Are AI tools artists or infringers? Court decides
In a landmark moment for the interplay between art and technology, the U.S. legal system today delved into the murky waters of copyright challenges posed by AI-generated artistry. As AI increasingly borrows inspiration from the annals of human creativity—often sans formal authorization or acknowledgment—a pressing question emerges: Where does copyright stand amidst this?
Stepping into this evolving discourse, U.S. District Court Judge William H. Orrick of the Northern District of California issued a defining judgment. At the heart of the matter were Stability AI, heralded for its seminal Stable Diffusion software; Midjourney, an AI image generator leaning on similar tech; and DeviantArt, the global image-sharing behemoth, which launched its AI-infused "DreamUp" in 2022. Challenging these tech titans in court were three artists: Sarah Anderson, Kelly McKernan, and Karla Ortiz, whose work lies at the epicenter of this debate.
Dismissal motion upheld
The trio of AI image generator firms submitted a motion to have the copyright infringement allegations by the artists dismissed. Today, Judge Orrick primarily accepted this motion, pointing out numerous defects in the complaint. Orrick elaborated on the reasons he deemed the artists' claim as faulty, notably highlighting that two of the artists, McKernan and Ortiz, hadn't officially registered their art with the U.S. Copyright Office.
Furthermore, out of the numerous works mentioned in the artists' grievance, only 16 had copyright claims by Anderson. The artists had argued that a portion of their images found their way into the vast LAION database, a colossal collection of images crafted by computer scientist and ML researcher Christoph Schuhmann along with his team. This database was the training ground for all three AI art generator tools.
Database's protective shield
Orrick points out a protective aspect of the LAION database for the AI companies, suggesting:
“The other problem for plaintiffs is that it is simply not plausible that every Training Image used to train Stable Diffusion was copyrighted (as opposed to copyrightable), or that all DeviantArt users’ Output Images rely upon (theoretically) copyrighted Training Images, and therefore all Output images are derivative images.”
Even if specifics are given, and even if the plaintiffs fine-tune their claims to focus on outputs derived from copyrighted images, I remain unconvinced about copyright claims rooted in a derivative theory without showcasing 'substantial similarity.' The reference cases seem to agree that the accused derivative should still somewhat resemble the original or possess its protected aspects."
Simply put, the vast array of artistic references utilized by AI image generators complicates the claim that a generated image infringes upon any specific copyrighted artwork, unless it predominantly mirrors that particular original piece.
Orrick has provided an opportunity for the artists to refine and resubmit their claims, focusing on specific copyrighted images they believe were infringed upon. However, the judge did permit one claim to proceed — the direct copyright infringement accusation against Stability AI for unauthorized duplication of Anderson’s 16 protected artworks.
AdvertisementJust in: NDCA Judge Orrick has issued his long-awaited ruling on defendants’ motion to dismiss in the closely watched AI copyright lawsuit filed by a group of artists against Stability AI, Midjourney and DeviantArt. Full ruling here: https://t.co/KEsAC2SwR9
The highlights:
1/— Aaron Moss (@copyrightlately) October 30, 2023
Photo and video upscaling/enhancement using AI is excellent. Most AI generated images I’ve seen are stupid. The “artists” winning competitions produce images that mostly have no artistic merit. Another renaissance group surrounding a portal to a sky with planets and moons so large they all would have collided long ago. Tee shirt art for the most part. Art’s in the eye of the beholder; this stuff has a long way to go IMO.
There are painters who can replicate a masterpiece so well the paint has to be chemically analyzed, whatever sophisticated method, to tell the difference from the original. They only get in trouble when the copy is claimed to be the original.
In any art, sculpture, architecture, the original has value; copies far less. I guess there is a market for AI generated “art,” but I see it as tangible as the NFT craze that lost so many people so much money.
Put something online and it’s there for anyone to use. I don’t see how the data collectors can be dinged for what they’re doing; AI images are so easy to copy, they have no significant value.
This is possibly the worst part of this judges comments.
“The other problem for plaintiffs is that it is simply not plausible that every Training Image used to train Stable Diffusion was copyrighted (as opposed to copyrightable), or that all DeviantArt users’ Output Images rely upon (theoretically) copyrighted Training Images, and therefore all Output images are derivative images.”
It should not be about whether every (in legalese meaning all, without exception) training image being copyrighted. That would be trivial do to defeat, by adding a non-copyrighted image to the training set, even if that means it is the only non-cypyrighted set.
It should be whether none, or to within a reasonable margin of error, are copyrighted or otherwise used without permission.
It is rather simple. In the absense of data indicating copyright absense or lack of permission (licensed), then it must not be included in the training set and the synthesis set.
But that is naturally not was these landgrabbers want. They want to indiscriminately bulk harvest and not have to go through the process of determining if the material should be off limits until approved.
It’s the usual “in by default” with the added twist of no opt out, because they will deny having your stuff or claim to have gotten it via other means that they will claim were not copyrighted, like someone reposting a cropped/downsized version or taking a photo of your art.
And on top of this there are some judges, like this one:
“highlighting that two of the artists, McKernan and Ortiz, hadn’t officially registered their art with the U.S. Copyright Office.”
That’s not how that works. Enormous heaps of stuff are copyrighted without going through the process of registering it there. They could not possibly handle all the material.
Nor do I think they would be able to handle the enormous volume of copyright status requests that it would take to deal with a ML Image Training set, specially considering they have been also hoovering something close to every available image on the internet.
So no, that is NOT a valid argument either.
I assume the judge is not that stupid, so he understands how wrong it really is on both counts. He likely also understands how big a slowdown it would be for (US based) AI companies if he judged properly. And that slowdown is currently very unwelcome, for fear of chinese getting ahead.
So a lot of “little people” will get their rights violated and the law broken. If it was Disney or some other large IP holders (RIAA and so on), it would be different, but a few individual artists will just get stomped and denied, not even a bit of “here is some shut up money, now shut up, we need to rush to MAGA this”.
The whole copyright infringement thing is BS in my opinion. I could careless if people were using copies of Microsoft software, movies or pc games. Evil company like Microsoft deserves to lose most of their money for colluding with the government. I support independent developers, not evil corporate conglomerates like Microsoft, Apple or Google who are pro censorship and anti free speech. Also I will not go a long with any company who supports the woke looney Marxist chit either. Those sick freaks deserve to go out of business for trying to destroy society.
If you are in bad faith, and you are not an artist, you copy others, otherwise you create your own style the used tool is not important, this has always happened in the art world.
How many people have remade Pollock’s dripping?
How is this ultimately any different from an art student studying existing art over their years in school, and then creating his own works?
If these sort of claims were legitimate, why have we never seen lawsuits against art schools, alleging they are nothing but copyright infringement factories?
Judges barely know how to turn a computer on, but they will decide that… lol
Most people believe that Artificial Intelligence is anything amazing, when it is like giving power to the dumbest person in class that just used internet search to write a project and will never grow from that because it is just algorithms written by some ‘master’ that tells it what to say.
AI contradicts itself, and argues over dumb stuff, can’t even summarize things without stating an opinion, or all works look the same.
Yes, it will always steal work, unless it was limited to just grab ‘not-copyrighted’ sources, and then, it is still not going to be intelligent nor great to use, it will look the same, write the same, and do the same.