AI’s role in copyright is growing clearer. For the first time, a federal court has ruled on copyrighting an AI’s work, holding that AI cannot have a copyright. Generative AI models, complex algorithms that can generate “original” content through words or images, have shown rapid improvement over the past few years. This has raised the question of copyright – who owns the rights to an image created by AI? The answer to this question has been of great concern to the field of generative AI, as copyright protects the creator of a work in many ways, primarily by ensuring ownership of ensuing profits. Given recent rulings, it seems quite likely that the future test for creating a copyright for an AI-generated piece of art will require some sort of transformative treatment applied by a human author, though it remains to be seen how stringent the standard will be.
Copyright Basics
Simply put, copyright is a protection against others using your creative work without permission. Copyright protects original works of authorship as soon as the author fixes the work in a tangible form of expression. This includes pictorial and graphic works, such as pictures and literary works. An owner of copyright has many protected rights, including the exclusive right to reproduce the copyrighted work, make new works based on the copyrighted work, and sell or distribute the work. In other words, as soon as an author creates a new work, such as an image or some text, they have exclusive ownership, and own the right to profits from that work.
Consequently, copyright protections are essential for any artist or creative hoping to make money from their work - not having a copyright on a creative work allows anybody to reproduce, copy, or otherwise rip off that work, and keep all of the profit. The film industry is a testament to how important, and lucrative, copyright protections are. Disney, for example, has fought tooth and nail to retain their valuable copyrights, being a significant contributor behind several changes to the duration of copyright. Notably, Disney was such a big part of the 1998 Copyright Term Extension Act, extending the duration of copyright by a further 20 years, that some have nicknamed it the “Mickey Mouse Protection Act.” If works created with the help of AI cannot be copyrighted, businesses will be unable to use generative AI technology for creative works, regardless of how powerful and cost-effective it becomes.
U.S. Copyright Office Speaks Twice
So far, the U.S. Copyright Office has denied a copyright to content generated by AI on two separate occasions. First, in 2018, Stephen Thaler applied for the copyright to an image titled A Recent Entrance to Paradise, which was generated solely using a generative AI model created by Thaler himself. Thaler contended that the human authorship requirement was unconstitutional and unsupported by either statute or case law. However, the U.S. Copyright Office declined Thaler copyright ownership, maintaining that copyright protections apply only to work created by a human author, citing precedents from the Supreme Court and lower courts. Notably, while Thaler had created the generative AI model himself, crafting prompts for the model to generate the image, he did not argue that the image was anything other than purely AI-generated. After receiving the U.S. Copyright Office’s final decision in February of 2022, Thaler filed suit in the U.S. District Court for the District of Columbia to challenge the ruling.
Following Thaler’s attempt, in February of 2023, the U.S. Copyright Office partially revoked copyright for a graphic novel which used AI-generated art. Zarya of the Dawn, by Kris Kashtanova, is a graphic novel created from images generated by an AI model called Midjourney, to which Kashtanova added dialogue. The U.S. Copyright Office viewed this work differently than Thaler’s because Kashtanova authored the work’s text, and selected and arranged the images in the work to create a compilation, both of which qualify as copyrightable works under 17 U.S.C. § 101(a).
However, the Office did not grant Kashtanova copyright of the images themselves, because the “traditional elements of authorship” originated from Midjourney. They reasoned that the author of a copyrighted work is the individual who actually formed the picture, acting as “the inventive or mastermind.” Because the generative AI model’s outputs were relatively random, merely influenced by the words that Kashtanova provided, there was a “significant distance” between her directions and the model’s final product. The U.S. Copyright Office distinguished AI from tools that artists regularly use, as those tools allow an artist to give visible form to an already-existing mental conception. Conversely, in the Copyright Office’s view, an AI model visualizes the image and creates it entirely on its own, with the artist merely influencing the image somewhat through the use of prompts. Importantly, the Office noted that, had Kashtanova made substantive edits to an image generated by Midjourney, there would have been human authorship for the image, and it could have been granted copyright.
Recent Decision - Thaler v. Perlmutter
Finally, on August 18, 2023, the U.S. District Court for the District of Columbia ruled on Thaler’s case, granting summary judgment to the U.S. Copyright Office. The court confirmed the U.S. Copyright Office’s decision, stating that “[t]he 1976 Act’s “authorship” requirement as presumptively being human rests on centuries of settled understanding.” In other words, no valid copyright has ever existed in a work without a human author, and thus, the work of Thaler’s AI model has no copyright to register. The court also addressed Thaler’s policy arguments, noting that, unlike human authors, non-human actors do not need to be incentivized with the promise of exclusive rights, and explaining that, based on legislative history, Congress never intended for the Copyright Act to include non-human authors. While this decision is at the federal trial level, and is not binding across all federal courts, it seems highly persuasive, and other courts are likely to follow this precedent.
Going Forward - What's in Store?
While the Thaler holding was fairly straightforward, it is still unclear where courts will draw the line when it comes to copyrighting human works that make use of AI tools. The copyright granted to Zarya of the Dawn seems hopeful for AI. In their letter, the U.S. Copyright Office mentioned that the images used in Zarya did not receive copyright because they were taken straight from an AI model, and had the images been edited by Kashtanova, with “significant” changes, they may have been eligible for copyright. However, the Office did not make the factors behind their reasoning particularly clear. The Copyright Office’s letter deemed Kashtanovka’s touch-up of one image’s lips to be too insignificant, while saying that fully editing a character’s face had the potential to be sufficient, depending on the image. This implies some sort of internal standard was applied to determine whether changes to the AI’s work were sufficiently substantive. Whether this standard is lenient remains to be seen. With Thaler planning on appealing his case, and Kashtanova designing a new work to test the limits of human authorship, courts will have ample opportunities to further analyze AI’s role in human authorship.
*The views expressed in this article do not represent the views of Santa Clara University.