Book Review: Musical Works, Copyright and Generative AI
The recently published doctoral dissertation Musical Works, Copyright and Generative AI-Legal Perspectives on Originality and Authorship (i-xiv, 1-275, 2025 Kluwer Law International BV), investigates and discusses copyright eligibility of AI-generated musical compositions. The author, Anton Ylikallio from the Vaasa Unit of the Helsinki University Faculty of Law, asks whether such works can qualify for protection by copyright law and if so, who may claim authorship? He approaches the research topic de lege lata and expressly emphasises that he does not advocate for granting copyright protection to AI-generated compositions.
Structure and methods
The thesis is divided into eight chapters. The introductory one is presenting the research premise and methodology (pp. 3-42) and the second is describing technical details generally and the development of generative AI within the musical field (pp. 43-70). The third chapter is running through the criteria for copyright eligibility particularly focusing on case law within EU (pp. 71-92). Chapters four to seven discuss authorial intent (pp. 93-106), a three-stage ontology of a work (pp.107-128), authorship of manual (pp.129-176) and AI-assisted template compositions (pp. 177-220). The conclusions are presented in chapter eight (pp.221-233).
Although the dissertation applies a traditional doctrinal legal approach, it is not a traditional legal dissertation. It namely combines musicological insights and aspects with law, which makes it both interesting, but also challenging for those readers who lack a musical background. This concerns particularly chapters five, six and seven regarding the three-stage ontology of a work, authorship of manual template compositions and AI-assisted template compositions. Here, the author is using his musical expertise to deconstruct a composition whereby he is supplementing the legal analysis.
The background for the multidisciplinary approach has a natural explanation. The author has also a background as a musician playing both concerts and teaching piano. He informs the reader that before autumn 2023, he shared his time between law and music, but at that time he switched to full-time legal research with the aim to finish his doctoral dissertation. He is now working as a university lecturer at the Vaasa Unit.
Substantial analysis
As a starting point for the dissertation, the author imagines an AI model designed to create musical compositions. The model is created by a group of developers, who make several decisions (inter alia) on training data and model architecture. He names this ‘Model A’. Model A allows a user to select the genre, style, mood, key signature and duration of the composition. The result is presented as an audio recording. The recording is called ‘Composition A’. Ylikallio now formulates two questions based on the above:
- Can Composition A and similar AI-generated compositions constitute musical works under copyright law? and
- If they can, which party or parties can claim exclusive right to them and on what grounds?
He then poses a few additional questions specifying the first two and ends up asking, under which conditions one can consider a sequence of instructions describing musical attributes (i) to be sufficiently specific, and (ii) sufficiently precisely reflected in the resulting composition to warrant copyright protection?
It is a matter of comparing pre-generation creative choices, as Ylikallio points out on several occasions, with the post-generation end result (for instance p. 40 and chapter 7).
Having described his starting point, argued his questions even more precisely and presented his musical background and approach, Ylikallio embarks on the research “journey” by describing AI-models in general, prompt engineering and music generation products, such as Suno, Udio, Google´s MusicLM and the improved version MusicRL as well as Meta´s MusicGen. It is interesting to note that already at this stage he is ventilating a warning: The media hype surrounding ´all things AI´ might exaggerate the impact of AI-powered systems. Oftentimes, the reviewer finds that in the public mind the speed of the output of generative AI-systems seem to be more important than the result.
Originality in CJEU case law
The following step of the dissertation focuses on the criteria for determining copyright eligibility. As indicated here, the object is a traditional legal journey of shedding light on the EU´s harmonisation of copyright requisites, the InfoSoc Directive and its interpretation by the Court of Justice of the European Union (CJEU), and post-Infopaq (C-5/08) case law.
In this part, references can be found to the Case C-393/09 BSA, in which the CJEU ruled that originality requirements are not met by features of a graphic user interface because of these being differentiated only by their technical function. In the Case C-403/08 Football Association Premier League it was stated that a sports event cannot be regarded as an intellectual creations and classified as a work within the meaning of the InfoSoc Directive. Basically, the same fate faced the Case C-406/10 SAS, which concerned the copyright eligibility of a software programming language.
Ylikallio then continues by commenting upon the Case C-145/10 Painer ruling concerning the copyright protection of portrait photographs. Here, the CJEU emphasises the importance of the ability for the author to express her or his free and creative choices reflecting her or his personality for the work to be deemed original and thus protected by copyright.
Having discussed the Case C-310/17 Levola Hengelo judgement on the taste of a cream cheese, the C-683/17 Cofemel judgement on clothing designs, the Case C-833/18 Brompton case on a collapsable bicycle and a few othersmore recent cases, he concludes by repeating the criteria of free and creative choices reflecting the personality of the author. Furthermore, he notes that there seem to be significant obstacles for obtaining copyright protection based on creative choices related to AI system development because of these only indirectly influencing the resulting composition.
Authorship
In chapter four, the authorial intent is discussed in the light of the concept of ‘author’ in copyright, i.e., a natural person engaged in the act of authoring. The latter meaning that she or he is converting a mental conception into a physical perceptible medium. In the dissertation the authorial intent is analysed in an interesting way through the ‘monkey selfie case’, where the macaque Naruto photographed itself using the photographer David Slater´s camera.
It becomes apparent that more effort must be put on proving that creative decisions have been involved when assessing the copyright eligibility of an AI-generated work.
The Ontology of a Work
Chapter five is focusing on the ontology of a work. The relationship between mental conception and external realisation is discussed. In the second section of the chapter, it is investigated what it means to produce an instruction set containing musical information and the role of interpretation in producing the musical result.
As earlier indicated for chapters five, six and seven, enlightening the reader with a musical background is of great help.
Having gone through the various steps in his research “journey”, Ylikallio concludes that Composition A can qualify as a musical work assuming the creative choices of the developers A or the user A are sufficiently specific and sufficiently reflected in the musical attributes. However, according to Ylikallio, User A´s contribution meets neither criterion, while Developers A might meet the criteria if the development of Model A involves multiple stages of iterative refinements focusing on specific elements of the model´s output.
Or – as the author puts it in slightly different phrasing – while some AI-assisted compositions may qualify for copyright protection under existing legal standards, many will fail to meet the originality threshold (that is, the work reflecting free and creative choices reflecting the personality of the author).
Conclusions
The dissertation is clearly an interesting and stimulating read. It is well articulated and consistent, and it sheds useful light on the various parts of a musical composition by deconstructing the same. Ylikallio’s way of putting together the generative AI-technology with musical composing and legal copyright criterion is excellent. Although having a musical background could have made the thesis easier to read, it is worthwhile spending time with also for those of us who are not musicians. This is of course generally the case with multidisciplinary research.
The conclusion is certainly reasonable. As well known, courts in the United States have not so far considered an AI-generated musical compositions as protected works. It is therefore interesting to note that the Internet court in Beijing has found the prompts of an AI-generated work creative enough to render copyright protection to the result.
It appears that for the time being there is no need to create an economic incentive for developing generative AI-models producing music or other works, as the development is breathtaking enough. Another matter is that only time will tell if all the trillions of dollars spent on AI will give a monetary payback (taking into account, for instance, all the energy used). In the meantime, we seem to be fascinated, but also struggling with the accelerating stream of output – be it music, fine art, literature, etc., produced by the various AI-models.
Under all circumstances it is crucial to keep in mind not only the legal perspective, but also the human one.
Kirjoittajat
