As far as I know, training generative AI on copyrighted works is fair use, though there may be some legal subtleties around this that need to be worked out. I'm guessing these copyright owners are trying to get some precedent in their favor with this suit.
IMHO AI training should be fair use and I hope this and other suits like it go nowhere. Honestly I don't think the authors have a leg to stand on here. If a generative AI is trained on a bunch of Stephen King novels it doesn't learn to reproduce the works themselves, ie. there is no copy of "Carrie" in the model data which it will spit out pieces of. A good model will learn to write in the style of King; this can be compared to a human author who also reads King's works and wants to write in the same style. Clearly there is no copyright violation in that case, so generative AI should be covered as well.
Obviously, because it's the equivalent of reading the damn book. These guys are upset that an entity read their book.
If I read Stephen King's book, and spit it back out in different pieces with different wording, that is legal. It doesn't make for a good book, but it's legal.
A person reading a book versus inputting parts of the work into a machine are different things and should be treated differently.
I'm not convinced one way or the other, in any case, but it's incorrect to claim these are equivalent. Even if human creativity worked in the way a LLM does (arguable) a person is way different than software.
In my mind, the only crime you can commit with input is piracy, which is I guess what we are talking about. However, if the LLM came across those legitimately, I can't see it as piracy on the output side. Unless the result is literal plagiarism.
This type of reasoning ignores that a generative AI essentially allows millions or billions of copies to be created in relatively trivial timeframe (minutes/hours/days).
It's not at all comparable to a real human being sitting down and reading a work and emulating it.
Back in reality, there are protections against reverse engineering products. It seems reasonable to consider that the process of training the AI on certain works (anything not in public domain) is a reverse engineering of a creative work and probably deserves some protections.
Back in reality, there are protections against reverse engineering products.
Not in the US. I am not familiar with other jurisdictions... are you referring to one of them?
In the US reverse engineering is allowed except under some very specific conditions such as obtaining the product illegally or breach of contract (ie you signed a contact saying you wouldn't reverse engineer).
Maybe you are thinking about patents? If you patent your product you are protected from reverse engineering with the tradeoff that the protection is for a limited time and you must divulge your secrets. If you do not receive a patent your trade secret is not (and should not be IMHO) protected from reverse engineering. But of course this wouldn't have anything to do with a copyright suit.
This type of reasoning ignores that a generative AI essentially allows millions or billions of copies to be created in relatively trivial timeframe (minutes/hours/days).
I don't think I follow your argument. Why is it relevant that a generative AI could spit out thousands or even millions of Steven King styled books so long as none of them are in violation of copyright? If I write something in the style of King because I've read a lot of his books and I like his work should he be able to sue me? Of course not. It doesn't matter if I write one book or a million; if there is no copyright violation there is no problem.
As far as I know, training generative AI on copyrighted works is fair use, though there may be some legal subtleties around this that need to be worked out. I'm guessing these copyright owners are trying to get some precedent in their favor with this suit.
IMHO AI training should be fair use and I hope this and other suits like it go nowhere. Honestly I don't think the authors have a leg to stand on here. If a generative AI is trained on a bunch of Stephen King novels it doesn't learn to reproduce the works themselves, ie. there is no copy of "Carrie" in the model data which it will spit out pieces of. A good model will learn to write in the style of King; this can be compared to a human author who also reads King's works and wants to write in the same style. Clearly there is no copyright violation in that case, so generative AI should be covered as well.
They'll know it's a copy when the monsters need to be defeated by preteen sex
Obviously, because it's the equivalent of reading the damn book. These guys are upset that an entity read their book.
If I read Stephen King's book, and spit it back out in different pieces with different wording, that is legal. It doesn't make for a good book, but it's legal.
A person reading a book versus inputting parts of the work into a machine are different things and should be treated differently.
I'm not convinced one way or the other, in any case, but it's incorrect to claim these are equivalent. Even if human creativity worked in the way a LLM does (arguable) a person is way different than software.
In my mind, the only crime you can commit with input is piracy, which is I guess what we are talking about. However, if the LLM came across those legitimately, I can't see it as piracy on the output side. Unless the result is literal plagiarism.
This type of reasoning ignores that a generative AI essentially allows millions or billions of copies to be created in relatively trivial timeframe (minutes/hours/days).
It's not at all comparable to a real human being sitting down and reading a work and emulating it.
Back in reality, there are protections against reverse engineering products. It seems reasonable to consider that the process of training the AI on certain works (anything not in public domain) is a reverse engineering of a creative work and probably deserves some protections.
Not in the US. I am not familiar with other jurisdictions... are you referring to one of them?
In the US reverse engineering is allowed except under some very specific conditions such as obtaining the product illegally or breach of contract (ie you signed a contact saying you wouldn't reverse engineer).
Maybe you are thinking about patents? If you patent your product you are protected from reverse engineering with the tradeoff that the protection is for a limited time and you must divulge your secrets. If you do not receive a patent your trade secret is not (and should not be IMHO) protected from reverse engineering. But of course this wouldn't have anything to do with a copyright suit.
I don't think I follow your argument. Why is it relevant that a generative AI could spit out thousands or even millions of Steven King styled books so long as none of them are in violation of copyright? If I write something in the style of King because I've read a lot of his books and I like his work should he be able to sue me? Of course not. It doesn't matter if I write one book or a million; if there is no copyright violation there is no problem.