed/stored in the weights more than other images.
-
These models don’t produce exact copies. Beyond the Getty issue, nytimes recently released an article about a near duplicate - https://www.nytimes.com/interactive/2024/01/25/business/ai-image-generators-openai-microsoft-midjourney-copyright.html.
I think some of the points he makes are valid, but they’re making a lot of assumptions about what is actually going on in these models which we either don’t know for certain or have evidence to the contrary.
I didn’t read Katherine’s article so maybe there is something more there.
She addresses both of those, actually. The Midjourney thing isn’t new, It’s the sign of a poorly trained model.
I’m not sure she does, just read the article and it focuses primarily what models can train on. However, the real meat of the issue, at least I think, with GenAI is what it produces.
For example, if I built a model that just spit out exact frames from “Space Jam”, I don’t think anyone would argue that would be a problem. The question is where is the line?
This part does:
It’s not surprising that the complaints don’t include examples of substantially similar images. Research regarding privacy concerns suggests it is unlikely it is that a diffusion-based model will produce outputs that closely resemble one of the inputs.
According to this research, there is a small chance that a diffusion model will store information that makes it possible to recreate something close to an image in its training data, provided that the image in question is duplicated many times during training. But the chances of an image in the training data set being duplicated in output, even from a prompt specifically designed to do just that, is literally less than one in a million.
The linked paper goes into more detail.
On the note of output, I think you’re responsible for infringing works, whether you used Photoshop, copy & paste, or a generative model. Also, specific instances will need to be evaluated individually, and there might be models that don’t qualify. Midjourney’s new model is so poorly trained that it’s downright easy to get these bad outputs.
This goes back to my previous comment of handwaving away the details. There is a model out there that clearly is reproducing copyrighted materials almost identically (nytimes article), we also have issues with models spitting out training data https://www.wired.com/story/chatgpt-poem-forever-security-roundup/. Clearly people studying these models don’t fully know what is actually possible.
Additionally, it only takes one instance to show that these models, in general, can and do have issues with regurgitating copyrighted data. Whether that passes the bar for legal consequences we’ll have to see, but i think it’s dangerous to take a couple of statements made by people who don’t seem to understand the unknowns in this space at face value.
The ultimate issue is that the models don’t encode the training data in any way that we historically have considered infringement of copyright. This is true for both transformer architectures (gpt) and diffusion ones (most image generators). From a lay perspective, it’s probably good and relatively accurate for our purposes to imagine the models themselves as enormous nets that learn vague, muddled, impressions of multiple portions of multiple pieces of the training data at arbitrary locations within the net. Now, this may still have IP implications for the outputs and here music copyright is pretty instructive, albeit very case-by-case. If a piece is too “inspired” by a particular previous work, even if it is not explicit copying it may still be regarded as infringement of copyright. But, like I said, this is very case specific and precedent cuts both ways on it.
The article dealt with Stable Diffusion, the only open model that allowed people to study it. If there were more problems with Stable Diffusion, we’d’ve heard of them by now. These are the critical solutions Open-source development offers here. By making AI accessible, we maximize public participation and understanding, foster responsible development, as well as prevent harmful control attempts.
As it stands, she was much better informed than you are and is an expert in law to boot. On the other hand, you’re making a sweeping generalization right into an appeal to ignorance. It’s dangerous to assert a proposition just because it has not been proven false.
-
Reproduction of copyrighted material would be breaking the law. Studying it and using it as reference when creating original content is not.
Reproduction of copyrighted material would be breaking the law. Studying it and using it as reference when creating original content is not.
I’m curious why we think otherwise when it is a student obtaining an unauthorized copy of a textbook to study, or researchers getting papers from sci-hub. Probably because it benefits corporations and they say so?
While I would like to be in a world where knowledge is free, this is apples and oranges.
OpenAI can purchase a textbook and read it. If their AI uses the knowledge gained to explain maths to an individual, without reproducing the original material, then there’s no issue.
The difference is the student in your example didn’t buy their textbook. Someone else bought it and reproduced the original for others to study from.
If OpenAI was pirating textbooks, that would be a wholly separate issue.
I was under the impression they mentioned at some point torrenting things
Don’t know about OpenAI, but Meta used pirated books to train its AI.
https://www.techspot.com/news/101507-meta-admits-using-pirated-books-train-ai-but.html
The fact that the “AI” can spit out whole passages verbatim when given the right prompts, suggests that there is a big problem here and they haven’t a clue how to fix it.
It’s not “learning” anything other than the probable order of words.
I really hate this reduction of gpt models. Is the model probabilistic? Absolutely. But it isn’t simply learning a comprehensible probability of words–it is generating a massively complex conditional probability sequence for words. Largely, humans might be said to do the same thing. We make a best guess at the sequence of words we decide to use based on conditional probabilities along a myriad number of conditions (including semantics of the thing we want to say).
Completely agree. And that should be the focal point of the issue.
Sam Altman is correctly stating that AI is not possible without using copyrighted materials. And I don’t think there’s anything wrong with that.
His mistake is not redirecting the conversation. He should be talking about the efforts they’re making to stop their machine from reproducing copyrighted works. Not whether or not they should be allowed to use it in the first place.
I agree that the issues
- whether AI output are derivative works of its input, and
- whether input to AI is fair use and requires no compensation
are separate, but I think they are related, in that AI companies are trying to impose whatever interpretation of copyright that is convenient to them to the rest of the society.
And indeed Meta pirated books to feed its AI.
https://www.techspot.com/news/101507-meta-admits-using-pirated-books-train-ai-but.html
humans studying it, is fair use.
So if a tool is involved, it’s no longer ok? So, people with glasses cannot consume copyrighted material?
No. A tool already makes it unnatural. /S
Copyright can only be granted to works created by a human, but I don’t know of any such restriction for fair use. Care to share a source explaining why you think only humans are able to use fair use as a defense for copyright infringement?
Because a human has to use talent+effort to make something that’s fair use. They adapt a product into something that while similar is noticeably different. AI will
-
make things that are not just similar but not noticeably different.
-
There’s not an effort in creation. There’s human thought behind a prompt but not on the AI following it.
-
If allowed to AI companies will basically copyright everything…
Your reply has nothing to do with fair use doctrine.
You are aware of the insane amounts of research, human effort and the type of human talent that is required to make a simple piece of software, let alone a complex artificial neural network model whose function is to try and solve whatever stuff…right?
Good point. I say the software can be copywrite protected, but not the content the program generates.
And that is human effort, not the AIs.
-
What’s the difference? Humans are just the intent suppliers, the rest of the art is mostly made possible by software, whether photoshop or stable diffusion.
I don’t agree. The publisher of the material does not get to dictate what it is used for. What are we protecting at the end of the day and why?
In the case of a textbook, someone worked hard to explain certain materials in a certain way to make the material easily digestible. They produced examples to explain concepts. Reproducing and disseminating that material would be unfair to the author who worked hard to produce it.
But the author does not have jurisdiction over the knowledge gained. They cannot tell the reader that they are forbidden from using the knowledge gained to tutor another person in calculus. That would be absurd.
IP law protects the works of the creator. The author of a calculus textbook did not invent calculus. As such, copyright law does not apply.
The model itself is a derivative work. It’s existence is what is under dispute. It’s not about using the model to produce further works
Then every single student graduating college produces derivative work.
Everything that required the underlying knowledge gained from the textbooks studied, or research papers read, is derivative work.
At the core of this, what are we saying? Your machine could only explain calculus because it was provided information from multiple calculus textbooks? Well, that applies to literally everyone.
Charge AI recurring tuition then.
It doesn’t break the law at all. The courts have already ruled that copyrighted material can be fed into AI/ML models for training:
This ruling only applies to the 2nd Circuit and SCOTUS has yet to take up a case. As soon as there’s a good fact pattern for the Supreme Court of a circuit split, you’ll get nationwide information. You’ll also note that the decision is deliberately written to provide an extremely narrow precedent and is likely restricted to Google Books and near-identical sources of information.
Have there been any US ruling stating something along the lines of “The training of general purpose LLMs and/or image generation AIs does not qualify as fair use,” even in a lower court?
You might want to read this post from one of the EFF’s senior lawyers on the topic who has previously litigated IP cases:
https://www.eff.org/deeplinks/2023/04/how-we-think-about-copyright-and-ai-art-0
I guess I can’t read anything and learn from it.
i don’t think it’s need rules against the law…
I do love how AI has gotten Corporate Giants to start attacking the Copyright System they’ve used to beat down the little man for generations
Maybe because it’s not the same corporations? We might be seeing a giant powershift from IP hoarders to makers.
Makers use the copyright system to their advantage as well though. If I write code and place it on github, the only thing stopping a mega corp stealing it is the copyright I hold.
Abolishing copyright is not a win.
Let’s not kid ourselves that the copyright is stopping mega corporations from stealing your github code.
What’s stopping them from hiring an engineer that basically rewrites your code? No one would ever know.
Copyleft enforcement is laughable at best and thats with legitimate non profits working on it (like FSF) and that’s when it comes to direct library use without modifications and there’s basically no history of prosecution or penalties for partial code copying (nor that there should be imo) that’s even when 1:1 code has been found!
I feel like copyright has been doing very little in modern age and have yet to see any science that contradicts my opinion here. Most copyright holders (like high 90%) are mega corporations like ghetty images that hardly contribute back to the society.
it definitely is, tho.
of course they could steal your code, but you could steal theirs. you could steal their software. you could steal their paywalled articles. you could steal all those things that are affected by artificial scarcity. They have much more to loose if copyright gets abolished.
In an ideal world, I’d make copyright to be 5 years for individuals and abolished for corporations. But the world is far from ideal and individuals have much to win if copyright gets abolished as a whole.
Science and the arts existed just fine without Intellectual Property. Abolishing copyright would be a huge win for society as a whole and a minor loss to corporations.
you know what? I like this argument. Software/Streaming services are “too complex and costly to work in practice” therefore my viewership/participation “could not exist” if I were forced to pay for them.
Hey if they want to set that precedent, so be it.
Oh, no no no.
Rules for thee, not for me.
That’s the other edge of this sword.
Not that I am a fan of the current implementation of copyright in the US, but I know if I was planning on building my business around something that couldn’t exist without violating copyright I would surely thought of that fairly early on.
“My profits from fencing your wallet could not exist if stealing your wallet were punished.”
“Ah, you’re right, how silly of me, carry on.”
The LCA principles also make the careful and critical distinction between input to train an LLM, and output—which could potentially be infringing if it is substantially similar to an original expressive work.
from your second link. I don’t often see this brought up in discussions. The problem of models trained on copyrighted info is definitely different than what you do with that model/output from it. If you’re making money from infringing, the fair use arguments are historically less successful. I have less of an issue with the general training of a model vs. commercial infringing use.
You’re responsible for infringing works, whether you used Photoshop, copy & paste, or a generative model.
I don’t disagree with that statement. I’m having trouble seeing how that fits with what I said, though. Can you elaborate?
It doesn’t really, I was just kind of restating what you quoted. Since no one factor of fair use is more important than the others, and it is possible to have a fair use defense even if you do not meet all the criteria of fair use, do you have data to back up your claims about moneymaking infringement?
Cool. Thanks. I can see it now. No, not really, just the pieces over time I’ve read on what wins fair use protections when challenged often talk about the interpretations involved and that profit making was generally seen as detracting from gaining fair use protections when the extent of the transformative nature was in question.
This mentions it, but of course it isn’t data on what has been granted protections vs. denials of protection. Harvard counsel primer on copyright and fair use
Noncommercial use is more likely to be deemed fair use than commercial use, and the statute expressly contrasts nonprofit educational purposes with commercial ones. However, uses made at or by a nonprofit educational institution may be deemed commercial if they are made in connection with content that is sold, ad-supported, or profit-making. When the use of a work is commercial, the user must show a greater degree of transformation (see below) in order to establish that it is fair.
Another reason why copyright should be shortened… Society has changed massively in the last 100 years, but every expression of our modern society is locked behind copyright.
I keep thinking how great it would be if the federal government made a central server system to access digital content for free via taxes.
All public domain and publicly funded research and content, all in one place. Could also host owned content for people/entities and pay out royalties automatically based on consumption.
There are ways to make this fairly affordable to everyone via taxes, but maybe the big opportunity is it could also allow companies to train AI on all the data for a fat, but fair subscription. The value of that could easily pay for enough to shrink any tax costs for the public.
In general, if the US government were smart (and not currently tearing itself apart) it would be creating a generative AI public service like the postal service, potentially even relying on public government documents and the library system for training.
Offer it at effectively cost for the public to use. Would drive innovation and development, nothing produced by it would be copyrightable, and it would put pressure on private options to compete against it.
We can still have the FedEx or DHL of gen AI out there, but they would need to contend with the public option being cheaper and more widely available for use.
In addition to the US government actually needing to do work, the senators would need to understand how to turn a computer on and off.
So the government gets to ignore copyright, but nobody else?
No thanks.
Yeah, that’s pretty much literally written into the copyright laws…
Yeah, but the attempts to kill off the USPS have been surprisingly bi-partisan
I literally don’t think the US will ever see a new socialized industry ever happen
Edited for clarity, to make it more obvious I’m not blaming just one party for this
if the US government was smart
So… This may be an unpopular question. Almost every time AI is discussed, a staggering number of posts support very right-wing positions. EG on topics like this one: Unearned money for capital owners. It’s all Ayn Rand and not Karl Marx. Posters seem to be unaware of that, though.
Is that the “neoliberal Zeitgeist” or what you may call it?
I’m worried about what this may mean for the future.
ETA: 7 downvotes after 1 hour with 0 explanation. About what I expected.
It’s interesting as it’s many of the MPAA/RIAA attitudes towards Napster/BitTorrent but now towards gen AI.
I think it reflects the generational shift in who considers themselves content creators. Tech allowed for the long tail to become profitable content producers, so now there’s a large public audience that sees this from what’s historically been a corporate perspective.
Of course, they are making the same mistakes because they don’t know their own history and thus are doomed to repeat it.
They are largely unaware that the MPAA/RIAA fighting against online sharing of media meant they ceded the inevitable tech to other companies like Apple and Netflix that developed platforms that navigated the legality alongside the tech.
So for example right now voice actors are largely opposing gen AI rather than realizing they should probably have their union develop or partner for their own owned offering which maximizes member revenues off of usage and can dictate fair terms.
In fact, the only way many of today’s mass content creators have platforms to create content is because the corporate fights to hold onto IP status quo failed with platforms like YouTube, etc.
Gen AI should exist in a social construct such that it is limited in being able to produce copyrighted content. But policing training/education of anything (human or otherwise) doesn’t serve us and will hold back developments that are going to have much more public good than most people seem to realize.
Also, it’s unfortunate that we’ve effectively self propagandized for nearly a century around ‘AI’ being the bad guy and at odds with humanity, misaligned with our interests, an existential threat, etc. There’s such an incredible priming bias right now that it’s effectively become the Boogeyman rather than correctly being identified as a tool that - like every other tool in human history - is going to be able to be used for good or bad depending on the wielder (though unlike past tools this one may actually have a slight inherent and unavoidable bias towards good as Musk and Gab recently found out with their AI efforts on release denouncing their own personally held beliefs).
That was a novel perspective for me. Thanks.
I see way too many people advocating for copyright. I understand in this case it benefits big companies rather than consumers, but if you disagree with copyright, as I do, you should be consistent.
Copyright law should benefit humans, not machines, not corporations. And no, corporations are not people. Anthony Kennedy can get bent.
I hate the MAGMA companies as much as anyone, but AI such as LLMs, especially the open source stuff Facebook and Stable diffusion is making, is beneficial to us all.
Abolishing copyright in the way that allows for the existence of Gen AI benefits people far more than it does corpos
Well said.
You don’t have to be against copyright, as such. Fair Use is part of copyright law. It exists to prevent copyrights from being abused against the interests of the general public.
But I am against any copyright beyond forcing attribution to the original creator.
…which AI fails to do.
AI creators, at least the open source ones, are usually pretty open about where they got the training data for their model
At the very least, every AI should be able to spit out a comprehensive list of all the material it used for training. And it should be capable of removing any specific item and regenerating its algorithm.
This is a fundamental requirement of the technology itself to function. What happens if one the training materials has a retraction? Or if the authors admit they used AI to generate it? You need to purge that knowledge to keep the AI healthy and accurate.
I’d say the main reason is companies are profiting off the work of others. It’s not some grand positive motive for society, but taking the work of others, from other companies, sure, but also from small time artists, writers, etc.
Then selling access to the information they took from others.
I wouldn’t call it a right wing position.
Wanting to abolish the IRS is a right-wing policy that will benefit the rich. That doesn’t change when some marketing genius talks about how the IRS takes money from small time artists, writers, etc. Same thing. It’s about substance and not manipulative framing.
That isn’t remotely similar…
The IRS takes a portion of income. This is taking away someone’s income, then charging access to it.
Like it or not, these people need money to survive. Calling it right wing to think these individuals deserve to be paid for someone taking their work, then using it for a product they sell access to, is absolutely insane to me.
I don’t know how this is supposed to make sense.
One is a percentage of income that everyone pays into.
The other is stealing someone’s work then using that person’s work for profit.
Recognizing that stealing someone’s work is not a right-wing position.
How is this complicated?
I see. Thanks for explaining.
This view of property rights as absolute is what right-libertarians, anarcho-capitalists, etc… espouse. Usually the cries of “theft” come when it gets to taxes, though. Is it supposed to be not right because it’s about intellectual property?
Property rights are not necessarily right-wing (communism notwithstanding). What is definitely right-wing is (heritable) privilege and that’s implied in these views of property.
ETA: Just to make sure that I really understand what you are saying: When you say “stealing someone’s work” you do mean the unauthorized copying of copyrighted expression, yes? Do you actually understand that copyright is intellectual property and that property is not usually called work? Labor and capital are traditionally considered opposites, of a sort, particularly among the left.
So… You think their art or writing was created by what then? Magic? Do you think no time was expended in the creation of books, research, drawings, painted canvases, etc?
Do you think they should starve because we currently live in a world driven entirely around money?
I don’t get your point even remotely.
Every single poster here has relied on disruptive technologies in their life. They don’t even realize that they couldn’t even make these arguments here if it was not for people before them pushing the envelope.
They don’t know the history of their technology nor corporate law. If they did they would just roll their eyes every time an entrenched economic interest started saber rattling about the next disruptive technology that is going to steal their profits.
The posters here are the people who complained about horsewhip manufactures that were going out of business because of cars. They are ignorant and act like the few sound bytes they heard make them an expert.
As an aside, when I browse TheGatewayPundit comments on AI articles, it is a lot more open, against legislation, and woke than I would expect!
Yeah…it’s pretty weird. Feels like some folks have really dived into LLMs regardless of ethics and will do any amount of hand waving to avoid criticism of a for-profit company openly attacking creatives’ livelihood with their own uncompensated works. In an ideal world where it wasn’t a case of “earn or die cold and alone in the streets”, sure, but this is just robbing those workers of the fruits of their labor and burning the ladder while
I think the “neoliberal zeitgeist” thought may be correct as neoliberal ideology devalues anything and everything that is not solely profit-driven, including just about everything that humans have historically found to make life meaningful.
It’s important to recognize that IP is conceptually fucky to begin with. They’re seeing what it’s claimed to be (creator ‘ownership’ of their creations) rather than what it really is (corporations using the government to enact violence on non-violent people).
It means nothing interesting. The position they feel they’re taking is “corporation bad” which is in line, they just haven’t analyzed how IP works in the real world.
I don’t know what you’re on about, the majority of the thread is pro open source AI and anti-capitalist, which is as left a stance as it gets, it’s not called “copyleft” for no reason. No one here wants to see AI banned and the already insane IP laws expanded to the benefit of the few corpos like the NYT at the expense of broader society.
IDK. I have seen a number of pro-corpo copyleft takes. It’s absolutely crazy to me. The pitch is that expansive copyright makes for expansive copyleft. It seems neo-feudal to me. The lords have their castles but the peasants have their commons.
Fair enough, seems like they’re down voting us anyway
I’d be fine with this argument if these generative tools were only being used by non-profits. But they aren’t.
So I think there has to be some compromise here. Some type of licensing fee should be paid by these generative AI tools.
You’re basically arguing for making any free use of them illegal, thereby giving a monopoly to the richest and most powerful capitalists.
Humans won’t be able to compete, and you won’t be able to use the means of generation either.
I’m arguing for free commercial use being illegal, absolutely.
And that fee should scale based on who is using it for commercial purposes. Microsoft and Google should be paying far, far out the ass for their data.
deleted by creator
And who does that serve?
Do you mean the whole thing or something specific like free commercial use being illegal?
I think the answer to both are the people who created the art, text, etc that these generative AI tools are going to make mostly obsolete.
Open source or open use AI will be practically illegal. Research will be practically impossible. It will be exclusively controlled by super rich and powerful corporations.
It won’t benefit the creators. It will only benefit those with the most capital that can buy up the training data needed and then can set the market so they make almost all of the money. For example you’d need to buy all the user content from reddit, facebook and twitter to train an AI. That will cost many millions because it’s a precious commodity (and only they own it). So only a few will control the “means of generation” and they will (have to) use it to make profit for themselves. This will make it practically illegal to make a free or an independent AI because you don’t have access to training data. This sets the rules and will lead to incredibly bad outcomes. For example anti-consumerist thinking or dissent could be suppressed, or other more subtle biases. Anything that reduces profit from advertising or threatens the shareholders. And they can manipulate the training data behind closed doors.
But that is how it’s going to go and it’s going to make the effects of AI generation on our civilization extra bad. We are so fucked :(
Fine. Shut them down.
And what about the open source models? Or the AI companies in countries that have more lax copyright laws? (Japan for example)
This technology exists now. We can’t put the genie back in the bottle. Copyright came out of the printing press, which allowed cheap copies to be made. Now a new technology has emerged so we likely need a new set of rules to replace the role that copyright performed, which was incentivizing artistic creation
Simple – you can’t charge for AI generation
Right? Like…I don’t give a shit. That’s not a threat or a fact that bothers me at all. They are only a tool for amassing more power and money. So what the fuck do I care.
I’m just trying to think about how refined AI would be if it could only use public domain data.
ChatGPT channels Jane Austin and Shakespeare.
That’s not really how it would work.
If you want that outcome, it’s better to train on as massive a data set as possible initially (which does regress towards the mean but also manages to pick up remarkable capabilities and relationships around abstract concepts), and then use fine tuning to bias it back towards an exceptional result.
If you only trained it on those works, it would suck at pretty much everything except specifically completing those specific works with those specific characters. It wouldn’t model what the concerns of a prince in general were, but instead model that a prince either wants to murder his mother (Macbeth) or fuck her (Oedipus).
That’s how it should be, but public domain has been crippled by Disney and co.
Of course they will exist. China will own them all.
Huh. You’d think in a situation where copyright is threatened by a lack of AI regulation, Disney would be all over this. Oh wait. They’re trying to use generative AI to make movies cheaper. Nevermind.
If they can’t afford a thing they want, that’s too bad.
The fact that their dream-AI ‘cant exist’ without stealing from everyone there is only one message to bounce back there from the rest of us;
‘good’
Sounds like a win to me
We are about to witness an incredibly power grab. They will be claiming practically the entirety of human intellectual works for example also for contributions made by billions of users on social media. Basically they will monopolize the entire power of GenAI for themselves.
This wil practically make free use of that power illegal. Generative AI will eliminate more and more jobs in the coming decades while we won’t be allowed to use it at all.
