Exploring The Copyright Infringement Implications of AI-Generated Music

By Piyush Senapati

Piyush is a student at National Law University, Jodhpur. This piece is one of the winning entries from the Copyright Law Blog Writing Competition, 2023.

Introduction

Artificial Intelligence (AI) generated works are creating shockwaves all over the art and entertainment world due to their copyright implications- and the music industry is no stranger to this, as is evident by the concerns raised in this regard by musicians, both abroad and in India. Through the lens of Indian jurisprudence, this article seeks to analyse in what circumstances can AI generated music infringe the copyright of musicians, and whether such music can qualify for the fair dealing defence against copyright infringement.

How AI creates Music

The prospect of computers generating music is nothing new, with Lady Ada Lovelace herself predicting that they “might compose elaborate and scientific pieces of music of any degree of complexity or extent.” Understanding the exact mechanism by which AI creates music is necessary as the process itself raises copyright concerns- AI generated music is based on nothing but using already copyright music as ‘training material’ for the system. AI applications use databases of pre-existing songs, and use inferences from them to generate new content. Once a user feeds a musician’s existing music into the AI system, the machine learns to detect their stylistic patterns and then produces new music in that artist’s voice and style; but with lyrics, compositions, and other specifications of our choice. Thus, at the effort of only a few clicks, one can create an endless array of new music.

Testing AI-Generated Music on the Touchstone of the Average Listener’s Test

The case of R.G Anand v. Delux Films (hereinafter, R.G Anand) and subsequent judgements established that the test to determine copyright infringement is to see whether the viewer, having seen both the works gets an unmistakable impression that the subsequent work is a copy of the original. Therefore, the question of copyright infringement in cases of music generated through AI applications would ultimately boil down to whether an average listener would be under the impression that it is a copy of the music that’s used to generate it.

Before we test AI generated music based on the average listener’s test, it is important to note that music involves two copyrights– copyright over the musical works (rhythm, composition, and lyrics) and secondly copyright over the sound recording itself, also called the master recording. Thus, our analysis of this question will have to be undertaken for both.

Where copyright infringement of the musical works is concerned, it is unlikely that an average listener would recognise a meaningful percentage of the original works in the AI song. The reason for this boils down to the mechanism by which AI applications create music. Substantial portions of the musical works, do not manifest expressly in the final output as AI creates music by using the sound recording and mining the performative nuances therein, rather than using the musical works themselves. However, AI generated songs will nonetheless have to be analysed on a case-to-case basis, as certain musical works portions could appear in the final output, leading to copyright infringement if recognisable to the average listener.

Coming to copyright violations of the sound recording, there are two broad ways by which AI uses sound recordings to create music. The first is through independent fixation, i.e., recording an independent version of the song. The second is by manipulation of the actual copyrighted sound recording through encoding and decoding, wherein a portion of the same manifests in the final output. The American case of Bridgeport Music, Inc. v. Dimension Films held that independent fixation would not amount to copyright infringement, no matter how similar the new version is to the original recording, as one cannot violate sound recording copyright by creating an independent version of the recording. Similarly, the Court in Gramophone Company of India v. Super Cassette Industries (hereinafter, Gramophone Company) held that creation of a different version of the sound recording does not amount to copyright infringement. However, in the second instance where the AI generated music involves a reproduction of the original copyrighted sound recording, there can be a case of infringement if a substantial portion of the same manifests in the output and is recognisable to the audience.

Therefore, whether or not an AI generated song would violate copyright would depend on a number of factors ranging from the type of musical copyright involved, the method used by the AI to generate the song and recognisability of the original music manifesting in it.

AI Generated Music and The Defence of Fair Dealing

 The ‘Fair dealing’ exception to copyright infringement covers bona fide use of copyrighted works without any prior permission or remuneration to the copyright owner, as opposed to blatant copying with ulterior motives. Some examples include using copyrighted works for research or review purposes. Supposing that an AI generated song fails the average listener’s test, it is pertinent to analyse whether it could fall under this exception. The Court in Civic Chandran v. Ammini Amma (hereinafter, Civic Chandran) laid down the factors for determining whether a case falls under the fair dealing exception as- the quantum of the matter taken from another work, the purpose for which it is taken, and the likelihood of competition between the two works. More importantly, the allegedly infringing work must be transformative, i.e., different from the original work in character, expression, and meaning to fall under this exception. Where AI generated music is concerned, depending on the facts and circumstances of the case, the factors mentioned in Civic Chandran could have either allowed or disallowed the same to qualify for the fair dealing exception. However, what is concerning is that the judiciary has allowed the transformative factor to override the other factors outlined in Civic Chandran– it has been held that if the work is transformative, it would be considered as fair dealing and the quantum of copying and potential competition between the works would be immaterial. Given that AI generated music can fall under the abovementioned definition of transformative, this heightened emphasis on the transformative factor could provide leeway for AI based musicians to evade infringement claims even if they heavily copy and compete with original music they trained their systems with.

Conclusion

While there is ample scope for AI music producers to evade copyright infringement, artists viewing AI generated music with suspicion and hostility is natural, particularly in a country like India having a poor track record of musical copyright protection. While one side of the camp would hail AI for opening the floodgates for musical creativity, the other side would understandably fear its impact on musicians’ rights and livelihoods. To arrive at a balance between these two competing aspects, numerous solutions have been suggested, such as new licensing and royalty splitting arrangements. While the feasibility of these solutions is to be tested, it can be expected that the Indian judiciary will soon be grappled with infringement claims against AI generated music. In adjudicating the same, Courts must similarly endeavour to arrive at a balance between fostering creativity and protecting musicians.

---

Samarth Goyal, ‘Can AI Replace Musicians on Stage?’ Hindustan Times (1 April 2023) https://www.hindustantimes.com/entertainment/music/can-ai-replace-musicians-on-stage-101680330849579.html accessed 17 June 2023.

Emily Howard, ‘OK Computer: How Ada Lovelace is Being Bought to Musical Life’ The Gaurdian (2 November 2019) https://www.theguardian.com/music/2019/nov/02/ada-lovelace-emily-howard composer#:~:text=Lovelace%20valued%20music%20and%20mathematics,they%20might%20even%20compose%20music accessed 25 June 2023.

Kaishik Pal, ‘How Can an AI Model Create Music?’ (Techopedia, 28 April 2023) How Can an AI Model Create Music? – Techopedia (accessed 20 June 2023).

Khasim Lockhart, ‘AI-Generated Music: The Biggest Impact on the Music Industry Since Napster?’ (Lexology, 24 April 2023) AI-Generated Music: The Biggest Impact on the Music Industry Since Napster? – Lexology (accessed 19 June 2023).

[1979] 1 SCR  218 (SC).

MRF Limited v. Metro Tyres Limited [2019] 262 DLT 734 (Del), XYZ Films v. UTV Motion Pictures [2016] 67 PTC 81 (Bom).

Kanika Raheja, ‘India: Copyright Protection in Musical Work’ (Mondaq, 2 August 2021) https://www.mondaq.com/india/trademark/1097778/copyright-protection-in-musical-work#:~:text=This%20property%20includes%20the%20exclusive,music%20copyright%3A%20master%20and%20composition (accessed 20 June 2023).

Eric Sunray, ‘Sounds of Science: Copyright Infringement in AI Music Generator Outputs’, (2021) 29(2) Cath. U. J. L. & Tech 185.

Ibid 208.

Ibid 208.

Ibid 209.

[2005] 410 F.3d 792 (6th Circ.).

[2010] 44 PTC 541 (Del).

Sunray, ‘Sounds of Science’ (n 7) 209.

Jai Vignesh K, ‘Doctrine of Fair Dealing in Indian Copyright Law’ (Surana & Surana, 2 September 2022) Doctrine of Fair Dealing in Indian Copyright Law – SURANA & SURANA (suranaandsurana.com) (accessed 21 June 2023).

[1996] 16 PTC 329 (Ker).

The Chancellor Masters and Scholars of the University of Oxford vs Narendra Publishing House &Ors [2008] 38 PTC 385 (Del).

Syndicate of the Press of the University of Cambridge v B. D. Bhandari [2009] 39 PTC 642 (Del).

Shrija Verma, ‘Indian Music Industry and the Copyright Controversy’ (The IP Press, 29 January 2021) Indian Music Industry and the Copyright Controversy – The IP Press (accessed 27 June 2023).

Eammon Ford, ‘Intelligence Test: Techno-Plagiarism and The Battle for Music In a New Age of AI’ (Synchtank, 23 February 2023) AI Music and Copyright – Where Do We Go From Here? (synchtank.com) (accessed 25 June 2023).