How is it fair? Do you expect 9,000 from Google, Meta, OpenAI, and everyone else? Were your books imitated by AI?
Infringement was supposed to imply substantial similarity. Now it is supposed to mean statistical similarity?
Infringement was supposed to imply substantial similarity. Now it is supposed to mean statistical similarity?
The suit isn't about Anthropic training its models using copyrighted materials. Courts have generally found that to be legal.
The suit is about Anthropic procuring those materials from a pirated dataset.
The infringement, in other words, happened at the time of procurement, not at the time of training.
If it had procured them from a legitimate source (e.g. licensed them from publishers) then the suit wouldn't be happening.
https://www.documentcloud.org/documents/26084996-proposed-an...
> reproducing purchased and scanned books to train AI constituted fair use
Library Genesis has one copy. It then sends you one copy and keeps it's own. The entity that violated the _copy_right is the one that copied it, not the one with the copy.
Of course, American law is different. But is it the case that copies made for the purpose of using illegally obtained works are not infringing?
Well, the question here is "who made the copy?"
If you advertise in seedy locations that you will send Xeroxed copies of books by mail order, and I order one, and you then send me the copy I ordered, how many of us have committed a copyright violation?
> "Who made the copy?"
This begs the question. With digital media everybody involved makes multiple copies.
The portion the court said was bad was not Anthropic getting books from pirated sites to train its model. The court opined that training the model was fair use and did not distinguish between getting the books from pirated sites or hard copy scans. The part the court said was bad, which was settled, was Anthropic getting books from a pirate site to store in a general purpose library.
--
"To summarize the analysis that now follows, the use of the books at issue to train Claude
and its precursors was exceedingly transformative and was a fair use under Section 107 of the
Copyright Act. And, the digitization of the books purchased in print form by Anthropic was.
also a fair use but not for the same reason as applies to the training copies. Instead, it was a
fair use because all Anthropic did was replace the print copies it had purchased for its central
library with more convenient space-saving and searchable digital copies for its central
library — without adding new copies, creating new works, or redistributing existing copies.
However, Anthropic had no entitlement to use pirated copies for its central library. Creating a
permanent, general-purpose library was not itself a fair use excusing Anthropic’s piracy."
"Because the legal issues differ between the *library copies* Anthropic purchased and
pirated, this order takes them in turn."
Questions
As an author do you think it matters where the book was copied from? Presumably, a copyright gives the author the right to control when a text is reproduced and distributed. If the AI company buys a book and scans it, they are reproducing the book without a license, correct? And fair use is the argument that even though they violated the copyright, they are execused. In a pure sense, if the AI company copied (assuming they didn't torrent back the book) from a "pirate source" why is that copy worse then if they copied from a hard book?
isn't digitizing your own copies as backups and personal use fine? so long as you dont give away the original while keeping the backups. similarly, dont give away the digital copies.
No? I think there are a lot more details that need to be known before answering this question. It matters what they do with it after they scan it.
Yes
> it means that yes what I did was technically a violation but is forgiven
Not at all. All "affirmative defence" means is that procedural the burden is on me to establish that I was not violating the law. The law isn't "you can't do the thing", rather it is "you can't do the thing unless its like this". There is no violation, there is no forgiveness as there is nothing to forgive, because it was done "like this" and doing it "like this" doesn't violate the law in the first place.
There was no issues with the physical copies of books they purchased and scanned.
I believe the issue of USING these texts for AI training is a separate issue/case(s)
The entire point of deep learning is to copy aspects from training materials, which is why it’s unsurprising when you can reproduce substantial material from a copyrighted work given the right prompts. Proving damages for individual works in court is more expensive than the payout but that’s what class action lawsuits are for.
Given that books can be imitated by humans with no compensation, this isn't as strong as an argument as you think. Moreover AFAIK the training itself has been ruled legal, so Anthropic could have theoretically bought the book for $20 (or whatever) and be in the clear, which would obviously bring less revenue than the $9k settlement.
And in general, when an LLM is able to recreate text that's a training error. Recreating text is not the purpose. Which is not to excuse it happening, but the distinction matters.
Real-world absurd example: A company hires a bunch of workers. They then give them access to millions of books and have the workers reading the books all day. The workers copy the books word by word, but after each word try to guess the next word that will appear. Eventually, they collectively become quite good at guessing the next word given a prompt text, even reproducing large swaths of text almost verbatim. The owner of company Y claims they owe nothing to the book owners, because it doesn't count as reading the book, and any reproduction is "coincidental" (even though this is the explicit task of the readers). They then use these workers to produce works to compete with the authors of the books, which they never paid for.
It seems many people feel this is "fair use" when it happens on a computer, but would call it "stealing" if I pirated all the books of JK Rowling to train myself to be a better mimicker of her style. If you feel this is still fair use, then you should agree all books should be free to everyone (as well as art, code, music, and any other training material).
Can you provide an example of someone being successfully sued for "mimicking style", presumably in the US judicial system?
I won't rehash the many arguments as to why the output is also a violation, but my point was more the absurd view that stealing and using all the data in the world isn't a problem because the output is a lossy encoding (but the explicit training objective is to reproduce the training text / image).
However, AI has been show to copy a lot more than what people consider style.
Music has had this happen numerous times in the US. The distinction isn’t an exact replica, it’s if it could be confused for the same style.
George Harrison lost a case for one of his songs. There are many others.
https://ultimateclassicrock.com/george-harrison-my-sweet-lor...
That's called extreme overfitting. Proper training is supposed to give subtle nudges toward matching each source of text, and zillions of nudges slowly bring the whole thing into shape based on overall statistics and not any particular sources. (But that does require properly removing duplicate sources of very popular text which seems to be an unsolved problem.)
So your analogy is far enough off that I can't give it a good reply.
> It seems many people feel this is "fair use" when it happens on a computer, but would call it "stealing" if I pirated all the books of JK Rowling to train myself to be a better mimicker of her style.
I haven't seen anyone defend the piracy, and the piracy is what this settlement is about.
People are defending the training itself.
And I don't think anyone would seriously say the AI version is fair use but the human version isn't. You really think "many people" feel that way?
To generate working code the output must follow the API exactly. Nothing separates code and natural language as far as the underlying algorithm is concerned.
Companies slightly randomize output to minimize the likelihood of direct reproduction of source material, but that’s independent of what the neural network is doing.
And it's not really about randomizing output. The model gives you a list of likely words, often with no clear winner. You have to pick one somehow. It's not like it's taking some kind of "real" output and obfuscating it.
It’s very rare for multiple outputs to actually be equal so the only choice is to choose one at random. Instead its become accepted practice to make sub optimal choices for a few reasons, one of which really is to decrease the likelihood of reproducing existing text.
Nobody wants a headline like: “Meta's Llama 3.1 can recall 42 percent of the first Harry Potter book” https://www.understandingai.org/p/metas-llama-31-can-recall-...
But copyright was based on substantial similarity, not causal links. That is the subtle change. Copyright is expanding more and more.
In my view, unless there is substantially similarity to the infringed work, copyright should not be invoked.
Even the substantial similarity concept is already an expanded concept from original "protected expression".
It makes no sense to attack gen-AI for infringement, if we wanted the originals we would get the originals, you can copy anything you like on the web. Generating bootleg Harry Potter is slow, expensive and unfaithful to the original. We use gen-AI for creating things different from the training data.
Copyright isn’t supposed to apply if you happen to write a story that bares an uncanny similarity to a story you never read written in 1952 in a language you don’t know that sold 54 copies.
So you're agreeing with me? The courts have been pretty clear on what's copyrightable. Copyrights only protect specific expressions of an idea. You can copyright your specific writing of a recipe, but not the concept of the dish or the abstract instructions itself.
(They can still sue for damages, but they can't claim copyright over your game itself.)
But otherwise, you're essentially asking if you can somehow bypass license agreements by simply refusing to read them, which would obviously render all licensing useless.
In the event that you try to play games to get around that acknowledgement: Courts aren't machines, they can tell that you're acting in bad faith to avoid license restrictions and can punish you appropriately.
> Most paid software generally makes you acknowledge that you have read and accepted the terms of the license before first use, and includes a clause that continued use of the software constitutes acceptance of the license terms.
Huh. If only I'd known that.
Why do you think that is?
Why did you think it made sense to respond to the question "Why do you think X is true?" with "Did you know that X is true?"?
Thus, isn't the settlement essentially Anthropic admitting that they don't really have an effective defense against the piracy claim?
The authors can still sue for damages though (and did, and had a strong enough case Anthropic is trying to settle for over a billion dollars).
Or you could sue him on a theory of unjust enrichment, in which case, if he lost, he'd owe you nothing, and if he won, he'd owe you all of his winnings.
It's not clear to me why the same theory wouldn't be available to Adobe, though the copyright question wouldn't be the main thrust of the case then.