A.I. Copyrights May Be Moving in the Right Direction
In the wake of the internet becoming "Ghiblified", the U.S. at least is moving forward in a helpful direction with answering what can be copyrighted if made with A.I.
In 2022, the art world experienced its first true inflection point with A.I. when Jason M. Allen submitted “Théâtre d’Opéra Spatial” (pictured below) to the Colorado State Fair’s art competition in 2022—and won. Allen had created the artwork using Midjourney, a generative AI platform for digital images.
This sparked much controversy. The Colorado State Fair upheld the win, stating that Allen didn’t break any rules by submitting an A.I. generated piece. However, they changed their rules the following year.
In the meantime, Allen tried to get the piece copyrighted, and the U.S. copyright office denied his application. He then filed a lawsuit against them in 2024, seeking declaratory judgment to overturn the decision and clarify legal standards for AI-assisted work. This week, the U.S. Copyright office has done just that and further officially clarified its standards. I think it’s a step in the right direction.
In short: they are upholding their decision that human authorship and creativity are essential for copyrighting a work. More details below.
A Ghibli Firestorm
It’s come at a good time, at least for artists in the U.S.
Last week, another firestorm hit the art world when Sam Altman (CEO, OpenAI) showed off how you can now create anime art out of any picture you upload to the latest iteration of ChatGPT. Soon, people were “Ghiblifying” all sorts of images, and even movies—having ChatGPT emulate the art style of the renowned Japanese artist Hayao Miyazaki from Studio Ghibli.
The irony of allowing people to generate images using Studio Ghibli’s style is not lost on fans. Under Miyazaki’s direction, the studio has been renown for its unflinching commitment to hand-drawn artistry and emotional storytelling, delighting people of all ages around the world. To have A.I. emulate art that was always meant to be deeply rooted in the human experience seems like a massive slap in the face to the very thing the A.I. is being used to emulate. The last week the art community has been in uproar, seeing this all as a massive ‘cheapifying’ of something rare and special, as the masses almost crashed ChatGPT’s server in creating “Ghiblified” images of anything and everything, including internet memes.
Miyazaki himself famously said in 2023 that he is “utterly disgusted” by A.I., saying even further: “I strongly feel that this is an insult to life itself.” (Worth noting he was shown a real bad example.)
Is all this fair to artists? This has been the question many have been asking the last few years. I’ve written on this subject regularly. I think that, in essence, A.I. companies have run roughshod over copyright laws and quite frankly shown tremendous disrespect to artists. But of course, there is the argument that all of this functions mostly as a gimmick—a Snapchat filter, really, that ultimately doesn’t mean anything more than having a bit of fun.
This is now perhaps the direction things are going with copyright laws.
The copyright trend as of now
The full breakdown of the U.S. Copyright’s office can be read at Reuters, but I will try and summarize it here and offer an opinion on what it means (and why I think they’ve done the right thing).
A. Copyright requires human authorship
By making this move, fully A.I.-generated art is automatically cheapened, rather than cheapening human-created art. Authors therefore uploading tons of A.I.-generated books to Amazon and flooding the market with ‘slop’ will therefore not be able to copyright their works, throwing into question whether or not they could even sell it.
The upshot is it changes the demand and how the tools are used. It makes “ghiblifying”, for example, merely a gimmick, which is fine (people should be able to have fun with art).
B. It must contain some degree of originality and cannot just be the result of time and effort
Good news as this addresses those who believe that by sitting and asking the A.I. to regenerate a new work until something good comes around counts as sufficient creative work.
C. A.I. can be used to assist in the creating process, but a fully-generated work cannot be copyrighted.
A.I. is here to stay, and it can be a very useful tool when the ethics are aligned to humanity. So this is a reasonable road to take. “Ghiblifying” something can be wonderful. Consider how the internet can become a more interesting place if our websites can adopt such a style. However, the next piece allows for the power to remain with the original artist.
D. If a person inputs their own original work into the A.I. and it generates a variation, as long as the variation can be perceived as linked to the human’s original work, it can be copyrighted.
This is good news as artists can therefore scale up their own work and offerings. For example, I could input my fantasy novel, When Twins War into the A.I. and ask it to generate some sort of board game out of it. Then I could copyright the resulting output.
This puts value back into the original work. “Ghiblifying” images therefore is really only a “snapchat” filter. Using it commercially, therefore, will require going back to the artist and getting the rights. That makes A.I. merely a tool to get something done and puts the power back into the hands of the original creator of the work.
E. Prompts alone are not enough
Some have argued that they ought to be able to copyright A.I. generated work because they came up with a unique prompt. This, however, is not enough. Good news as it means sufficient creative work must be done.
How it looks around the world
The E.U. seems to be going the same route as above.
The U.K. still seems to be working it out.
Japan is looking at it “case by case” but an article at Yankodesign claims the copyright laws are one of the reasons why Ghibli’s work could be used as above.
China (not known for strong copyright laws) simply states that “A person who uses AI to create the work is the author of that work.” This, of course, is much the opposite as above.
South Africa is yet to come to a firm conclusion. It may land somewhere in the middle. See here.
So where to now, Western Man?
Most of the discussion around A.I.-generated art has revolved around whether a large language model (LLM) such as ChatGPT is legally able to ingest copyrighted work under ‘fair use’ as allegedly LLM’s do not store the data but simply ‘learn’ from it.
I disagree it is fair use in principle because what is ‘learning’ the work is not a human being but a machine designed to ingest and then spit out variations, at scale, for a for-profit company.
In order to ‘learn’ or be ‘inspired’ as a human being, I have to always put money up front. I have to rent a movie or pay a streaming service. I have to buy a book or pay taxes for a local library. I have to pay for music I like. If ChatGPT ‘learns’ like a human does (it doesn’t, because it’s not human) then ChatGPT should have to pay for that inspiration too. But, of course, we have all learned by now that the Big Boys of Tech believe they’re exempt from any trivialities such as ethics.
Anyway, before I get too cynical about this, the good news is that dealing with the copyright issues on the end of the process some precedent can be set for the rest of the process. Making the output uncopyrightable except under circumstances that favor the creative process is a fantastic step at leveling the demand for the tools and how they’re used.
There are a lot of lawsuits waiting in the wings that will bring answers (and stability) to this whole ordeal. Obviously, the likes of Sam Altman are betting on using the argument that the tools have become so useful and so ubiquitous and so important even for “national security” that they obviously shouldn’t be expected to pay for copyrighted material as that will hurt “progress”. (Of course, progress in establishing proper ethics is also a type of ‘progress’ that tends to get conveniently ignored.)
Read more about Sam Altman’s statements at Ars Technica.