By Keyerra Harfield. Keyerra is a second-year law student at the Sandra Day O’Connor College of Law, where she is focusing on the intersection of intellectual property and contract law. Currently, Keyerra serves as a Co-Senior Fellow of Symposia at the McCarthy Institute, and as Treasurer for the Intellectual Property Students Association. Keyerra also serves as an Associate Editor on Jurimetrics. Prior to law school, Keyerra worked at Goldman Sachs where she collaborated with in-house attorneys and corporate executives on legal and compliance initiatives.
——————————————————————————————————————–
In the rapidly evolving landscape of generative artificial intelligence (GenAI), our approach to creating music will need to develop alongside this new technology. The evolution of the music creation process is not a recent phenomenon; it has evolved in tandem with technology throughout history.2 During the nomadic era, tribes created lightweight and portable instruments, including flutes made from bones and drums constructed from animal skins.3 The portable nature of these instruments allowed tribes to preserve and sustain shared traditions and songsacross vast distances.4 As societies became more sedentary, instruments grew more extensive and complex.5 This sedentism led to the development of instruments like the piano, violin, and harp during the Renaissance and Baroque periods.6 As societies advanced technologically, music became more organized, thanks to the development of musical notation, which made it possible to record and share compositions widely.7
The invention of recorded music in the late 19th century marked a monumental shift in what?, allowing sounds to be preserved and disseminated like never before.8 Thomas Edison’s phonograph then revolutionized the accessibility of music.9 The 20th century experienced swift technological advancements ranging from vinyl records and magnetic tapes to CDs and digital streaming.10 This evolution transformed how we create, share, and enjoy music.11 The digital age, in particular, democratized music production and distribution, enabling independent artists to reach global audiences without the backing of major record labels.12
U.S. copyright law also evolved with the technological advancements in music. At its inception, the 1790 Copyright Act did not explicitly reference music.13 However, it allowed musical compositions printed as books to be protected.14 This foundation paved the way for the 1831 Act, which formally recognized musical works as copyrightable.15 An amendment in 1897 expanded protection to public performances, reflecting music’s growing cultural and economic significance.16 Finally, the 1971 Sound Recording Act and the 2018 Music Modernization Act extended copyright protection to sound recordings, aligning copyright protections with the global success of recorded music.17
Today, we stand on the brink of another musical evolution: the integration of GenAI into music creation.18 GenAI technologies are not only changing how music is produced but also challenging traditional notions of creativity and authorship.19 GenAI algorithms can analyze vast datasets of songs, learn patterns, and use them to compose a new song in a matter of seconds.20 This development opens new possibilities for innovation but also raises important questions about the protection of human creativity, especially under copyright law. As we enter a new era of music creation, understanding how GenAI has evolved is essential. This paper analyzes the evolution of Generative Advasarial Networks (GANs) and discusses their impact on copyright protections critical to the economic landscape and the creation of music.
I. Overview of GenAI and GANs
At a high level, GenAI models leverage learned patterns to create new content across various domains, including text, images, music, and code.21 These models analyze vast amounts of data to understand underlying structures and then generate new instances that conform to these learned patterns.22 GANs, in particular, have gained prominence due to their unique structure and ability to produce high-quality, realistic outputs in various domains.23 The dual-network architecture of GANs enables this advancement by fostering a competitive yet collaborative environment that enhances the creative output.25 Today, GANs are transforming music creation by introducing groundbreaking techniques that emulate human compositions.24
GANs rely on an adversarial interplay between two neural networks: a generator that creates synthetic data and a discriminator that aims to distinguish real from generated samples.26 During training, the discriminator first learns to identify fake data by comparing it with accurate data while the generator remains fixed.27 The process then reverses, with the generator adapting its parameters based on the discriminator’s feedback to fool it better while the discriminator remains constant.28
Simply put, GAN models create music by recognizing and imitating the patterns that contribute to a song’s distinctive sound. They pit a Generator and a Discriminator against one another.29 Rather than relying on creative judgment or emotional reactions, these systems break down music into its recognizable components.30 A Generator system examines a song’s elements to simulate the interplay of its melody, rhythm, harmony, and timbre.31 These models use millions of musical samples to understand how notes and chords relate probabilistically.32 They explore how different instruments harmonize and the way stylistic details define a genre’s unique feel.33 This method utilizes neural network models to imitate specific elements of human cognition while avoiding human-like interpretations of music.34
In practice, training an AI involves providing it with thousands of recordings and musical transcriptions, which may include copyrighted material.35 The quality and quantity of these “training datasets” influence the AI algorithm’s accuracy.36 The model fine-tunes its internal connections, referred to as “weights,” until it accurately predicts the next note.37 Because neural networks learn through these weight adjustments rather than explicit programming, it is often unclear exactly how the model reaches its conclusions.38 Through a rigorous training process, GANs learn to generate music that adheres to complex musical structures, producing compositions that are aesthetically pleasing and technically sound.39 However, these advancements stem directly from copying and training on copyrighted materials, creating legal and policy complications.40
II. Gen AI in Music and Copyright Concerns
Moving beyond the underlying technological principles, the use of GANs in music creation exposes critical gaps in copyright law. Copyright law’s main objective is to encourage the creation and sharing of creative works for the public benefit.41 The law was designed with human creators rather than automated systems in mind.42 Although U.S. copyright law grants copyright holders exclusive rights to their creative works,43 these provisions do not adequately address the complexities of using vast repositories of copyrighted music to train GenAI systems.
With millions of distinct works potentially included in a single training dataset, licensing becomes impracticable. This leaves GenAI music platforms vulnerable to extensive litigation.
Conversely, enabling an overly permissive fair-use model could undermine the creative ecosystem and the economic right doctrine of copyright law.44 As a result of this uncertain legal landscape, industry players like YouTube and Spotify have responded by restricting AI-generated tracks and removing GenAI content upon detection. 45 Universal Music Group likewise reached an agreement with TikTok to bolster protections against AI-generated music, reflecting a broader shift toward new content guidelines in response to this rapidly evolving technology. In light of these developments, many scholars, policymakers, and industry stakeholders now advocate for updated legal frameworks that clarify authorship standards, mandate transparent documentation of training materials, and encourage negotiated licensing between AI developers and rightsholders.46 Organizations such as the American Society of Composers, Authors, and Publishers (“ASCAP”) have called for collective licensing mechanisms that preserve innovation without sacrificing fair compensation for creators.47 By promoting legislative solutions over lawsuits, Congress can mitigate both the risk of stifling AI progress and the potential concentration of power in large-scale AI platforms.48 Such targeted reforms would uphold copyright’s overarching goal while reflecting the realities of generative AI’s role in music production.
III. Conclusion
As we reflect on the journey through technological advancements and legal considerations, it is clear that GANs are poised to revolutionize the music industry.49 They offer opportunities to expand creativity, collaboration, and innovation beyond the advancements of the past century. While GANs enable the generation of complex music, they challenge traditional notions of composition and authorship. To fully harness the potential of AI in music, it is essential to address the accompanying copyright law challenges.
By fostering dialogue and collaboration among all stakeholders, we can create a dynamic environment that embraces AI’s possibilities while safeguarding the rights and contributions of human creators. This balance will ensure that the evolution of music continues to enrich our culture in meaningful and equitable ways. By embracing the potential of GANs for music generation and proactively addressing the associated copyright concerns, we can usher in a new era of musical creativity.
1 AI was used for the research, scoping, and outlining of this paper.
2 See generally Big Think, 40,000 years of music explained in 8 minutes | Michael Spitzer, YOUTUBE (Jul. 29, 2022), https://www.youtube.com/watch?v=Am18ZxKgi_g.
4 See id.; See also Bruno Nettl, The Study of Ethnomusicology: Thirty-one Issues and Concepts 12 (New ed. 2005).
5 See generally Curt Sachs, The History of Musical Instruments 276 (W.W. Norton & Co. Inc.,1940).
8 Katz, Mark, Capturing Sound: How Technology Has Changed Music 23 (Univ. of Cal. Press, 2010).
9 Clive Thompson, How the Phonograph Changed Music Forever, SMITHSONIAN MAGAZINE (Jan. 2016), https://www.smithsonianmag.com/arts-culture/phonograph-changed-music-forever-180957677/.
10 Michael Chanan, Repeated Takes: A Short History of Recording and Its Effects on Music 56 (Verso, 1995)
11 Id.
13 Larry Wayte, Pay for Play: How the Music Industry Works, Where the Money Goes, and Why, UNIV. OF OREGON (May 15, 2023), https://opentext.uoregon.edu/payforplay/chapter/chapter-23-copyright-theory-and-history/.
15 Punit Motiwala, World Music Day 2024: The History of Music and Copyright, COPYRIGHT ALLIANCE (June 20, 2024), https://copyrightalliance.org/history-music-copyright/.
18 Bob L. Sturm et al., Music transcription modelling and composition using deep learning, in Proc. of the Computer Simulation of Musical Creativity Conference 1 (2016).
19 See generally Aaron Vick, Harmony Or Discord? The Impact Of Generative AI On Music Copyright, FORBES (Aug. 1, 2024, 8:45 am), https://www.forbes.com/councils/forbestechcouncil/2024/08/01/harmony-or-discord-the-impact-of-generative-ai-on-music-copyright/.
20 Alberto Martinez Jr, How Artificial Intelligence is Revolutionizing Music Composition, FLOURISH$PROSPER MUSIC GROUP (Jan. 3, 2024), https://flourishprosper.net/music-resources/how-artificial-intelligence-is-revolutionizing-music-composition/.
21 Patent Landscape Report – Generative Artificial Intelligence (GenAI), WIPO https://www.wipo.int/web-publications/patent-landscape-report-generative-artificial-intelligence-genai/en/1-generative-ai-the-main-concepts.html.
22 See, e.g., Maria Rosales Gerpe, Why You Should Use Gen AI in Content Creation, DOCEBO, https://www.docebo.com/learning-network/blog/generative-ai-in-content-creation/ (last visited Feb. 1, 2025).
23 Youssef Skandarani et al., GANs for Medical Image Synthesis: An Empirical Study, PMC (Mar. 16, 2023), https://pmc.ncbi.nlm.nih.gov/articles/PMC10055771/.
24 Adam Roberts et al., A Hierarchical Latent Vector Model for Learning Long-Term Structure in Music, in Proceeding of the International Conference on Machine Learning 4361, 4362 (2018).
25 Abhishek Deupa et al., GAN Based Music Generation, SSRN (Aug. 30, 2024), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4941197.
27 Id.
28 Id.
31 Markus Schedl et al., Music Information Retrieval: Recent Developments and Applications, 8 FOUND. & TRENDS IN INFO. RETRIEVAL 170 (2014).
32 Honglak Lee et al., Unsupervised Feature Learning for Audio Classification Using Convolutional Deep Belief Networks, ADVANCES IN NEURAL INFO. PROCESSING SYS. 22 1 (2009) https://proceedings.neurips.cc/paper/2009/hash/a113c1ecd3cace2237256f4c712f61b5- Abstract.html.
34 See What is a neural network?, IBM (Aug. 17, 2020), https://www.ibm.com/cloud/learn/neural-networks; See also Aravind Pai, 6 Types of Neural Networks in Deep Learning, ANALYTICS VIDHYA (Feb. 2020), https://www.analyticsvidhya.com/blog/2020/02/cnn-vs-rnn-vs-mlp-analyzing-3-types-of-neural-networks-in-deep-learning/.
36 See Amal Joby, What is Training Data? How It’s Used in Machine Learning, LEARN G2 (July 30, 2021), https://tinyurl.com/mss9hf25.
38 Lou Blouin, AI’s mysterious ‘black box’ problem, explained, UNIV. OF MICHIGAN-DEARBORN NEWS (Mar. 6, 2023), https://umdearborn.edu/news/ais-mysterious-black-box-problem-explained.
39 Deupa et al., supra note 25.
40 See, e.g., Lauren Forristal, AI music startup Suno claims training model on copyrighted music is ‘fair use’, TECHCRUNCH (Aug. 1, 2024), https://techcrunch.com/2024/08/01/ai-music-startup-suno-response-riaa-lawsuit/.
41 Kevin J. Hickey, Cong. Rsch. Serv., IF12339, Copyright Law: An Introduction and Issues for Congress (2023), https://crsreports.congress.gov/product/pdf/IF/IF12339.
43 17 U.S.C. § 106 (1990) (granting copyright holders the right to reproduce, distribute, publicly perform, and create derivative works).
44 Case Western Reserve University, Economic Right, Moral Right, and Databases, CASE (last updated Oct. 20, 2006), https://case.edu/affil/sce/authorship-spring2004/economic.html.
45 Erin Ollila, AI content rules: YouTube, Spotify & Audible 2025, DESCRIPT (Mar. 5, 2024), https://www.descript.com/blog/article/ai-content-on-youtube-spotify-audible.
46 See, e.g., Generative AI Copyright Disclosure Act, H.R. 7913, 118th Cong. (2024).
47 See generally, The American Society of Composers, Authors and Publishers, Comment Letter on U.S. Copyright Office Request for Comments on AI and Copyright Issues 7-8 (Oct. 30, 2023) [hereinafter ASCAP Comment].
49 Bob L. Sturm, et al., Artificial Intelligence and Music: Open Questions of Copyright Law and AI Creativity, 62 Communications of the ACM 68, 69 (2019).