Book Publishers Sue Google For Copyright Infringement Over Gemini AI Training (theguardian.com) 108
Major publishers Hachette, Cengage, Elsevier, and author Scott Turow have sued Google, accusing it of using millions of copyrighted books to train Gemini without permission or payment, in "one of the most prolific infringements of copyrighted materials in history." The Guardian reports: The publishers argue that Google repurposed books that had been supplied for limited services such as Google Books, Google Play Books and Google Scholar. Those services allowed Google to use the works in specific ways -- for example, to display searchable snippets or sell ebooks -- but not, the lawsuit claims, to copy them for training commercial AI products. "Desperate to maintain its online dominance, Google abandoned its early motto of 'Don't be evil' and engaged in one of the most prolific infringements of copyrighted materials in history," the suit states (PDF).
According to the complaint, the tech company made copies of copyrighted books to train Gemini without permission or payment, despite internal discussions acknowledging the legal risks. The filing claims Google flagged internally that it could face "$10Bs-$100Bs in potential fines" for using texts provided by publishers for Google Play Books. The publishers say Google's actions are harming authors and the wider publishing industry, arguing that AI-generated content could negatively impact book sales.
It notes that, for example, Gemini could generate "a 100-page murder mystery set in a quiet seaside town filled with secrets, that substitutes for an original copyrighted murder mystery on which Gemini trained" in 20 minutes for 39 cents. "No publisher or author can compete with that." The lawsuit names a number of specific books that the publishers allege were among the copyrighted works used without permission, including NK Jemisin's The Fifth Season, and Lemony Snicket's Who Could That Be at This Hour?
According to the complaint, the tech company made copies of copyrighted books to train Gemini without permission or payment, despite internal discussions acknowledging the legal risks. The filing claims Google flagged internally that it could face "$10Bs-$100Bs in potential fines" for using texts provided by publishers for Google Play Books. The publishers say Google's actions are harming authors and the wider publishing industry, arguing that AI-generated content could negatively impact book sales.
It notes that, for example, Gemini could generate "a 100-page murder mystery set in a quiet seaside town filled with secrets, that substitutes for an original copyrighted murder mystery on which Gemini trained" in 20 minutes for 39 cents. "No publisher or author can compete with that." The lawsuit names a number of specific books that the publishers allege were among the copyrighted works used without permission, including NK Jemisin's The Fifth Season, and Lemony Snicket's Who Could That Be at This Hour?
Good (Score:4, Insightful)
Re: (Score:2)
they have trained on the novel, the same way a human reads and trains on a novel.
Well, two points there. First, it's not the same way a human reads a novel. Second, a human doesn't "train" on a novel, because training implies learning to duplicate its structure. The only humans "training" on a novel and proceeding to write their own are what we call cut and dried plagiarists.
Re: Good (Score:2)
Re: (Score:2)
IN America, Downloading is pirating, distribution is. Read the code.
There is no example of any AI repeating an entire work beyond fair use.
" go ahead and help themselves without finding out what was appropriate under current legal structure?"
do you do that every time you read a web page? a book?
Re: Good (Score:2)
Re: Good (Score:2)
Aspiring writers are told one of the most important things to do to learn writing is, get this, read. Not instructional manuals, not tutorials, while those are absolutely encouraged, what they are told to do is read the types of books they want to write.
You know, so they can preprogram their neural structures on what the concepts are around structuring the types of books they want to write.
So yes, people train on books, novels, technical manuals , fiction, non-fiction.
Some of those people have didactic memo
Re: (Score:1)
Re: (Score:3)
What an amazing non sequitur. You might as well have said, "But humans have hair, and LLMs don't."
Re: (Score:1)
Re: (Score:2)
" LLMs are doing everything that humans do, "
I don't think people say that' however when it comes to speech patterns, AI are like humans because its based on Zipf's law. Which shows people speak in a very predictable manner.
"LLMs need real thought,"
define "real thought" in a meaningful way. In fact, write a book to be celebrated as the first person to do that. In the mean time, stow your Scotsman.
" It's not because we're force-fed them in order to regurgitate them later. "
Well, we are as children. IF your p
Re: (Score:1)
No AI regurgitates a work. There is no example of AI repeating any work outside of fair use.
Amazing, to think that regurgitation isn't a known problem at all, and that there certaintly aren't several famous lawsuits based on several-paragraph near verbatim output.
Re: (Score:2)
How does training imply duplicating structure? That's a strange definition.
Let's think about a simple model. Take your Bayes classifier, you know the P(A|B) thing (most people here should know the math behind it). You won't see any structure of the spam mail in the learned weights, still it is trained on Spam mails.
Re: (Score:1)
Re: (Score:2)
Wait until I tell you what's happening at thousands of college campuses across the nation!
Re: (Score:2)
I actually conducted an experiment on a living human. The result of the experiment was that after reading about 7000 hours in foreign language, the human in question become almost fluent in the language the novel was written in. In school grades, the jump was about 3 grades. Note that the human had basic skills in the foreign language before the experiment.
How does the human use these skills you might ask. Before reading, the human could only write very basic sentences, often with broken grammar. After read
Re: (Score:1)
Re: (Score:2)
People are trained by reading all the time. People saying otherwise is just simply dumb.
Re: (Score:2)
"it's not the same way a human reads a novel."
its like a mother reading to a child, so the statement stands. Plus, it's not A novel.
" a human doesn't "train" on a novel, "
we absolutely do.
"implies learning to duplicate its structure"
Which is what children do.
"The only humans "training" on a novel and proceeding to write their own are what we call cut and dried plagiarists"
lol, no. Every thing you write is based on everything you read and learned. All humans.
Sitting in a librasy and writes a book bu cutting
Re: (Score:2)
They stole copies of all those works. Just as if you broke into a bookstore at night and stole all the books.
Is that simple enough for you?
Re: (Score:2)
That's not a good analogy. What Google did was more like, they went in to their own corporate library, made copies of a bunch of books, and took the copies into the room next to their own corporate library and studied them. The books never left the building, but the person who studied them did. Without seeing the text of the license Google agreed to (and also probably being a Copyright lawyer,) we will probably never know if Google actually did anything illegal here. Immoral, unethical? Likely, but ill
Re: Good (Score:2)
Die already, rent seekers!!! AI is here to replace you.
Re: Good (Score:2)
Re: (Score:2)
Re: (Score:2)
They didn't take anything.
" In an ideal world they'd be forced to remove it from their data "
You do not understand AI.
If you write a book, do you get permission from every author you have read in the past? why not? you book is built upon everything you learned, so clearly they own a piece, right?
Re: (Score:2)
If you write a book, do you get permission from every author you have read in the past? why not? you book is built upon everything you learned, so clearly they own a piece, right?
There's a big difference between how human authors operate and how LLMs operate. LLMs, by design, use their training material to predict the text that should be placed on the screen given the prompt provided. While it is not reproducing a work word-for-word, it relies on prior works explicitly for any future production.
For human authors, yes, prior reading can inform about structure, phrasing, pacing, etc., and there are genre and common structural elements, unless they are a plagiarist or parodist they a
Re: (Score:2)
There's a big difference between how human authors operate and how LLMs operate...
Is there? How exactly do LLM's operate? I don't mean from a theoretical perspective. I mean from a "can you empirically tell me how LLMs generate their output?" perspective.
Training is fair use, but... (Score:5, Interesting)
Training is "inherently transformative", and thus protected as fair use. BUT... Google should lose this.
The publishers have a contract with Google that spells out the specific purposes for which the materials are to be used. These are not books that Google purchased off-the-shelf. They were provided by the publishers for that specific use. Any other use -even an otherwise legal use is a violation of that contract.
Even if Google argues fair use in training their AI system, they violated the contract. Pay up.
Re: (Score:1)
Training is "inherently transformative", and thus protected as fair use. BUT... Google should lose this.
The publishers have a contract with Google that spells out the specific purposes for which the materials are to be used. These are not books that Google purchased off-the-shelf. They were provided by the publishers for that specific use. Any other use -even an otherwise legal use is a violation of that contract.
Even if Google argues fair use in training their AI system, they violated the contract. Pay up.
There are four criteria that are used to help determine whether something is fair use [cornell.edu] (in the US) and Google seems to fall on the wrong side of three on them:
(1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
(2) the nature of the copyrighted work;
(3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
(4) the effect of the use upon the potential market for or value of th
Re: (Score:2)
Training is "inherently transformative", and thus protected as fair use.
You should stop posting on the law until you understand it better. Go watch some lawyer videos on fair use or something. The Wikipedia article is good. Your concept is wrong.
Re: (Score:2)
Any you got your law degree from...? and you have read all the contracts.. when? and where in these contract violation in the contract?
Don't tell me you are some ignorant blowhard spewing nonsense that fits your bias?
You cannot copyright a genre (Score:1, Informative)
These publishers are part of the problem (Score:2, Insightful)
These LLM models have taken ALL of humanities knowledge and ALL of humanity should benefit, as seen previously on Slashdot: Bernie Sanders Unveils $7 Trillion Plan To Give Americans Control of AI Industry https://yro.slashdot.org/story... [slashdot.org]
Re: (Score:2)
Re: (Score:2)
Did they ever mention copyrights on Star Trek???
Why would that even come up?
We would absolutely have a form of copyright post scarcity. Because post scarcity one of the thing that differentiate people is what they created.
Re: (Score:2)
Where's the payout for coders? (Score:4, Interesting)
Seriously, why should book authors get more compensation for their work being trained on than the rest of us?
Re: (Score:2)
Nobody's going to lift a finger on our behalf. That's what the thieves are counting on. Which I guess explains the vigorousness of the music industry's response to "pirating" back in the day.
Re: (Score:2)
Re: (Score:2)
I suppose I need to pay you every time I point someone to you works to?
Do I have to pay you when I finish you book and give it to someone else?
Re: (Score:2)
Re: (Score:1)
I taught Gemini how nose-picking works. They owe me!
Re: (Score:2)
Re: (Score:2)
Many open source licences require attribution and even more require the preservation of the licence.
Re: (Score:2)
It's not just up to them, it's up to copyright law.
And why do you think you speak for all book authors over all time?
People like you think libraries should be shut down, fair use removed, and no one allowed to resell a book they read.
Literally the logical conclusion to your post.
Typically american⦠(Score:4, Insightful)
Consensus cracking (Score:1)
Not sure what the answer is? (Score:2)
And related to Authors and others, yea they got robbed, but when it comes to LLM generated material not sure how it gets stopped now.
You are already seeing huge mountains of garbage burying things of value.
And going forward anything put on the Internet will be scraped if it has a hint of value.
Re: Not sure what the answer is? (Score:2)
Google is not selling stories written by others. They, and the other companies in AI, are creating AI tools that learn from existing materials and are allowing users to create new works using them.
I am not sure I am happy with the extension of copyrights to any use of the material. I am also not happy with the idea that the author of a work has to authorise every different type of use of their material. A copyright should, in my opinion, be about who is allowed to make copies of a work and then sell it, or
Re: (Score:2)
You can't stop the LLM if it's published... but you can sue the company that scraped data it was not legally entitled to scrape, and the legal sanctions should involve destruction of the collected archive of training data and all copies of the resulting LLM as well as a financial penalty that is sufficiently large to dissuade future repetitions of the offense.
"But that would harm our bottom line" is not an acceptable defense against this. Don't steal. It's easy.
Of course, stealing's easier if you're a meg
Re: (Score:2)
Re: Not sure what the answer is? (Score:2)
Re: (Score:2)
And related to Authors and others, yea they got robbed, but when it comes to LLM generated material not sure how it gets stopped now.
That's not an argument.
"Yeah, that guy is dead now. We have a pretty solid idea who did it. But not sure if that'll make him alive again, so let's not bother with catching them."
Judgment the world wants (Score:3)
Google; You wre evil and will be broken up.
Copyright Cartel: 95-year copyright is an abomination. That goes down to a reasonable five years, and one year for journals.
Use AI to make-up a religion (Score:2)
Fair Use!!! (Score:1)
Re: (Score:2)
No it isn't, else I could take a scissors and cut and paste from major works and pretend it was my own original work. without other peoples original words, these LLMs are just so many door-stops. Data embedded and mapped into N-Dimensional Space as tokens.
Re: (Score:2)
I think the current law does allow for 'collage' which is what you are propose, however there are severe limitation on monetizing it.
Wanna bet? (Score:2)
that substitutes for an original copyrighted murder mystery on which Gemini trained" in 20 minutes for 39 cents. "No publisher or author can compete with that."
I think Nora Roberts could give them a run for their money.
I'm with the authors and publishers (Score:2)
Re: (Score:2)
This may be but current copyright law may not account for this particular sentiment - they will unpack the training into current language in the law and see what matches. If the "sentimental" request like above does not then it will be rejected and well you can try to augment the law.
Google may be safe (Score:2)
IANAL, but there's a precedent in the anthropic ruling [documentcloud.org] (page 11-ish), where the judge argues that using the books a-la-Google is fair use for training.
Anthropic got in trouble for using pirated books.
Google has scanned pretty much every book under the sun by buying it, digitizing, and destroying it and in that ruling that can be fairly used to train an LLM.
Google may settle just to get over it, but it has a decent fighting chance.
Re: (Score:2)
The lawyers in that case were on the take, the class members got fucked for pennies on the dollar for Dario.
The real lawsuit is the one by the opt outs.
So we need to pay to read? (Score:2)
How is reading copyright infringement? It's no different then what AI is doing.
Also, copyright is about disturbing works. There is no record of any AI distributing works beyond fair use.
If I write a book by cutting the words out of a library full of books, it's not copyright infringement. Same with AI.
Bunch of people don't understand AI. these same writer probably rally against libraries.
Re: (Score:3)
Nice wordplay. Meaning nothing, and trying to justify THEFT.
As I responded to someone above, it's the same as if you personally broke into a bookstore at night and stole all the books.
And speaking as an author, with two books published by a small press, YOU HAVE NO FUCKING RIGHT to read my novels without paying for them.
Steal now, make money, pay a bit later? (Score:3)
Training == collage? (Score:2)
Can training of LLMs be classified as "collage"? If so this would be allowed with some limitations on revenue from it. afaiu?
Re: Dictionaries Mysteriously Not Sued (Score:2)
Because the dictionary publishers didn't copy all those works into their training corpus
Re: (Score:1)
Re:Dictionaries Mysteriously Not Sued (Score:5, Informative)
Dictionary publishers do get sued. In 2001, the "New Oxford American Dictionary" added ghost (fake) word "esquivalience" to their boko and sued several online dictionaries that included the word (who only could have added through bulk copying the first dictionary).
In 1998, Larousse and Robert (two well-known dictionary publishers) sued Maxidico for plagiarism due due to striking similarities in definitions, including same mistakes/typos. Maxidico was sentenced to the equivalent of 1.5 million euros in damages (of the money of the time) and filed for bankrupcy.
Re: (Score:1)
Re: (Score:2, Insightful)
Dictionary publishers have never been accused of downloading massive torrents of pirated copies of books and processing them.
Google on the other hand HAS been accused of that, and the decade of litigation related to that ultimately rules that the limited things google was doing with it was fair use. The dictionary companies are likely paying for enhanced access to that google data now.
The AI companies are singing the same fair use tune, but its really quite different. Google was doing it (at the time) to al
Re: (Score:2)
No, its not. In fact, I'd wager that they only have one copy of each unique word in their database*.
* - you have to assume that they also have development, testing, and pre-production copies of that same database.
Re: (Score:2)
Who is they? And what database are you referring to?
Dictionaries often contain quotes from original source material, or references to where it was first used, or first used a certain way.
Re: (Score:1)
Re: (Score:1)
Re: (Score:2)
"No. It is not copyright infringement"
Go ahead, prompt for that story and publish your own 'moonlit princess". It is not a court case you'd win; the details taken from the Disney version are beyond excessive.
" and there's no reason to hold copyright so sacred anyway. Are you seriously wanting to protect hundred year old fairy tails from being retold?"
That's an entirely separate discussion. Legally it is infringement. Whether it should be is completely separate question, or how long it should be are separate
Re: (Score:2)
Re: (Score:2)
The chatgpt story is a ripoff of the Disney version though.
The scene I quoted with the three "magical gaurdians" bestowing three gifts at a celebration crashed by the sorcerous who places the curse -- that is not from the original source material, its not novel either, its a scene ripped straight from the Disney version. And its not an isolated issue with the AI version.
There is no question that Disney doesn't "own" sleeping beauty, but they do own their telling of it. This was an obvious ripoff of that pa
Re: (Score:1)
Re: (Score:2)
"It is not a court case you'd win; the details taken from the Disney version are beyond excessive."
We have decades of court cases that show you are wrong. STFU.
"beyond excessive."
based on what?
Re: (Score:2)
You seriously telling me this is NOT copyright infringement?
It's not copyright infringement. Copyright doesn't protect ideas, only the specific form an idea takes. This is why, for example, there's many sci-fi seafaring tales, but in space books, TV shows and films.
Heck, I finally got around to watching that Wednesday series on Netflix and they basically did that whole school for exceptional outcasts trope, which is so heavily used that it actually ends up divided [tvtropes.org] into [tvtropes.org] several [tvtropes.org] sub [tvtropes.org] tropes. [tvtropes.org]
Re: Dictionaries Mysteriously Not Sued (Score:2)
Re: Dictionaries Mysteriously Not Sued (Score:2)
Re: (Score:2)
Yes, it does in America. Please read US copyright code. fair use is a things. I can give you a used book for free, and you can read it and give it to someone else for free and so on. No copyright violation. Yu can access works for free through Libraries.
Copyright expires.
Re: (Score:3)
To be fair I don't think Google is being accused of using BitTorrent, that was Facebook/Meta. Google actually scanned a large number of books themselves, for the Google Books project. That was found to be legal, although that only covered scanning, searching, and making excerpts available, not AI training.
As for your Sleeping Beauty example, since that story dates back to the 1600s, and appears to be a based on an even older story going back to the 1300s, any copyright protection it may have enjoyed has lon
Re: (Score:2)
Copy protection on the original sleeping beauty is long expired.
Copy protection on the Disney version is good for years yet.
In the original versions, the scene i quoted? There is no scene like that in the original version. But it is beat for beat straight from the Disney version. If you want to tell a sleeping beauty story, you absolutely can, the original source material not copy protected, you can faithfully tell that story, including the unconscious rape of the princess to impregnate her so that she fina
Re: (Score:2)
"f downloading massive torrents of pirated copies of books and processing them."
Downloading isn't piracy, please read copyright code.
" to generate new content, "
What a misleading phrase.
" much of which non-factual in nature,"
It's called Fiction. Your library has a whole section.
"and often very arguably explicitly creatively derivative."
Not really.. or no more then people who write.
prompt: "Make a story like sleeping beauty" ... 2 seconds later we have "The Moonlit Princess" and we'll just self-publish that
Re: (Score:2)
It's been 16 years since I've felt inspired to comment on Slashdot, but I'm breaking silence to congratulate you on achieving your goal (I assume) of being the worst, most ignorant version of yourself. It must take a concerted effort to become such a cartoonish portrayal of the worst stereotype imaginable of a red-pilled incel. To be so completely wrong in a public forum, and to have no shame about it, is genuinely impressive.
Not in a positive way, of course. You're a small-minded tool with a pathologically