The music sector is regularly confronted with disruptive technological innovations that impact the industry, the public, and the artists. Certain historical changes perceived today as progress could, at the time, destabilise the performers’ jobs, weaken their income and profoundly modify their way of working. It was the case with radio, the LP, the minicassette, the CD, and computer music, more recently with downloading and streaming, and now, artificial intelligence.
The evolution of the normative framework can help balance these effects for the different parties concerned. For music performers, the Rome Convention and the WPPT have provided welcome solutions regarding broadcasting and communication to the public. Unfortunately, these instruments have failed to regulate download and streaming effectively, with Article 10 of the WPPT, as currently implemented, not allowing artists to benefit from a fair share of the revenue generated by the online exploitation of their recordings.
Recent advances in generative artificial intelligence and the techniques implemented for machine learning suggest an analogy with human learning mechanisms. However, this analogy quickly reaches its limits. Indeed, the volume of data ingested and the speed at which the machine collects and assimilates these data are incommensurate with what the human mind is capable of. Machine learning consists of appropriating all the creations of the human mind currently accessible and encoding them into algorithms to generate new contents based on the knowledge acquired. This new paradigm radically differs from the slow and gradual knowledge acquisition process at work in humans.
The AI services recently made available to the public translate into a competitive and fast-growing market with strategic implications and considerable profit prospects. However, this new ecosystem is not regulated by any adequate normative framework protecting the community of creators whose work and talent are exploited in proportions beyond comprehension.
AN INADEQUATE COPYRIGHT FRAMEWORK
The existing copyright and neighbouring rights normative frameworks were not designed to address the particular problems posed today by generative AI, whether for incoming or outgoing data. One should, therefore, not assume that the transfer to a producer of a performer’s exclusive rights covers the right to authorise or prohibit the use by AI of that performer’s recorded performances, irrespective of whether such use includes an act of reproduction.
The performers’ moral right introduced by the WPPT in 1996 does not help. It is limited to “the right to claim to be identified as the performer of his performances, except where omission is dictated by the manner of the use of the performance and to object to any distortion, mutilation or other modification of his performances that would be prejudicial to his reputation”. In the AI environment, performers need and deserve a more robust moral right, broad enough to allow them to individually oppose the use of their works, sounds, voices, images, likenesses or styles for either TDM Purposes or the generation of audio products by AI (or with its assistance), including deep fakes.
The copyrightability of content produced by AI or with its assistance is a new and complex question that gives rise to discordant decisions depending on the country. At this stage, deciding firmly between the copyrighting of AI- generated content (or with its assistance) and its classification as public domain remains challenging.
THE EU LEGISLATION PROVIDES NO SATISFACTORY RESPONSE
Art. 4 of Directive 2019/790, which provides an exception to the exclusive right of reproduction for “text and data mining” (TDM), allows rights holders to reserve their rights through “machine-readable means in the case of content made publicly available online”.
Firstly, it is necessary to clarify whether the TDM exception of Art. 4 mentioned above is compliant with the three- step test enshrined in the Berne Convention, the WCT, the WPPT, the Beijing treaty and the EU acquis, which allows limitations to an exclusive right only if the following cumulative criteria are met:
a. in certain special cases;
b. that do not conflict with the normal exploitation of the work; and
c. that do not unreasonably prejudice the legitimate interests of the author / right-holder.
As far as the AI-generated output is intended to enter the market and compete with human creations on unfair terms, compliance with steps b and c is highly questionable.
Another problem with this article is that five years after the adoption of Dir. 2019/790, we still lack standardised machine-readable means.
Finally, the number of rights holders in a recorded performance makes the “opt-out” mechanism a complicated machinery for which no consensual solution has emerged yet.
Art. 52 of the European AI Act requires a minimum level of transparency regarding the sources used for content generation via “transparency obligations for providers and users of certain AI systems and GPAI models”. We welcome this first step and believe that any content generated by AI or with its assistance should be documented with detailed information on the sources used and a guarantee that all creators’ rights have been respected. It is also essential that members of the public are informed of the nature of the content and know whether they are dealing with the work of a human mind or an AI product.
RECOMMENDED REMEDIES
On the input side
Preventing human displacement
A vibrant culture of human artistry is an inseparable component of the foundation of a civilised society. This precept demands a fair but firm set of controls that balances the demands of commercial enterprise with the imperative of every society to protect and preserve its cultural soul from displacement by the inexorable advances of technology.
It is unacceptable, therefore, that music performers can be the victims of large-scale exploitation of their works, sounds, voices, images, likenesses or styles without their free, prior and informed consent and against no financial compensation. Performers should have the right to authorise and effectively prohibit the scraping and analysis of their works, sounds, voices, images, likenesses or styles by an AI system, including after the transfer of their exclusive rights, and to receive financial compensation for such use.
It is also vital to ensure that performers enjoy the same level of protection against unauthorised use of their performances by AI, whether based on a literary or artistic work, an expression of folklore, or AI-generated material.
On the output side
Preventing market distortion
Once AI has ingested and analysed performers’ works, sounds, voices, images, likenesses or styles, it can use these data to output new contents at a scale representing a considerable market distortion and an objective threat to the careers and livelihoods of all present and future artists.
We need a sustainable legal and economic environment that efficiently prevents AI-generated sound and audiovisual materials from distorting the market with prices much lower than those of human creations protected by copyright and neighbouring rights.
Financial compensation based on the output
To the extent that AI-generated content draws its value from human creations exploited on a large scale, it is entirely relevant to consider mandatory compensation mechanisms benefiting the creative community and applying to all generative AI tools.
Innovative remuneration mechanisms based on the output should, therefore, be considered. Any AI-assisted generation of musical content should be subject to fair payments to performers as their work and talent constitute the knowledge base at the origin of such content. Such fair payments, however, must not operate to normalise or unduly encourage the supplanting by generative AI of the work of individual human beings. We need a payment system that will honestly compel a producer who is contemplating the use of generative AI to weigh the economic advantages of human-produced products and performances against the convenience of a generative AI-produced products.
This may require the creation of a sui generis intellectual property regime as stand-alone legislation for adequate compensation. The United Nations Declaration on the Rights of Indigenous Peoples (UNDRIP), which affirms the right of Indigenous Peoples to restitution or just, fair and equitable compensation for resources taken and used without their free, prior and informed consent, provides a wise principle that we recommend following.
Existing private copying compensation mechanisms may also represent a valuable model for collecting payments from the users and distributing them to the performers concerned.
We also strongly support the language in Principle 11 of the G7’s International Draft Guiding Principles for Organizations Developing Advanced AI Systems and urge the EC and the G7 to continue supporting explicit language on respecting material protected by intellectual property rights, including copyright-protected content, and ensuring transparency of data sets, as part of promoting safe, secure, and trustworthy Artificial Intelligence (AI) technology worldwide.
Version of 27 June 2024