Who (If Anyone) Owns AI-Generated Content?

A_Recent_Entrance_to_Paradise-300x225Whose content is it anyway? This is one of the questions that many hope will be answered by a federal court in Thaler v. Perlmutter. In June 2022, computer scientist Dr. Stephen Thaler sued the U.S. Copyright Office to redress the denial of his application to register copyright in his AI system’s visual output under the Office’s “Human Authorship Requirement.” A few months later, the Office again enforced this requirement, reversing its decision to register Kristina Kashtanova’s illustrated comic book, Zarya of the Dawn, after it became clear that AI was used to generate those images. While Thaler now asks a U.S. federal court to determine whether an AI system can author copyrightable work, and to effectively overrule the Office’s “Human Authorship Requirement,” it remains to be seen whether the court will tackle those broad issues, or instead narrowly focus on whether the Office had reasonable grounds to deny Thaler’s application.

Thaler is no stranger to the U.S. federal court system. Earlier this year, the U.S. Supreme Court declined to review Thaler v. Vidal, Case No. 21-2347 (Fed. Cir. 2022), a ruling that found Thaler’s Device for the Autonomous Bootstrapping of Unified Sentience (DABUS)—an AI system—was not capable of inventorship because it was not a natural person. Thaler’s instant suit against the Office concerns a different AI system dubbed Creativity Machine, an artificial neural network that can generate two-dimensional artwork in a manner that arguably mimics human creativity. In November 2018, Thaler filed a copyright application with the Office to register Creativity Machine output: a two-dimensional image titled “A Recent Entrance to Paradise” (the “Work”). Thaler listed himself as claimant and Creativity Machine as author, and the Office refused the application because it ran afoul of the Human Authorship Requirement.

In so doing, the Office relied on Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 56 (1884)—a seminal copyright decision involving a photograph of Oscar Wilde that established the copyrightability of photographs when that technology was new. The Office also relied on Urantia Found. v. Kristen Maaherra, 114 F.3d 955, 957–59 (9th Cir. 1997), where the Ninth Circuit held that “human creativity” was required “for a work to be copyrightable,” such that a book allegedly “‘authored’ by non-human spiritual being” was unprotectable. The Office further relied on Naruto v. Slater, 888 F.3d 418, 426 (9th Cir. 2018), the well-known “Monkey Selfie” case, where a monkey’s self-portrait photograph was denied authorship because the Act’s terms “all imply humanity and necessarily exclude animals.”

District Court
Thaler sought judicial relief in the U.S. District Court for the District of Columbia against the Office, and Shira Perlmutter in her official capacity as Register of Copyrights and Director of the Office. In his complaint, Thaler pleads a single count of an “Administrative Procedure Act Violation for Denial of Plaintiff’s Application.” In support of his motion for summary judgment, Thaler argued that the authorities relied upon by the Office were either inapt or misinterpreted, and that the Office’s “Human Authorship Requirement” is contrary to the Act’s express meaning and broader purpose. Specifically, Thaler argued that the Office mistakenly relies on dicta from decisions that pre-date modern computers, and that the Act does not require human authorship, as evidenced by 17 U.S.C. § 201(b), the provision of the Copyright Act that allows works made for hire to be generated by employees but authored by employers (e.g., companies). Thaler further argued that he owns copyright in the Work as a work made for hire because his AI system was an “employee” for purposes of the Act, and was exclusively and entirely controlled, operated and owned by Thaler. In the alternative, Thaler argued that he owns the Work as the “fruit of the tree” or by right of first possession, relying on common law property principles.

The Office cross-moved for summary judgment and filed a response to Thaler’s motion, arguing that its decision was based on settled law and was not “arbitrary and capricious”—the standard for overturning agency determinations. Specifically, the Office defended its Human Authorship Requirement, contending that the Supreme Court “has consistently referred to a human person when discussing the term ‘author’ as used in the Copyright Clause of the Constitution and in copyright statutes,” and relies on the same cases cited in support of its refusal to register Thaler’s application. In doing so, the Office countered Thaler’s argument that 17 U.S.C. § 201(b) demonstrates that the Act permits non-human authorship, reasoning that such provision “only states that the employer or ‘other person’ who commissioned the work is considered to be the author for purposes of the Act, it does not contemplate that any non-human is the literal author of the work.” The Office also responded to Thaler’s assertions that he owns the Work as one made for hire, arguing that a machine cannot be employed by or contract with Thaler, and that the common law principles on which Thaler relies (e.g., governing copyright transfers in intestate succession or corporate dissolution) are inapposite, if not preempted by the Act.

With the parties’ cross-motions for summary judgment fully briefed, a ruling is expected soon.

What Lies Ahead for Thaler
While Thaler’s chances of success in his dispute with the Office are all but impossible to predict, there is good reason to be skeptical. First, it is typically difficult to prove that agency determinations are “arbitrary and capricious” and the same appears true here. Second, Thaler’s application seems at odds with his own arguments; while Thaler identified the Creativity Machine as the author of the Work in his application, he now argues that he owns the Work as one made for hire, which would mean Thaler (not Creativity Machine) was the author; indeed, a court may have trouble siding with Thaler in view of such inconsistencies. Third, Thaler does not argue (and perhaps has waived the argument) that he exercised sufficient creative control over the AI system’s creation of the Work to qualify as its author. Indeed, Thaler does not argue that he used specific text, source code or other inputs to control the creative details of the AI’s output; instead, he largely relies on broad arguments that the Act theoretically does not preclude non-human authorship, and that AI authorship is consistent with the legislative intent.

Though still undecided, we may come to wonder how the case would have turned out had Thaler made this argument. Unlike Kashtanova, who input English language text prompts into third-party AI platform Midjourney to generate images for Zarya of the Dawn, Thaler developed the Creativity Machine himself, and was better positioned to use his technical knowledge of his own system to “control[] and guide[] [the Creativity Machine] to reach [his] desired image[.]” (See Kashtanova Decision, at 9.) One also wonders how a court might view the Office’s 1984 decision to grant a copyright registration in an AI-generated work, The Policeman’s Beard is Half Constructed, which lists the computer program “Racter” as its author. Thaler did not make this point. If the Office saw fit to register Racter’s work, why not register Creativity Machine’s work? Would the decision to register one, but not the other, be arbitrary and capricious?

Even so, that may not be the end of the story for Thaler. Thaler argues in support of summary judgment that the Office’s “Human Authorship Requirement” is inconsistent with allowing a non-human (e.g., a company) to “author” a work made for hire, which, in turn, arguably suggests that the Office’s Human Authorship Requirement is inconsistently (and perhaps even arbitrarily) applied. The Office’s response—that companies can be “considered” authors under the Act, but cannot be “literal author[s]”—raises more questions than answers: where does this distinction between legal authors and “literal” ones come from? Is it supported by the text of the Act? And why is this legal fiction that supports corporate authorship unable to also support AI authorship?

While it remains to be seen how the court will rule, if Thaler’s DABUS case is any guide, Thaler would likely appeal an adverse ruling, and it also seems likely that the Office would do the same. Future cases are also likely to raise more difficult questions about copyright ownership in AI-generated content, particularly with respect to third-party use of commercial AI platforms governed by terms of service. For example, if a user guides and controls an AI platform to generate a specific expressive outcome, in a manner that would be sufficient to satisfy the “test” articulated by the Office in refusing registration of Zarya of the Dawn, but does so after agreeing to terms of service that divest the user of ownership, who is the author of that output? And who owns it? These and other questions have yet to be answered.


A New Dawn for Copyright in AI-Generated Works?

Ability of AI to Invent Struck a Resounding and Uncompromising Blow