Flooded with half a million registrations every year, the U.S. Copyright Office sits front row to witnessing the upcoming trends in copyright law. The trend making its way through policy updates, registration contests, and ensuing litigation? The use of artificial intelligence (AI) to create expressive material. Last year, the Office launched an initiative focusing on AI's impact on copyright law and issued guidance on the registrability/copyrightability of works created using AI. At the root of the issue are the following questions: 1) whether copyright protection can/should/does subsist in works generated using AI, 2) how does AI impact copyright registration/protection, and 3) at what point does an AI-generated work cease qualifying for copyright protection. Let's discuss.
A Brief Primer on Copyright Law.
Copyright protection is enshrined in our Constitution. See U.S. Const., art. 1, s. 8, cl. 8 (giving Congress the power "[t]o promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries"). Under the Copyright Act of 1976, copyright protection subsists in "original works of authorship fixed in any tangible medium of expression. . ." 17 U.S.C. § 102(a). Copyright protection does not extend to "any idea, procedure, process, system, method of operation, concept, principle, or discovery, regardless of the form in which it is described, explained, illustrated, or embodied in such work." Id. at § 102(b). For the purposes of this article, we are focused on the idea of "authorship." In the context of AI, the question becomes do "authors" have to be human? Over the past couple of years, the U.S. Copyright Office has addressed this issue with increasing frequency.
The Office Issues Guidance.
In its March 2023 Registration Guidance, see 37 CFR 202, the Office took (and continues to take) the position that only humans qualify as "authors." Relying on a (timely!) 1884 U.S. Supreme Court decision regarding photographs, Burrow-Giles Lithographic Co. v. Sarony, the Office stated that SCOTUS "repeatedly referred to such 'authors’ as human, describing authors as a class of 'persons' and a copyright as 'the exclusive right of a man to the production of his own genius or intellect.'" The Office noted that federal appellate courts have reached similar conclusions, pointing to two cases from the Ninth Circuit from 1997 and 2018--slightly more current than 1884 but still predating the current AI craze.
In the context of AI, at least one federal district court addressing the issue has held that "United States copyright law protects only works of human creation." See Thaler v. Perlmutter, No. 22-cv-1564, 2023 WL 5333236, at *3 (D.C.C. Aug. 18, 2023).
According to the Office's guidance, the main consideration is "whether the AI contributions are the result of 'mechanical reproduction' or instead of an author’s 'own original mental conception, to which [the author] gave visible form.'" Stated differently, the Office asks whether the work is basically one of human authorship with the AI being an "assisting instrument" or whether the traditional elements of authorship were actually conceived and executed by a machine. In determining the copyrightability of any given work, the Office will examine how the specific AI tool operates and how it was used to create the final work. Not all AI tools are created equal for the purposes of copyright law, and so, the Office will review the applications on a case-by-case basis.
Under the Office's guidance, a work's "traditional elements of authorship" must be produced by a human. If the user gives direction to the AI tools, but the technology ultimately exercises control over turning the instructions into the output, then the output is not the product of human authorship. But, if there is sufficient human authorship even if some of the work is generated by AI, then the Office left room for the possibility of registration. For example, the Office said that a human could arrange AI-generated content in a sufficiently creative way to qualify as an "original work of authorship."
The Office gave the following directives:
An applicant must disclose that AI was used in the process and must describe the human author's contributions to the work in the "Author Created" field of a copyright application.
Individuals may claim copyright protection for their own contributions to a work produced using AI technology.
AI-generated content that is more than de minimis must be explicitly excluded from the application in the "Limitation of the Claim" section. Applicants should describe the AI-generated content.
Applicants who previously submitted applications that are pending and which contain AI-generated content should go back and re-assess whether they need to update their applications to properly disclose AI content.
5. Applicants who had registrations issued that contain AI-generated work should file a supplementary registration making new disclosures.
6. Applicants who fail to file a supplementary registration could risk the status of their registration (i.e., the Office could cancel the registration, or a court can disregard the registration in future infringement litigation).
The Office is continuing to monitor developments.
The SURYAST Saga: Application of the Office's Guidance.
At the end of last year, the U.S. Copyright Office affirmed its decision to not register a work that was created using AI. Here's what happened and how the U.S. Copyright Office applied its guidance to the work at issue.
In December 2021, Mr. Sahni applied for copyright protection for a 2-D photograph-artwork titled "SURYAST" (below). In his application, he identified himself and “RAGHAV Artificial Intelligence Painting App” (RAGHAV) as the authors of the work. Because Mr. Sahni identified AI as an author, the Copyright Office requested additional information regarding his creation process. Mr. Sahni provided his process: he chose a photograph that he authored, ran it through the RAGHAV app, input Van Gogh's "The Starry Night" as his style input, and chose a variable that would determine the amount of style input to apply to the photograph.
The Copyright Office refused the register the work, explaining that the work "lack[ed] the human authorship necessary to support a copyright claim." Mr. Sahni asked the Office to reconsider its refusal to register the work, arguing that "the human authorship requirement does not and cannot mean a work must be created entirely by a human author." The Office affirmed its decision not to register the work on the basis that the work was a derivative work and "does not contain enough original human authorship to support a registration.”
Mr. Sahni again asked the Office to reconsider. Mr. Sahni made three arguments in favor of registration: 1) RAGHAV served merely as an assistive software tool that facilitated the creative decisions by Mr. Sahni in selecting his original photograph, The Starry Night, and the variable value for the style transfer. 2) Mr. Sahni claimed that there were human-authored elements of the work. 3) The work was not a derivative work, and the original photograph was just an early stage of what the work would ultimately become--"akin to a painter making a sketch before completing a painting."
On December 11, 2023, the Office issued a letter again affirming its denial of registration for SURYAST. First, the Office reviewed how AI was used to generate SURYAST, and specifically how RAGHAV worked. RAGHAV was trained on a dataset of 14 million images and a dataset of "content and style images" so that it could learn how to transfer styles from the "content and style images" to the 14 million images. RAGHAV takes two image inputs, a style image and a base image, and a numerical value indicating the amount of style transfer. RAGHAV then generates a new image based on the three inputs. RAGHAV does not simply layer the style image on the base image like a filter. Here, Mr. Sahni selected the three inputs: his photograph as the base image (below), The Starry Night as the style image (below), and a variable amount for the style transfer. RAGHAV then generated the work (below).
Left: Mr. Sahni’s base image. Right: The Starry Night.
SURYAST
The Office found that the work was not the product of human authorship. The Office found that "the expressive elements of pictorial authorship were not provided by Mr. Sahni." Instead, after Mr. Sahni selected the three inputs, "the RAGHAV app, not Mr. Sahni, was responsible for determining how to interpolate the base and style images in accordance with the style transfer value." The Office rejected Mr. Sahni's attempt to describe RAGHAV as "an assistive tool," noting that the record previously described RAGHAV as generating a new image based on features learned from other images. Here, RAGHAV's predictive functionality is what determined the stylization of the work. Even though Mr. Sahni selected the variable for the amount of style transfer, the Office said that choice is the kind of de minimis authorship that is insufficient to warrant copyright protection. The Office ended by noting that Mr. Sahni could apply to register his photograph if it meets all the statutory requirements but could not register the AI-modified version. Ultimately, "[b]ecause Mr. Sahni exerted insufficient creative control over RAGHAV's creation of the Work, he cannot register it."
Conclusion.
Going forward, we can expect more decisions from the Office and more litigation in the courts over the extent of protection for works generated through the use of AI. As AI evolves, copyright policy and law will follow suit. For now, we know that these issues are front row and top of mind for both creators and the Copyright Office.
Comments