The Art of Attribution and Three Unlikely Theories of AI Authorship

by Tim Knight and Dennis Crouch.

In 2018, Dr. Stephen Thaler, creator of the ‘Creativity Machine’ AI system, sought copyright registration for an AI-generated image, listing the Creativity Machine as author. The Copyright Office rejected the application, citing the necessity of a human author under copyright law. After two failed requests for reconsideration, Thaler sued to compel registration. United States District Judge Beryl Howell ruled against Dr. Thaler. The case is now pending before the D.C. Circuit Court of Appeal, with the parties recently presenting oral arguments before Judge Patricia Millett, Judge Judith Rogers, and former patent litigator Judge Robert Wilkins. The graphic work itself is undeniably original and fixed in tangible form – key traditional elements for copyright protection. 17 U.S.C. 102 (“Copyright protection subsists, in accordance with this title, in original works of authorship fixed in any tangible medium of expression”).  And unlike naturally occurring beauty, the work exists only because of human creative endeavors. The rub though is that Dr. Thaler’s human creativity was directed toward developing the AI system, a major step removed from originating the artwork itself.

All the parties appear to agree that the facts in this case are particularly narrow based upon Dr. Thaler’s admission that there was no direct human involvement in the painting’s creation.  Thus, this is not a situation involving direct human-AI collaboration in the creation of a work, but rather a human creating an AI who then creates the work.

The question presented in Thaler is whether an original work generated by an AI system in the absence of a traditional human author is copyrightable. Thaler’s attorney, Professor Ryan Abbott, argued that the Copyright Act does not require a human author and that the work is deserving of registration because the AI “did the thing one traditionally associates with authorship.” At oral arguments, Abbott provided three options for the court to find this work was copyrightable:

  1. Operation of Law: The autonomous machine is the author of the work because it is factually the work’s proximate creator and originator, but Dr. Thaler is deemed the author under the operation of law because he is the owner, user, and programmer of the machine.
  2. Work-for-Hire Doctrine: The autonomous machine is the work’s author, but Dr. Thaler is deemed the author under the work-for-hire doctrine or operation of law.
  3. Indirect But For Originator: Thaler is the author because, while indirect, he is the originator of the work, even though he did not make a direct “traditional” contribution.

As we discuss below, these theories are closely aligned, but each are unique.

Professor Abbott’s first theory posits that the AI system serves as the author by being the proximate cause of creating the work. At the same time, Dr. Thaler acquires ownership through virtue of possession and accession property doctrines. This approach follows traditional property law concepts, suggesting that creation and ownership of the autonomous machine confers ownership of its output. Mr. Abbott did shift his stance slightly in the middle of the argument, saying that Thaler is the owner by first possession. However, conceptually, this theory stumbles against copyright law’s fundamental premise that rights flow from authorship, not mere possession or creation of the tools. Art is not a bundle of sticks or wild foxes. While a person stumbling across an original painting might become owner of its physical form, that does not trigger the broader intellectual property rights sought here.  Similarly, the person who rents out painting supplies does not become the author by accession whenever those supplies are used to create a work. The copyright sphere is independent and unique from traditional property doctrine. Mixing these principles into copyright law to grant authorship to programmers of AI is quite unsatisfying.  Judge Millett also raised significant skepticism against this argument, hinting that the common law property arguments were not fleshed out.

While receiving less attention during oral arguments, the second theory, the work-for-hire approach, represents perhaps the most established legal framework for granting copyright to a person who is not the author. Unlike patent law, where the human inventor is always listed on the patent document, the work-for-hire doctrine allows a legal fiction where employers or hiring parties become the constructive authors.

Works Made for Hire.—In the case of a work made for hire, the employer or other person for whom the work was prepared is considered the author for purposes of this title.

17 U.S.C. § 201.  An important feature here is that employers are often non-human people—corporate employers—that legally become the author through this statutory construction.  Yet, applying this doctrine to AI-generated works stretches its boundaries in unsatisfying ways. The work-for-hire doctrine was designed to handle relationships between “traditional” human authors and their employers, not the relationship between a machine learning system and its creator.  This argument was briefly touched on in oral arguments. In briefings, Mr. Abbott argued that “[t]he level of control, lack of independence, and overall operation and direction of the AI system Dr. Thaler exercised creates, on balance, a clear enough level of control to justify employee status for purposes of copyright.”  However, applying the traditional work-for-hire factors reveals fundamental incompatibilities. For example, there is no method of payment, no employee benefits, no tax treatment, and no discretion over working hours.  Many questions about work-for-hire make little sense if they are at all applicable to AI. But most fundamentally, AI cannot be hired because that is a process predicated on a consensual contractual relationship which can only exist in American law between legal entities — certainly not between a human programmer and an AI. Unlike traditional work-for-hire situations where a human author’s rights may transfer by operation of law to a constructive corporate author, Thaler’s approach would invert this framework. The work-for-hire doctrine always begins with human creativity that is then attributed to another entity through legal fiction. What Thaler proposes is fundamentally different—he seeks to have human  ownership attached to underlying non-human creativity. This reversal undermines any meaningful analogy to the work-for-hire doctrine. While work-for-hire serves to reallocate rights flowing from human creativity, Thaler seeks something entirely different—to conjure human authorship from non-human origins.

The third and most nuanced theory introduces the concept of ownership through Dr. Thaler indirect causation of the origination of the work.  At first glance, this theory might appear identical to the first theory, as Judge Millett raises, because origination and proximate cause are closely related in the artistic sense. However, there are subtle but important distinctions on the programming side.  Under the first theory, the AI is designated as author, and Dr. Thaler is then deemed the author because he owns the machine. Instead, under this causation theory, Dr. Thaler is the author because he set the events in motion to create the work. The strategy attempts to sidestep the traditional authorship requirement by focusing on Dr. Thaler’s role as the originator of the creative process, even if he did not directly originate the work itself. Mr. Abbott took time at oral argument to explain what originator means in the programming sphere and distinguish it from the traditional author. While Thaler’s programming established the framework for autonomous creation, this raises complex questions about the relationship between “but for” causation (Thaler’s creation and initiating operation of the AI system) and proximate causation (the AI’s independent generation of the work). The difficulty lies in determining whether Thaler’s indirect role in originating the creative process is sufficiently immediate and substantial to constitute copyright authorship, or whether the autonomous nature of the AI system somehow breaks the chain of causation that copyright law traditionally requires. Perhaps more problematically, adopting this “but for” causation theory would force the Copyright Office to conduct complex, fact-intensive analyses for each AI-human collaboration—an administratively burdensome approach that would strain the registration system and likely produce inconsistent results.

In sum, all of the three arguments presented by Mr. Abbott have major holes, and each is unsatisfying in its own right. This sentiment seemed to be echoed by the Judges hearing the case.

= = =

The oral arguments in Thaler explored the contours of “traditional authorship” through two illuminating hypotheticals posed by Judge Millett: the Kodak camera and the printer malfunction. These examples help map a spectrum of human involvement in creative works and highlight the challenges in placing AI-generated works within existing doctrine.

The Kodak camera hypothetical probes whether a tool manufacturer could claim authorship rights. While cameras are essential to photography, they remain tools requiring human direction and creative input. As Judge Millett emphasized, Kodak’s role in creating the mechanism for photography places them several steps removed from the actual creative process—they neither compose the shot nor choose the subject matter that traditionally grounds copyright ownership. This stands in marked contrast to AI systems which, at least in theory, autonomously make traditionally authorial decisions about composition, style, and content. Yet this distinction raises a deeper question: as camera manufacturers increasingly incorporate AI features that make creative choices (adjusting composition, selecting optimal moments to shoot), does the line between tool and author begin to blur? The evolution of “smart” cameras may soon challenge our understanding of where creative decisions truly originate.

The printer malfunction hypothetical introduces a different dimension of the authorship inquiry—the role of intention versus accident in copyright doctrine. When a paper jam creates a “smeary image,” traditional copyright principles must grapple with unintended results from intended actions. Copyright law has long recognized that authors need not intend specific outcomes to claim authorship rights, only that they made intentional creative choices in the process. See Alfred Bell & Co. v. Catalda Fine Arts, 191 F.2d 99 (2d Cir. 1951) (“Having hit upon such a variation unintentionally, the ‘author’ may adopt it as his and copyright it.”). But AI-generated works exist in an uncomfortable middle ground—neither purely intentional nor truly accidental and operated through an anthropomorphized intermediary. The causation chain in Thaler’s case raises similar questions: does his role in creating and deploying the AI system constitute sufficient intentional creative choice, or does the AI’s autonomous operation represent a break in the chain that severs any meaningful connection to traditional authorship?

The oral arguments mirror the discussion explored Dan Burk’s 2020 Houston Law Review article titled “Thirty-Six Views of Copyright Authorship by Jackson Pollock.” Through a series of hypotheticals involving paint application, Burk develops a framework for analyzing authorship that focuses on causation, intent, and volition – concepts that emerged repeatedly during the Thaler oral arguments. Burk’s exploration of mechanical painting tools provides a direct parallel to Judge Millett’s Kodak camera hypothetical, particularly in how both examine whether intermediary devices break the chain of causation between human creativity and fixed expression. His analysis of accidental paint splatters and unintended artistic effects maps closely to the printer malfunction scenario, revealing how copyright doctrine treats unplanned outcomes. Most significantly, Burk demonstrates that authorship requires both a causal chain linking human mental formulation to physical fixation, and sufficient human decisional authority over the creative process – even when that process incorporates random or autonomous elements. This framework helps explain why Thaler’s attempt to claim authorship through mere ownership or programming of an AI system, without direct causation of the work’s expression, strains traditional copyright doctrine.

The panel’s most pointed questions focused on preservation of issues for appeal. When Abbott argued that the district court erred in finding waiver of the indirect control theory, Judge Millett pressed him to identify where in his appellate brief he had disputed the district court’s waiver finding. This exchange highlighted how procedural hurdles might prevent the court from reaching the broader doctrinal questions about AI authorship.  In supplemental briefing, Abbott pointed to pages in the briefs where the issues were, in his view, properly preserved.

Based on oral arguments, the D.C. Circuit appears poised to affirm the district court’s ruling against Thaler, either on procedural grounds or by finding that the Copyright Office properly refused registration.  During arguments, Judge Millett specifically noted Supreme Court precedent suggesting that Congress, not courts, should take the lead on adapting copyright law to technological advancement. And, based upon prior precedent, humans have always been seen as a central element of the creativity required for copyright protection.  A handful of precedential cases loom large over the dispute. In Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884), the Supreme Court established that copyrightable works must be “representatives of original intellectual conceptions of the author,” distinguishing between mere “mechanical reproduction” and expression reflecting human creative choices.  While that case allowed for copyrighting of photographs, it focused on the creative elements added by the human artist-photographer. The Ninth Circuit built on this foundation in two key cases: Urantia Found. v. Maaherra, 114 F.3d 955 (9th Cir. 1997) explained that “some element of human creativity must have occurred” for copyright protection, while Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018) found that terms like “children” and “widow” in the Copyright Act “imply humanity and necessarily exclude” non-human authors. Most explicitly, in Kelley v. Chicago Park Dist., 635 F.3d 290, 304 (7th Cir. 2011), the Seventh Circuit held that “authorship is an entirely human endeavor” and “authors of copyrightable works must be human.”

While Thaler’s case may fail, the underlying question of protection for AI-generated works has become increasingly pressing as millions of individuals and businesses integrate AI tools into their creative workflows. Whatever the outcome, the case leaves unresolved a critical question of human-AI collaborative works: what level of human contribution is necessary for copyright protection when a work results from human-AI collaboration? Consider an author who provides detailed prompts to an AI system, iteratively refines the output, and makes selective editorial choices – does this constitute sufficient creative input for copyright protection? And, could the human claim copyright to the entire work, or should the AI contributions be excluded as if prior existing work? These questions will require attention from future courts and potentially Congress, even if Thaler’s particular theories regarding purely AI-generated works fail to persuade.

* Tim Knight is a graduate of Grinnell College and a third-year student at the University of Missouri School of Law. Knight’s article on copyright’s discovery accrual rule was recently published in the Missouri Law Review.

Leave a Reply