On February 22, 2023, the US Copyright Office made a long-awaited decision on the fate of Kristina Kashtanova’s copyright for their graphic novel, Zarya of the Dawn (2022). The drama began last Fall when Kashtanova received a copyright registration for the comic book despite having failed to disclose that images contained in it were generated with the assistance of the artificial intelligence image generator Midjourney. Following Kashtanova’s announcement on social media that they had received a registration for the entire work, including the AI images, the Copyright Office backtracked on the registration and sought further information from the artist as to the details of their authorship.
After receiving additional information from Kashtanova’s attorney, the artist was granted protection for the text and compilation of the comic book, but not for the images. The Copyright Office determined that, despite Kashtanova’s labor in crafting prompts and selecting the final images to be used in the comic book, based on prior case law and established policy standards, “the images… generated by the Midjourney technology are not the product of human authorship.”
The Copyright Office therefore restated its established position that it “will refuse to register a claim if it determines that a human being did not create the work.”
The Office justified its decision on the basis that prompts are more like suggestions from which the AI then makes its own unpredictable choices. As a result, the process is not controlled by the artist and so, from the Office’s perspective, Midjourney is the real author of the images generated. And because Midjourney is not human, no copyright can be conferred. It was on this basis that the Office determined that the artwork incorporated into Kashtanova’s comic book is not subject to the protection of copyright.
This decision suggests that, for the moment at least, all artists who rely on generative AI platforms to produce images will fail to receive copyright in those images. However, it could also have implications for any artistic process that generates images in an unpredictable way.
In contrast to technologies like photography or Adobe Photoshop, which the Copyright Office regards as “tools” and therefore sufficiently precise to confer authorship (and therefore copyright) on the artist, it is the “unpredictable” nature of Midjourney’s creative process that sets it apart. Of course, by distinguishing Midjourney for its lack of predictability, the Office sets the stage to exclude other forms of art that rely on chance in the future.
It is worth stressing that, in this instance, Kashtanova’s application had not sought to register the prompts themselves but rather the images generated by Midjourney. But even if the prompts and their specific sequencing are found to be protectable, the Copyright Office’s stance disqualifying AI as a “tool” with which to create copyrightable work warrants scrutiny. If multiple people input exactly the same prompts into an AI, the AI will come up with different creations every time. Any of the resultant outputs can of course be further modified by the artist, but how much of the final product is due to the artist shaping the output to their own vision and how much is a response to the AI’s choices? Does it even matter?
In this instance, the Copyright Office’s decision leaves the door open for artists to pursue protection for their prompts.
Given that previous case law establishes that the amount of creativity needed to justify copyright protection is very low, it seems likely that even minimally creative prompts to an AI image generator may in fact be copyrightable. As the field of prompt engineering explodes into a new and potentially valuable skill set, AI is quickly becoming inextricably entwined with the creative process throughout the creator economy. How it is manipulated therefore represents a new frontier both artistic and legal.
In the Kashtanova decision, The Copyright Office concluded that the amount of control the artist exercises over the AI outcome is not enough for the AI to be treated as a tool of creative expression. But this sounds suspiciously similar to the arguments made 150 years ago as to the validity of photography as a copyrightable medium. Back then, the recognition of photography as a “tool” hinged on how effective it was in facilitating the expression of an artist’s intentions. The same analysis can be applied here, not least because the Copyright Office relied on the ruling of the Supreme Court in the case of Burrow-Giles Lithographic Co. v. Sarony.
In that case, the Court confirmed photographer Sarony’s copyright in his photograph of Oscar Wilde based on its finding that the photograph was “an original work of art, the product of plaintiff’s intellectual invention…” In explaining its ruling, the Court stated that “[t]he only reason why photographs were not included in the extended list in the act of 1802 is, probably, that they did not exist, as photography, as an art, was then unknown…”
In refuting Burrow-Giles’s argument that Congress did not have the right to revise the copyright law to include protection of photographs in the first place, the Court confirmed that “the constitution is broad enough to cover an act authorizing copyright of photographs, so far as they are representatives of original intellectual conceptions of the author [emphasis author’s own].” Therefore, contrary to the Copyright Office’s use of this historic case to support its conclusion that AI does not qualify as a “tool,” the Burrows-Giles Court clearly supports the opposite conclusion. Indeed, the Court relied on instructive language from an even older English case to define “author” as “the person who effectively is as near as he can be the cause of the picture which is produced.”
Certainly, an artist using an AI image generator is both the person who is, as near as any person can be, the cause of the picture which is produced.
Owing to the iterative process of refining prompts, the final outcome is also representative of the intellectual conceptions of an artist. Given the artist’s ability to use AI to generate images that meaningfully reflect their creative intentions, AI is as much a tool as a camera that captures a photographer’s artistic vision.
Another dubious conclusion of the Copyright Office is that an unpredictable process, mechanical or not, necessarily negates an artist’s ability to express intellectual conceptions. We know and accept that even if a sculptor applies the same chisels and techniques to every block of marble, the process is unpredictable. As a result, each sculpture is always unique. Likewise, every time a painter applies paint, the result will be distinct and different. It is the very unpredictability of an artist’s approach and the media they use that we are referring to when we invoke the value of the artist’s hand, or indeed the aura of art. The artist Bob Ross is widely credited with encouraging a generation of painters with the words: “[i]t’s the imperfection that makes something beautiful, that’s what makes it different and unique from everything else.”
Perhaps it is the very unpredictability of AI that sustains art in the age of human-machine interaction.
If we are willing to accept the imperfect expressions of human hands, why not afford the same latitude to the unpredictable outputs of algorithms as they respond to human guidance? Either way, the simple reality is that artists are flocking to AI as a tool to help them realize their creative intentions. The law simply needs to catch up!
We already know that only a human can hold copyright, so if the original author is not human, no underlying copyright is created, and thus no valid copyright can be transferred to or claimed by anyone else. Relying on this fact, the Copyright Office pre-emptively foreclosed a potential workaround whereby a human user could claim authorship by treating the AI essentially as an employee or contractor. If the AI is treated as the hired help of the artist, then the artist could become the ultimate owner of the final work through a pseudo-work-made-for-hire structure. As the law currently stands, an employer (another person or corporate entity presumably operated by humans) obtains a valid copyright interest in work made by people working for the employer. This is called work-made-for-hire.
If treating AI as a tool does not respect the level of autonomy exercised by the AI, perhaps it is more appropriate to treat the AI as an employee or independent contractor, able to hold a provisional copyright interest subject to its eventual transfer to a human artist.
Deeming the AI to be “employed” by its human user could allow for a limited use of work-for-hire concepts to transfer all intellectual property rights to the human user, thereby allowing a human artist copyright protection for the AI-generated expression.
Article 1 Section 8 of the US Constitution provides Congress with the power to create limited protections to encourage the development of the arts and sciences. Though up to now only pertinent to human expression, the artist and legal communities would be wise to use these established constitutional principles as a touchstone. By staying grounded in the constitutional origins of intellectual property protection frameworks, we can ensure continuity and scalability as we explore whether and how to expand protections to include creative expression resulting from human interactions with AI.
This line of reasoning at the very least harks back to the Burrow-Giles Court’s decision, which reaffirmed Congress’s powers to expand the definition of copyright. Whether this opens the door to treating AI as a creative tool or as an employee or independent contractor, we must bear in mind that our choices will also have an impact on the international intellectual property community.
AI has already advanced to the point of demanding that we question not only the laws we have constructed to protect humans and their creative expressions, but what it means to be human in the first place.
So much, if not all, of what we think, imagine, and create is based on concepts, words, observations, and teachings of others. These are our own personal data sets, while AIs are machines of human invention. As we adjust our existing legal frameworks, we must reconsider what it means to “think,” to “create,” and to “imagine.” Are the programmed functions of AI algorithms simply mimicking human thought, or do they now come so close to actual thought processes that it is no longer possible (or rational) to distinguish human thought from an AI’s algorithmic calculations? Moreover, how do we use the law and policy decisions to reconcile the gaps? With new developments coming at a blistering pace, it is likely that these questions will become more complicated and urgent before we have any real answers.
The Copyright Office’s Kashtanova decision offers some guidance on what artists can do to make their AI-generated artworks more likely to be copyrightable. Perhaps the most helpful is that if an artist makes “substantive edits to an intermediate image generated by Midjourney, those edits could provide human authorship and would not be excluded from [registration].”
Three further takeaways for artists are:
With the Kashtanova decision, the Copyright Office made its first foray into AI thoughtfully, albeit predictably on the conservative side, and with an open acknowledgment of the uncertain future. It also conceded the possibility that different analyses would apply if it is shown that other AIs work differently from Midjourney. That openness will be crucial as we navigate the avalanche of mind-bending legal, ethical, moral, philosophical, and technological issues on the horizon.
Sarah Conley Odenkirk is a Partner at Cowan DeBaets Abrahams & Sheppard LLP, and Co-Head of the Los Angeles office and the Art Law and NFT Practice Group. She is also an Adjunct Professor at the University of Southern California Law School teaching Art Law; and serves on the board of directors of the Andy Warhol Foundation for the Visual Arts.