In The Age of AI: And Our Human Future, Henry A. Kissinger, Eric Schmidt, and Daniel Huttenlocher opined that AI is shepherding a world where decisions are made by humans, by machines, or through an “unfamiliar but also unprecedented” collaboration between them. As co-authors ourselves, we know firsthand that it can be difficult to ensure that human collaborators feel their voices are fairly represented, or that labor is shared equally. But why is creative collaboration between humans and machines (qua Artificial Intelligence) so difficult?
The current US legal analysis, based on developing copyright precedent, suggests that the answer hinges on the degree of human involvement. A critical assessment of whether human involvement is measured qualitatively or quantitatively requires both philosophical inquiry (evoking the words of Walter Benjamin, who proclaimed: “that which withers in the age of mechanical reproduction is the aura of the work of art”), as well as a practical policy analysis.
The issue of non-human authorship is not new to the US Copyright Office. Prior to considering AI, the Office notably denied registration in a 2018 headline-grabbing case involving a simian “selfie.” Fast forward to February 2023, when the Office granted Kris Kashtanova, author of the graphic novel Zarya of the Dawn (2022), copyright protection for their book’s text and compilation, but not the AI-generated graphic components. In its letter to Kashtanova, the Office explained that the copyright registration for Zarya would not extend to the images created with the assistance of Midjourney, because the Office only registers works of original authorship created by human beings. In its view, Kashtanova did not sufficiently control Midjourney’s creative process, and the tool’s output could not be sufficiently predicted—thus, the Office concluded that the images were authored by the AI, not by Kashtanova. (For more detail, refer to Odenkirk’s article, “You Can’t Copyright AI Art (Yet)”.)
In more recent guidance, the Office further explained that when reviewing the registration application for a creative work containing AI-generated material, it will consider whether the AI content is simply the result of “mechanical reproduction,” or whether instead, it gives “visible form” to the human author’s “original mental conception.” If the AI, and not the human, determines the expressive elements of its output, the Office will conclude that the generated material is not copyrightable.
However, the Office also noted that, as with Zarya, AI-generated material could be incorporated into a copyrightable human-authored compilation; if a human selects or arranges AI-generated material in a sufficiently creative way, the resulting work may constitute an original work of human authorship entitled to “thin” protection (i.e., protection with respect to layout or arrangement, as well as of any other creative human-authored elements, such as Kashtanova’s text for Zarya).
Moreover, the Office further acknowledged the possibility of a human author modifying AI-generated material to such a degree that the result could qualify for protection as a human-authored derivative work. Because authors of derivative works only receive copyright protection in their own sufficiently-creative contributions, this scenario results in the human author being able to enforce their rights only in the specific portions of the work that actually reflect their human touch. However, unlike the nebulous guesswork of determining whether making changes to another human author’s work requires a license or is a permissible “fair use” (a blurry line currently at issue before the US Supreme Court in the Warhol Foundation v. Goldsmith case), a human author adapting an AI work would not need any such permission, since the underlying work would not qualify for copyright protection in the first place.
Recently, Kashtanova sought registration of a different AI-assisted creation, entitled Rose Enigma, depicting a cyborg’s face with roses emerging from the top. In their application to the Office, Kashtanova notes the use of Stable Diffusion 1.5, emphasizing that rather than feeding the AI only text prompts, Rose Enigma is instead based on Kashtanova’s hand-drawn image, with changes that they carefully controlled through use of the AI’s various mechanisms. Drawn to its logical conclusions, this application asserts a provocative legal theory: that Stable Diffusion made a derivative of Kashtanova’s work, not the other way around.
The different factual context for Kashtanova’s new application will require the Office to refine its position regarding the copyrightability of artwork made by human-machine interaction and the requisite level of human touch. And yet the decision as to Rose Enigma will be specific to that work; the Office’s recent guidance acknowledges that registration for AI-assisted works will continue to present novel and fact-specific use cases that necessitate a case-by-case analysis.
Notwithstanding the difficulty that human creators will have in proving the degree to which they dominate AI-generated output, we are at a tipping point. Artists must consider what it means to use AI as a tool, and how to prove that their final output qualifies for copyright protection. More radically, artists, scholars, and policy makers might also consider (as Amy Adler has) whether copyright protection even remains relevant. The Kissinger, Schmidt, and Huttenlocher text argues that continued development of this technology is not only changing thought, perception, and knowledge, but expanding the notion of reality itself. Perhaps the US copyright framework is due for revision to reflect this new reality. At the intersection of AI and art, the human touch—as it relates to the use of technology—needs to be respected and legally acknowledged, in some form or another.