The products of creative work are not merely societal decorations, they are of great (economic) importance. Music, for example, does not simply happen – it has to be made or created. And it is common knowledge that neither makers nor creators are prepared to work for free.
Copyright law is one of the few available instruments for monetising the creative process. It protects literary and artistic works – including music. The Copyright Act does not only regulate how works are protected, it also regulates accessibility to works and the exceptions to protection. Prohibiting access to works entirely would hinder both science and the creative process by making it difficult to engage with what is passed down. As a result, copyright issues are always caught in the tension between the need for protection and the need for freedom.
Artificial intelligence (AI) is the latest challenge in this tug-of-war. From the copyright law perspective, there are two especially significant aspects when considering generative AI systems. Firstly, input: In order to train AI, works protected by copyright are imported into a database. Secondly, output: AI produces new immaterial goods.
Input – feeding the AI database
Artificial intelligence can certainly produce something new, but it cannot do so from nothing. It needs “raw materials” which it sources from its database. The database contains works, including works protected by copyright.
Under Swiss law, the author alone has the right to decide whether, when and how their works are used and, especially, whether, when and how they are copied (reproduced). Therefore, the providers of AI systems basically need permission, i.e. a licence, from the rightholders to copy their works into the databases for the purpose of training the AI algorithms.
Using a work to feed the AI database is lawful where it is covered by a statutory limitation clause. An example of which is the exception covering the use of works for purposes of scientific research (Article 24d CopA): there is no reason in principle to disallow the use of AI for scientific research. But normal AI models are not designed for this purpose. Therefore, the scientific exception, as we call it, does not apply in most cases. Accordingly, permission has to be obtained from the rightholders for the corresponding reproductions.
How to proceed with the country-specific legal differences is not clear. Under European law, the limitation clause for text and data mining is more permissive than the Swiss scientific limitation clause, and also applies to AI database feeding. Rightholders do however have an opting-out option. In other words, an author may nonetheless decide whether, when and how the AI provider may use their works by opting out.
The European model seems questionable, on the one hand, given the lack of transparency (the AI black box) which makes it difficult to check whether the works being used are covered by an opt-out. On the other hand, however, the model can bring the players in this growing industry to the negotiating table – to negotiate remuneration or a licence fee for the creatives.
Output – artificial or artistic creation?
It is not for copyright law to decide what is and what is not art. But it is its task to determine how and under which premises art is protected. And this question is especially interesting in connection with AI-generated creation, or output as it is known.
The Copyright Act defines protected works as literary and artistic intellectual creations with individual character. To qualify for protection, a work must be intellectual in the sense that it must be immaterial but also that it must be the product of an intellectual effort. At the same time, the intellectual effort must be a creative effort, in other words, the result must be new insofar as it must originate in the mind of the author.
To qualify as an intellectual creation, the premise is that the work must be an immaterial creation of the mind. Although AI does generate immaterial output, it is not a creation of the mind – which is the sole preserve of the human being. Nor does such output originate in the mind of a natural person (the author), it proceeds instead from an – admittedly complex – algorithm. For this reason, purely AI-generated output is not protected by copyright.
It is a different matter, however, when AI does not produce a finished end-product but merely serves as a tool and the creative process is controlled by a human being. This would be the case, for example, where AI simply supplies an idea which then serves as the starting point for the creation of a new work. One would then have to verify on a case-by-case basis whether the new work is a literary and artistic intellectual creation with individual character. If the premises for copyright protection are satisfied, the work qualifies for protection even if it is an AI-assisted creation.
Outlook
Individual rights management is already complex enough as it is, and the complexity is aggravated in connection with AI reproductions. SUISA has collectively managed copyrights on behalf of its members and principals for a hundred years. Conversely to individual rights management, collective administration offers the advantage that individual rights can be bundled together and exploited collectively.
SUISA will not break with this tradition when dealing with the use of works for AI. SUISA will do everything in its power to ensure that authors, publishers and lyricists of musical works are lawfully remunerated as soon as possible. However, before protected musical works can be duly licensed for use in AI systems, several points of legal uncertainty will have to be resolved, be it through court rulings, scientific research, the enactment of legislation, or by benchmarking laws and exchanging information internationally.
So far SUISA has actively engaged in this discourse and will continue to do so. By actively contributing its expertise to the legal and political debates, cultivating and strengthening its international relations within umbrella organisations and with foreign sister societies directly, participating in domestic stakeholder associations, and thereby constantly developing its AI strategy, SUISA is confident that it will also master this new challenge in the interests and on the behalf of its members and principals.
Media release, 11 March 2024: “Artificial Intelligence – SUISA is committed to securing fair remuneration for its membersˮ (PDF, 92 KB)
