The U.S. Copyright Office took the easy way out when it decided recently that images created with artificial intelligence are not entitled to legal protection because they don’t require sufficient human creativity. That is disappointing to people who are producing highly innovative work with AI. If they make their images publicly available, they can be stolen by anyone and used for any purpose.
The C.O. ruled in the case of Kristina Kashtanova, a New York City artist who used the AI platform Midjourney to produce the images in an 18-page comic book, “Zarya of the Dawn.” After initially granting permission to register her work — which is required for an infringement lawsuit and serves as a deterrent against illegal use — the C.O. rescinded approval for the images when it learned from her social media posts that they were created with AI.
But the C.O. didn’t stop there. It imposed a new standard — unsupported by the Constitution, statutes and case law — to justify the denial of copyright protection and discourage those using AI to apply for it.
In a letter to Kashtanova’s attorney, the C.O. said her images cannot be registered because she could not control and predict how Midjourney would respond to her word prompts. Such a “control-and-predict” rule had never been required for registration. The C.O. also issued a policy statement requiring applicants to declare whether the images were created with AI so those can be excluded from copyright protection.
The C.O. said a human and not a machine must be primarily responsible for creating a copyrightable image. It cited an 1884 decision, Burrow-Giles Lithographic Co. v. Sarony, in which the Supreme Court ruled that a photograph of author Oscar Wilde involved human creativity and was entitled to copyright protection even though a camera captured the image.
Sarony’s photo shows the talent of an experienced photographer. Nevertheless, he relied on a mechanical device the way Kashtanova used a computer and Midjourney to create the images she envisioned for her comic book.
Burrow-Giles was decided almost a century and a half before AI became widely available and is obsolete as a precedent, at least as applied to AI. That may be why the C.O. didn’t explain well why taking a photo was enough for legal protection but Kashtanova’s design plans and the hundreds of prompts she wrote over many months were not.
The C.O. also cited a 1991 Supreme Court decision, Feist Publications v. Rural Tel. Serv., where the court said that “only a modicum of creativity is necessary” for copyright protection. But the C.O. then brushed aside Kashtanova’s imagination and computer skills by endorsing Feist’s statement that the “sweat of the brow” — how much time and effort were spent creating an image — is irrelevant.
The C.O. said that “Rather than a tool that Ms. Kashtanova controlled and guided to reach her desired image, Midjourney generates images in an unpredictable way. Accordingly, Midjourney users are not the ‘authors’ for copyright purposes of the images the technology generates.” And the C.O. added, “The fact that Midjourney’s specific output cannot be predicted by users makes Midjourney different for copyright purposes than other tools used by artists.”
This ignores two issues: An AI creator can get close to what they are looking for with the right artistic descriptions and technical commands (and sometimes a reference image) and can build on the portions of the prompt that are working while trying different words to fix what is not. Also, AI managers are constantly improving how closely images follow directions.
If the C.O. is going to use “control-and-predict” to evaluate AI images, it must apply it to photographs or else it looks like the new standard was created just to keep out AI. But then, the analogy breaks down. Digital cameras can capture multiple photos per second. If the subject — a person, a pet, a waterfall — is moving, each of the photos will be different. The photographer, however, cannot control and predict in advance what a particular photo out of hundreds taken will look like.
To comply with the C.O. rule, a photographer would have to take one photo at a time. After each one, either the subject or the camera must move. Only then, when the proposed image is under “control” and the photographer knows what the camera will capture, should the next photo be taken.
This approach doesn’t appreciate the creative process. Photographers, authors, painters and other artists don’t always know how their work will turn out. They may want to experiment and improvise to see where their project leads. Some of the best photographs and paintings are unexpected.
Kashtanova’s attorney, Van Lindberg of Taylor English Duma, wrote a comprehensive and persuasive 13-page letter to the C.O. arguing for registration. He wrote that “All the images … were designed by Kashtanova. The visual structure of each image, the selection of the poses and points of view, and the juxtaposition of the various visual elements within each picture were consciously chosen.” And he added, what she did was “more intensive and creative than the effort that goes into many photographs.”
But it seems likely that Kashtanova never had a chance. The C.O. receives half a million registration applications per year, a staggering 10,000 per week. It can take three to nine months for an application to be reviewed and, if approved, for a certificate to be issued.
The backlog would be worse if AI images could be registered. If there were no general rules to apply, the staff would have to determine the extent to which there was “human involvement” in the creation of the image. A decision could take a year or more.
The C.O. is not going to welcome registration of AI images. Congress or the courts will have to take the initiative. Meanwhile, those creating artistic images with AI know that if they share their images, they will have no control over how they are used.
____
(Richard Labunski (labunski@uky.edu) is a retired journalism professor from the University of Kentucky and author of “James Madison and the Struggle for the Bill of Rights.”)
___