Defining Authorship for the Copyright of AI-Generated Music
I. Introduction
On October 13, 2023, music publishers Universal Music (UMG), ABKCO, and Concord sued the artificial intelligence company Anthropic for its AI music-creating chatbot, “Claude.” [1] The lawsuit alleged that Anthropic violated artists’ rights by feeding Claude a database of copyrighted lyrics for over 500 mainstream, contemporary songs. [2] As the case moves forward this year and Anthropic continues to defend its development of Claude, this lawsuit represents the most recent instance in which the music industry has sought to redefine the extent to which copyright protects existing creative work from AI-generated production and to establish a legal framework that AI-generated work must respect.
While in the Anthropic case, the artificial intelligence company itself is the entity being sued, the fundamental question regarding how modern copyright regulations apply to AI music does not center on what motivates a lawsuit — clearly, it is the 500 songs “Claude” drew upon — but who should be sued. That is, who is the “author” of the composed work that takes ownership of the AI’s infringements on copyright? Should the developer of the AI program, as seen in the Anthropic case, always be held accountable for infringements on the copyrighted music produced by its AI system? Alternatively, should responsibility fall on the user of the AI system or the AI system itself?
The United States Copyright Office currently specifies that AI-created works do not fall under copyright. [3] Without an updated understanding of how AI software relates to pre-AI notions of authorship developed in the Constitution, we cannot hold any entity related to AI-generated music — whether this be the developer, the user, or the system itself — responsible for infringing on existing human artists. Thus, this paper hopes to clarify whether there is a material difference between the violation of copyright occurring from AI-generated music versus the violation of copyright from human-made music. Authorship should apply to AI systems, and these machines and their developers must be held accountable for creating and infringing on copyrighted works of music. To accommodate this expansion in the definition of authorship, current U.S. Copyright Office guidelines surrounding the creation of AI-generated music must be revised and expanded to consider both humans and non-humans.
II. The History of Copyright’s Relationship to AI-Generated Works
Currently, under the U.S. Copyright Office provision, AI-generated art cannot be protected under copyright because of the human authorship clause, which states that a human must be deemed the author of a creative work for copyright to apply. [4] More specifically, the human authorship clause originated from the third of three requirements outlined by the original 1909 Copyright Act that a work of art must meet to qualify for copyright protection. The first requirement for copyright protection states that the work must be “fixed in a tangible form of expression,” the second is that it must be an “original work,” and the third is that it must be “created by an author.” [5] Consensus agrees that AI-generated music meets the first two requirements. For a work of art to be “fixed,” it must be “sufficiently permanent or stable to permit it to be perceived, reproduced, or otherwise communicated for a period of more than transitory duration.” [6] A piece of music, as it stands completed, fulfills this requirement. Moreover, AI-generated music is also original, under the definition that “originality” refers to work that possesses at least a minimal amount of creativity. Here, AI-generated music is considered original, as only an indistinguishable copy of another’s work would disqualify the AI-generated result from meeting the originality requirement. While an AI may take unmodified elements like lyrics from other artists, nevertheless, it amalgamates them to create a distinct, original product, meaning it achieves originality. [7]
The last and the most contentious clause an AI-generated work of music must fulfill is the authorship clause. An author is defined as: “he to whom anything owes its origin; originator; maker; one who completes a work of science or literature.” [8] Neither the Constitution nor the Copyright Act defined authors as explicitly human, in part because the framers of the Constitution may never have expected the rise of artificial intelligence. [9] Significantly, it was only through intervention by the Copyright Office that the authorship clause was modified. In its 2021 Edition of the Compendium, the U.S. Copyright Office updated these original definitions by specifying that for a work of art the “work must be created by a human being,” and “[w]orks that do not satisfy this requirement are not copyrightable.” [10]
Recent litigation upholds the U.S. Copyright Office’s modified position. In Thaler v. Perlmutter (2023), “the court concluded that copyright protection is reserved for works of human creation.” [11] After Stephen Thaler attempted to apply for copyright protection on an AI-generated visual work created by an AI algorithm known as the “Creativity Machine,” the Court conducted an in-depth analysis of current definitions of authorship in legal texts such as the Constitution. To protect the basis of artistic creativity as stemming from human creations only, the Court upheld authorship as requiring human origins. [12] However, in an attempt to protect creativity itself, the Court allowed AI-generated programs to remain unregulated, thereby enabling them to copy and infringe upon the creative works of human artists. The verdict in Thaler has also been applied to the world of music. As of March 15, 2024, the Copyright Office issued its latest formal guidance after a TikTok user named “Ghostwriter977”’s AI software produced a hit song from superstar musicians Drake and The Weeknd’s music. In the Copyright Office’s statement, the Office continued to affirm that works created by “AI without human intervention or human involvement cannot be copyrighted.” [13]
However, the U.S. Copyright Office’s definition clashes with modern scenarios involving AI music programs. Indeed, the Court acknowledged in Thaler the “lingering uncertainties related to AI authorship and copyright.” [14] The problem with maintaining the current Copyright Office guidelines is that no one could have ever predicted that artificial intelligence systems could one day produce music — or any form of art — as skillfully and similarly as professional artists. For instance, the “Ghostwriter977” TikTok reached millions of eager listeners across social media platforms and opened the doors for other AI systems to create more art based on its songs. Examples such as this TikTok account represent the beginning of a new revolution in AI-generated art, and the human authorship clause enables a market of non-regulated, AI-generated music, which creates ethical and economic concerns for human artists who cannot protect their music from non-human, AI-induced infringements.
Additionally, the Court has not addressed gray zones such as when human authors prompt AI programs to generate music. For instance, consider the situation in which a user writes a piece using lyrical elements from the hit songwriter Ed Sheeran, while another user prompts an AI program to develop a song in the style of Ed Sheeran by drawing on an existing database of Sheeran’s lyrics. In the former scenario, the user is operating directly under the threat of copyright infringement because of the human authorship clause, but in the latter scenario, under current legal guidelines, artists like Ed Sheeran cannot sue the AI developer, user, or system for copyright. Professor Robert Denicola of the Nebraska College of Law summarized the complicated dynamics at hand. He wrote, “If a person uses a computer to assist in the manipulation of expression created by the user, the result is copyrightable. If a user’s interaction with a computer prompts it to generate its own expression, the result is excluded from copyright.” [15] Currently, the U.S. Copyright Office Statement of Policy published on March 16, 2023, expresses the same sentiment as Profesor Denicola: “If a human submits a prompt to an AI system asking it to create a work, but the AI system has full control over the expressive elements of its output, such work is simply not the product of human authorship and is not copyrightable.” [16]
Because the Copyright Office’s human authorship clause does not make room for non-human creators and does not regulate human-prompted, AI-generated musical compositions, the legal scope of copyright is only broad enough to regulate human beings and does not extend to AI-generated music. However, the acceleration of AI technology complicates this definition, especially those works prompted by human users. Moreover, the flood of AI-generated music on platforms such as Spotify, YouTube, and TikTok means that every day, the creativity of human artists is encroached on by artificial systems. An entity must be held accountable if it creates works of art that intervene in a marketplace regulated by copyright, so a responsible party inevitably must be chosen in these niche situations to be designated as the one held responsible for copyright infringements.
III. Definition of an Author, Definition of AI
By comparing the definition of authorship and the definition of an AI system, as specified in the Constitution and other past litigation, there are numerous entities pointed to as the author of AI-generated music. Proving these definitional similarities is contingent on whether the actions performed by the AI system correspond with the actions performed by a human author of copyrightable art. If AI fulfills the requirements of authorship in this scenario, then there is a strong argument that the human authorship clause is legally obfuscating and outdated. I argue that current copyright guidelines should be revised to reflect pre-AI copyright guidelines outlined in the Constitution and the 1976 Copyright Act, which do not require human authorship.
According to 15 U.S. Code § 9401, artificial intelligence is defined as “a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments.” Later, in the notes of that code, it further specifies that AI is “[a]n artificial system developed in computer software, physical hardware, or other context that solves tasks requiring human-like perception, cognition, planning, learning, communication, or physical action.” [17] When an AI system creates a work of music, it is fulfilling its mandate to “make…decisions” and “solv[e] tasks requiring human-like perception.” The act of generative music production requires notes and lyrics to be arranged in a specific order and manner, which requires decision-making and the ability to create an original work. AI is not a sentient being, but these qualifications about decision-making and “human-like perception” illustrate that when it comes to the production of music, the AI can act in a way identical to a human. [18] In the case of composing music and creating works of art in the age of AI, the scope of copyright should rest on whether the work is imbued with the human characteristics of creativity and decision-making, not whether the creator of the work has a pulse.
Critically, the non-human element of AI systems does not preclude such systems from being considered authors under constitutional law. The Constitution and the 1976 Copyright Act both mention the word “authors” but never define it to mean “human author.” [19] Instead, in the late-19th century case Burrow-Giles Lithographic Co. v. Sarony (1884), the Court defined “author” to mean “he to whom anything owes its origin; originator; maker.” [20] This case concerned whether photographs could be copyrighted, and the Court held that photographs were copyright-able because they could be “traced quite directly back to the governing consciousness and sensibility of the photographer, the person behind the lens who posed the subject just so and altered the lighting just so.” [21] The words “origin” and “traced” illustrate that the Court intended to define an author not according to whether it is human but according to its ability to be held responsible for a work’s creation. That is, one must be able to point to the AI system as the original entity responsible for creating the music. Legal experts such as Nina Brown dismiss the possibility of the AI system serving as an author — calling the computer a piece of “chattel” without the ability to act as an author — but arguments such as Brown’s, which do not reflect the current technological capacities of AI, only apply under the current legal framework in which the human authorship clause remains written over original copyright laws. [22] As long as an AI system can be pointed to as responsible for the creation of the music in question, then AI-generated music can theoretically fall under the definition of “origin” in pre-AI copyright protections outlined by the Constitution. It is only since the 2021 updates by the U.S. Copyright Office, that “a copyrightable work must be created by a human being.” [23]
Moreover, the Copyright Act also already grants non-human authorship in the context of hiring, in which employers usually exercise their rights and obligations as a corporate entity that holds legal personhood. Stated in the “work for hire” statement, the Act specifies: “[i]n the case of a work made for hire, the employer or other person for whom the work was prepared is considered the author.” [24] If non-human entities already legally exist in the world of industry and can be held accountable for acting on subjective decisions such as hiring other humans, then this framework surrounding non-human authorship can be applied to the world of art and music, in which AI entities are making creative decisions regarding the music of humans.
IV. Designating Authorship
Consider three parties that may be held responsible for copyright infringements by AI-generated music: the developer of the AI, the user of the AI, and the AI system itself. By analyzing the mechanisms by which each entity controls the creation of AI-generated music and how it compares to the definitions of author established by the Constitution and the 1976 Copyright Act, we can understand who may be considered an author and why the AI system itself is likely the most logical choice, conditional on a revision in U.S. Copyright Office guidelines to permit non-human authors.
A. The Developer of the AI
At first, the developer of the AI seems to be a natural choice to point to as the author — this is the situation in the Anthropic case, in which Anthropic is being held accountable for Claude’s music. The problem is that the developer controls the creation of the AI software but not the AI-generated music itself, which distances them from the final product. [25] Furthermore, developers of AI already have legal protections and copyright regulations on the AI software, so affording them protection over the end product can be considered a “double dipping” concern. [26]
The End User
The end user of the AI system can also be designated as the author. This seems logical because the end-user is the one controlling how AI software delivers artistic results. Much like a person fixes the artistic idea of capturing an image by pressing a button on the camera to take a picture — which was a relevant distinction in the famous Burrow-Giles case — the AI user chooses the words, lyrics, or melodic samples to prompt the AI system. [27] But, generative AI differs from the camera in the Burrow-Giles case: rather than a singular snapshot prepared fully by the user of a camera, generative AI creates original content that was not already apparent to the user. In other words, the problem with designating the user as the author is that the AI system itself performs the specific action of composition alone, without the user guiding it. The human is not involved with the actuation of the art, and the AI makes the final creative decisions after being initially prompted.
The AI System
The final, most likely possibility is that the AI system should be the legal entity held accountable for copyright. This is because the human on the developer side and the human on the user side are both less involved with the creation of the musical work than the AI system itself. Scholar Ryan Abbott draws similarities between AI systems in copyright and patent law. Regarding AI usage in patent law, he writes, “[c]reative computers invent because they are instructed to invent” — meaning, the function of the AI or the “creative computer” is to create. [28] Just as patents protect inventions, copyright protects works of artistic expression. Though one may argue that artistic expression requires human origins, the Constitutional definition of authorship — which is what currently counts for determining the parameters of copyright — mentions no specificities that would restrict an AI system from qualifying as the creating entity behind AI-generated music.
V. The Stakes of Copyright Regulation of AI-Generated Music
The Copyright Office and other legal experts argue that the human authorship clause remains the standard for the copyright of artistic work because creating art is inherently a human process. [29] These agencies seek to protect the integrity of human-created music by only including human authors in their framework for copyright. Ironically, these well-meaning guidelines have not adjusted to the current climate of music production, in which AI programs are taking over the roles of top artists such as Drake or the Weeknd and poaching the content of smaller, independent artists who do not have the visibility and firepower to fight against AI systems. [30] The lacuna of copyright when it comes to AI-generated music has led to an even greater deterioration of integrity within the art world and has greenlighted a higher propensity for non-human art to flood the market. This reality should deeply concern copyright proponents who encourage human expression and creativity.
Copyright was created to protect artists and ensure that their art retained marketable value, but the lack of regulation of AI-created music establishes a parallel “black” market where artists’ ideas are copied and repurposed by AI systems capable of creating millions upon millions of instances of marketable, unprotected music. The economic implications of this shift are extreme. The “Ghostwriter997” case is proof that AI systems can create highly successful songs on social media platforms such as TikTok. This form of non-human-driven success disincentivizes and crowds-out innovation by human artists. [31] Humans should no longer fear the risk of other humans copying their work as much as they should fear an AI system creating permutation upon permutation of unregulated music to be published online to people without the original author’s consent.
VI. Conclusion
Under the U.S. Copyright Office provision, AI-generated art cannot be protected under copyright because of the human authorship clause. But, by tracing the history of copyright law as it pertains to the human authorship clause and engaging in a definitional analysis of authorship, we find that pre-AI interpretations of authorship can apply to non-human entities. Thus, by eliminating the human authorship clause, the copyright laws surrounding AI-generated music become clearer and better regulated. In a world in which the works of generative AI are encroaching on the creative and economic rights of artists, understanding how AI systems can serve as legal entities creates a mechanism to hold them legally responsible for their infringements.
One issue remains to be investigated further. If the AI system is named the entity accountable for copyrighted music, it is not clear how damages from copyright infringements can be paid off, as the AI cannot engage in that transaction itself. This situation may call upon the developer of the AI or some independent regulatory agency to become involved as the arbiter of restitution and held responsible for altering the AI’s design as a consequence of its infringements. The nuance of the issues at hand in determining origin, intent, and how damages may be sought is proof that legal clarifications are needed to add regulation and protection in a new age of AI-generated art: removing the human authorship clause is the first step to establishing a new framework.
Bibliography
[1] Blake Brittain, “Anthropic fires back at music publishers' AI copyright lawsuit,” Reuters, January 17, 2024, https://www.reuters.com/legal/litigation/anthropic-fires-back-music-publishers-ai-copyright-lawsuit-2024-01-17/.
[2] Daniel Tencer, “AI trained its AI to rip off copyrighted lyrics, music publishers allege in escalating court battle,” Music Business Worldwide, February 15, 2024, https://www.musicbusinessworldwide.com/anthropic-trained-its-ai-to-rip-off-copyrighted-lyrics-music-publishers-allege-in-escalating-court-battle/.
[3] “Federal Register Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence,” accessed May 6, 2024, https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence.
[4] U.S. Copyright Office, Compendium of U.S. Copyright Office S § 313.2 (3d ed. 2017), Chapter 306.
[5] Mackenzie Caldwell, “What Is an ‘Author?’ - Copyright Authorship of AI Art Through a Philosophical Lens,” Houston Law Review 61, no. 2 (December 11, 2023), 419, https://houstonlawreview.org/article/92132-what-is-an-author-copyright-authorship-of-ai-art-through-a-philosophical-lens.
[6] 17 U.S.C § 101, Cornell Legal Information Institute, https://www.law.cornell.edu/uscode/text/17/101.
[7] Nina I. Brown, “Artificial Authors: A Case for Copyright in Computer-Generated Works,” The Columbia Science and Technology Law Review 20, (Fall 2018), https://doi.org/10.7916/stlr.v20i1.4766.
[8] Burrow-Giles Lithographic Co. v Sarony, 111 U.S. 53 (1884), 58.
[9] Caldwell, “What Is an ‘Author?’,” 422.
[10] Neal F. Burstyn, “Creative Sparks: Works of Nature, Selection, and the Human Author,” The Columbia Journal of Law & The Arts, 39(2), 281. https://doi.org/10.7916/jla.v39i2.2089.
[11] Atreya Mathur, “Case Review: Thaler v. Permutter (2023),” Center for Art Law, December 11, 2023, accessed May 6, 2024. https://itsartlaw.org/2023/12/11/case-summary-and-review-thaler-v-perlmutter/.
[12] Ibid.
[13] Rachel Reed, “AI created a song mimicking the work of Drake and The Weeknd. What does that mean for copyright law?” Harvard Law Today, May 2, 2023, https://hls.harvard.edu/today/ai-created-a-song-mimicking-the-work-of-drake-and-the-weeknd-what-does-that-mean-for-copyright-law/.
[14] Mathur, “Case Review.”
[15] Robert Denicola, “Ex Machina: Copyright Protection for Computer-Generated Works,” Rutgers University Law Review 69 (2016), 251, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3007842.
[16] Cassidy Kitterman, “No Human Authorship, No Copyright,” TALG Law, September 19, 2023, https://talglaw.com/no-human-authorship-no-copyright/.
[17] 15 U.S.C § 9401, “Artificial intelligence (AI),” Cornell Legal Information Institute, https://www.law.cornell.edu/wex/artificial_intelligence_(ai).
[18] Russ Pearlman, “Recognizing Artificial Intelligence (AI) as Authors and Inventors Under U.S. Intellectual Property Law,” 24 Rich. J. L. & Tech. no. 2, 2018.
[19] Ibid.
[20] U.S. Constitution, art. 1, sec. 8, cl. 8.
[21] Brown, “Artificial Authors,” 30.
[22] Ibid., 33.
[23] Burstyn, “Creative Sparks.”
[24] 17 U.S.C. § 201, “Ownership of copyright,” Cornell Legal Information Institute, https://www.law.cornell.edu/uscode/text/17/201.
[25] Caldwell, “What Is an ‘Author?’,” 433.
[26] Brown, “Artificial Authors,” 23.
[27] Caldwell, “What Is an ‘Author?’,” 435.
[28] Ibid., 433-434.
[29] Pearlman, “Recognizing Artificial Intelligence.”
[30] Brown, “Artificial Authors,” 5.
[31] Reed, “AI created a song.”