Studio Ghibli, Bandai Namco, Square Enix Demand OpenAI Stop Using Their Content To Train AI 50
An anonymous reader shares a report: The Content Overseas Distribution Association (CODA), an anti-piracy organization representing Japanese IP holders like Studio Ghibli and Bandai Namco, released a letter last week asking OpenAI to stop using its members' content to train Sora 2, as reported by Automaton. The letter states that "CODA considers that the act of replication during the machine learning process may constitute copyright infringement," since the resulting AI model went on to spit out content with copyrighted characters.
Sora 2 generated an avalanche of content containing Japanese IP after it launched on September 30th, prompting Japan's government to formally ask OpenAI to stop replicating Japanese artwork. This isn't the first time one of OpenAI's apps clearly pulled from Japanese media, either -- the highlight of GPT-4o's launch back in March was a proliferation of "Ghibli-style" images.
Altman announced last month that OpenAI will be changing Sora's opt-out policy for IP holders, but CODA claims that the use of an opt-out policy to begin with may have violated Japanese copyright law, stating, "under Japan's copyright system, prior permission is generally required for the use of copyrighted works, and there is no system allowing one to avoid liability for infringement through subsequent objections."
Sora 2 generated an avalanche of content containing Japanese IP after it launched on September 30th, prompting Japan's government to formally ask OpenAI to stop replicating Japanese artwork. This isn't the first time one of OpenAI's apps clearly pulled from Japanese media, either -- the highlight of GPT-4o's launch back in March was a proliferation of "Ghibli-style" images.
Altman announced last month that OpenAI will be changing Sora's opt-out policy for IP holders, but CODA claims that the use of an opt-out policy to begin with may have violated Japanese copyright law, stating, "under Japan's copyright system, prior permission is generally required for the use of copyrighted works, and there is no system allowing one to avoid liability for infringement through subsequent objections."
Re: (Score:2)
Fascinating perspective. Thanks for your thoughtful comment.
Re:I have multiple opinions (Score:4, Insightful)
The AI company /is/ selling the result. Via a monthly fee, but still.
The AI user is also selling the result. Probably via YouTube or social media monetization, but still.
And the media platform is also selling the result, by auctioning ads around the slop.
It's turtles all the way down, and none of them give a rats ass about the IP holder. So there is merit to the complaints. If the IP holders manage to get a leg up, this is most probably going to result in some sort of ContentID system, with takedowns and copyright claim abuses and the whole lot of it.
I don't even know where I would stand in this. I honestly don't give a rats ass about any media companies nor theri precious IP, with exactly the one exception of Ghibli. Slopping them seems to be if not sacrilegious, then at least hopelessly bad taste and tone deaf... But nobody is going to ask me anyway, it's going play out like any other one money against another money battle.
Re: (Score:2)
And the person learning by reading/watching may just as well sell the result.
Re: (Score:2)
The AI company /is/ selling the result. Via a monthly fee, but still.
And you're allowed to providing the work is transformative. It is accepted copyright law that you can't copyright style. Simply selling isn't the only issue here, copyright law runs for more than 1 page for a reason.
Re:I have multiple opinions (Score:4, Interesting)
It's not exactly like training a person, because human brains and computers are different, additionally a human looking at art is generally what the art is for and the human getting greater understanding of art at the same time is a bit of a side-effect.
That said, it's not exactly like making a copy either. It's generally undefined in legal terms, because laws have been written on the assumption that a person looking at it or some kind of literal copying procedure are the only important things which are done with art, but it exists in a space somewhere in between the two, having elements of both.
Re: (Score:2)
No, even a "trained" person is not allowed to just reproduce copyrighted works and get paid for it, even as a mishmash. And that's not even getting into the degree to which the "AI" might just be a glorified index and storage system, which means it really does have a copy of the copyrighted work in it and therefore the model weights themselves are a derivative work.
I think there should be no question that a semi-arbitrary algorithm optimized to reproduce a dataset under certain operations is a derivative
Re: (Score:2)
The term "exactly" is wrong, but the idea is that it is legally the same and in particular to debunk the idea of "AI is just photobashing".
AI learns concepts, humans learn concepts.
AI models are simple, that's why concepts are represented simple and counting fingers is a challenge. Human brains are complex, which allows them not only to grasp the concept of "A hand has five fingers" but also the concept of "Let's give the alien hand another number of fingers" and finally the concept of "Even the alien has t
Re: (Score:2)
Of course it does. Maybe the word is not clear, let's give a few examples:
- cat, dog
- red, blue, pink, black, white
- upside down
- smiling
- horror style
- line art
While some are subjects and others are style elements, these are concepts the model knows and can arbitrarily combine to create something new. It only knows them as concepts, which means you don't have the one horror style, but for every new seed a new horror style (except you combine enough concepts to clearly communicate an unique style).
Re: I have multiple opinions (Score:1)
You cannot copyright an art style. You can have trademarks for your characters though, so OpenAI could be violating trademark law. This is how Disney protects their characters. Copyright only prevents exact duplication of works. Piracy is copyright violation, but AI using content for training is not. Governments could change laws or create new laws about AI training, but I have not heard of any so far.
Re: (Score:2)
a human looking at art is generally what the art is for
You're speaking like a consumer not like an arts student or a human explicitly learning an art. You're comparing the wrong human practices.
Think of it this way: I didn't read a math textbook for the suspenseful plot.
Re: (Score:2)
"Think of it this way: I didn't read a math textbook for the suspenseful plot."
Talk for the reflexive relation with you and not forall.
Re: (Score:3)
I tend to agree, but I'm not sure why selling the result should trigger copyright.
If a human artist studies and replicates the style of other artists, and sells it, is that a copyright violation? If it is, it should be for the AI as well. If it's not, not.
I thought copyright applied to specific works, not styles. AFAIK law has never protected "styles". (But I'm no expert on this.)
Re: (Score:2)
Re: (Score:3)
The immaturity of no guardrails. (Score:2)
AI has great potential to do serious work, but the mass public seems to only care about making annoying slop
AI won't have the potential to do serious work until it is forced to grow the hell up.
Right now it's a fucking toy. Ask it to do dumb shit, it does dumb shit. Ask it to do dumber shit, it does dumber shit. With the extra bone-us of it learning how to do the dumbest shit.
The only other brain that acts that way, is a childs. Time to stop pretending otherwise.
Re: (Score:1)
Computers are not people.
Re: (Score:1)
But there is "fair use" under the Copyright Act of 1976, which allows for the usage of copyrighted material without explicit permission.
That's what's being debated and what AI companies are hoping for. One of the questions, for fair use, is whether new work competes with the original. And the Copyright Office hasn't really answered whether AI use is fair
Re: I have multiple opinions (Score:1)
I'd suggest that having only direct sales be considered infringement is too open.
Presumably, ImageBots are considered by someone at ImageBotCompany to generate some value even if it's intangible future value, for example: brand awareness - sure no cash has changed hands but if there were no expectation of future value from brand-awareness exercises, they wouldn't be happening.
So I would argue we're seeing "look at this thing our ImageBot made. Our intention is to make it so similar that it's undeniably in t
Re: (Score:2)
No it is not. AI is not trained on how to draw, it already knows how to draw.
It is instead trained on how to draw particular styles. It does not need to learn how to draw using my style in order to come up with it's own style. The only reason to learn to draw my style is to draw my style.
People are different. We can develop our own styles based on other things. That is why it is OK for a person to copy someone else's style AI never does that. They never develop their own style, so have no need to le
Re: (Score:1)
You wouldn't steal a car (Score:5, Insightful)
You wouldn't steal a handbag.
You wouldn't steal a television.
You wouldn't steal a DVD.
Downloading someone's content for AI training is stealing.
Stealing is a crime.
I'm glad someone is finally calling all this data harvesting for what it really is.
Re: (Score:1)
I think a better analogy would be somebody who sets a used couch on the curb in front of their house, and then gets upset because an upholstery business picks it up so they can refurbish and sell it, rather than it being picked up by a neighbor who just wants a couch. These companies made their content public, and now they are upset because the wrong kind of user picked it up and used it for commercial purposes.
Re: (Score:3)
"Made it public"? I mean, they published it. You can play Pac-Man, if you have a quarter. You can watch _Spirited Away_, if you bought the DVD or are willing to pay for a streaming service that carries it.
But you've never been allowed to then publish your own works featuring Pac-Man or Spirited Away Monster just because you paid a quarter. How is this different?
Re: (Score:2)
It's different because AI doesn't republish your copyrighted work. It has always been legal for humans to draw inspiration from a copyrighted work, and use it to create their own original work. That's essentially what AI does.
In your Pac-Man example, you could play the game, and create your own game with a fish swimming through a maze, being chased by sharks. It might be very similar to Pac-Man in how you play it, but it wouldn't infringe on Pac-Man's copyright.
Re: (Score:1)
By that logic, all an AI company needs to do is buy a single copy of a work. Then they're free to do what they want with the content (doctrine of first sale).
As to your other other questions,
I would buy a knockoff.
I wouldn't steal a television or take one that fell off the back of a truck. However, I would repurpose a computer monitor to show televised content.
I would transcode DVD content into MP4.
Stealing is a crime; it's not just the AI companies that are "stealing." It's hard to throw a rock without hit
You got 3nds of the Kool Aid didn't you (Score:3, Insightful)
No, I wouldn't steal a car, handbag, television, or a physicle DVD. That would deprive someone of their property and be a violation of Criminal law.
Downloading a copy of someones IP without their permission does NOT deprive them of their physical property and is not covered by Criminal law. Using a portion of someones IP for parody or satire is allowed in small amounts.
Making a complete copy and using it for profit is Copyright infringement and covered under Civil law. Operations that make and distribute
Re: (Score:2)
You know, that you're citing the ad everyone made fun of seriously?
Re: (Score:2)
You wouldn't steal a handbag. You wouldn't steal a television. You wouldn't steal a DVD. Downloading someone's content for AI training is stealing. Stealing is a crime.
I'm glad someone is finally calling all this data harvesting for what it really is.
You literally used the same logic that the mandatory warnings about piracy that they used to shove onto the beginning of every DVD they sold. Funny how that logic only applied to us plebians just trying to enjoy a fucking movie that we already god damned paid for, but doesn't apply to the jackholes that wanna suck up every bit of content / data that humanity has ever created so they can replace us at everything from drawing to our actual jobs.
You wouldn't steal a font (Score:3)
"LOL," said the thieving techbros, (Score:2)
What are they training? (Score:2)
How to make succulent looking eggs or giant mechanized robots?
Re: (Score:1)
No, they will not sue a bunch of Slashdotters for posting graffiti. What they will probably do is sue a megacorp using their IP to drive traffic to its own product.
They want it both ways (Score:2, Interesting)
I have mixed feelings about this. If you create a copyrighted work, and you post it on the web publicly, you are implicitly giving people permission to make and use copies, because you can't view the content any other way.
The web has a mechanism for indicating that you want your work not to be used by other sites or bots: robots.txt. If these companies want to keep their work private, they should indicate so in robots.txt. The problem is, they want it both ways. They want web search engines to see their con
Re:They want it both ways (Score:5, Informative)
b) You seem to be suggesting that they are actually encouraging indexers to consume their content? I don't really follow what point you're trying to make of that.
c) robots.txt is not security protocol or something, it is just asking nicely. Some crawlers respect it, but AI scrapers completely ignore them.
Re: (Score:1)
Indexers do, and always have, consumed the content that they index. They store it, in many parts unaltered, in their own database, in order to produce a new product (a search engine). Those search engines also reproduce that content by showing thumbnails and sentences that are taken from the original content. I agree with the AI companies' assertions that what they are doing, is a new kind of indexing.
Re: (Score:2)
a) None of this is available "publicly" on the internet
You are making assumptions about how the material was acquired. Copyright protections do not protect against learning for the ability for transformation. There's a reason your FBI warning at the start of a DVD doesn't say: "FBI Warning: If you remember any of the content of this movie in any capacity you will go to jail." Copyright has limits.
Re: (Score:2)
The web has a mechanism for indicating that you want your work not to be used by other sites or bots: robots.txt.
The mechanism is copyright protection. Copyright exists because creators want to share their works under their own terms, and they don't want someone else to take away control of said works. This can be for a profit motive or for other, personal reasons.
If people or companies kept their works to themselves only, there would be little to no reason for copyright at all.
Re: (Score:2)
It's true, and courts have held that AI training is fair use, as long as no laws were broken in obtaining the material. https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fnatlawreview.com%2Fartic... [natlawreview.com]
Criminal, not civil (Score:2)
When are the CEOs, whose leaked emails *said* that it was "too much trouble" to pay royalties, and trained on stuff from stolen archives, going to be held *criminally* liable for receiving stolen goods?
Re: (Score:2)
Oh, to answer your question, they won't.