Works generated through complex AI systems, such as machine learning and text-to-image generation models, have recently stirred up many discussions and even given rise to lawsuits (here and here). Voices emerged questioning whether current EU copyright laws should be amended in light of the many AI-generated works that have come about. One important question has been whether copyright law should be extended in order to protect such works. The academic debate has revolved mostly around copyright law rationales, the human-centred authorship requirement, as well as the notions of creativity and originality. In this upcoming paper, this author takes a different approach to this hot topic. The research positions copyright law within the EU’s constitutional limits to consider whether the EU legislative competences allow for the expansion of copyright protection to purely AI-generated works.
In EU copyright law, a central requirement for protection is human authorship and, specifically, the human’s clear stamp of free and creative choices in the final output (here and here, among many others). In many computational creativity projects in the fields of art, journalism and music, the heavy reliance on AI stretches the causation bond between the human author and the final creative output to breaking point. Consequently, it is not clear whether copyright protection would still subsist in many of these newly emerged works.
Proponents of extending copyright protection for AI-generated works suggest that absent copyright protection for such works, creativity would be stifled and various industries where purely AI-generated works are abundant will suffer underproduction. AI processes will be able to produce a large number of works extremely quickly. Faced with the choice between using an AI-generated work, which according to the status quo of EU copyright law today is likely to be free from copyright protection, and a human-authored work, for which a user needs to secure permission, some have suggested that users will prefer the former. Therefore, such AI-generated works are said to compete directly with human-authored works and thus might be capable of disturbing the market for low creativity works, which is where apparently many artists nowadays make a living.
In light of this, this author poses the following question: should the internal market goal justify opening EU copyright law to AI-generated works?
Following the principle of conferral, the Union can legislate only within the limits of the competences conferred upon it by the Member States. There is no specific legal basis tackling copyright, meaning that EU copyright law-making has not been based on copyright-related reasoning, but instead on the goal of establishing an internal market as per Article 114 TFEU. To that end, the EU would typically introduce secondary copyright legislation whenever the differences between national laws risk interfering with the free movement of goods and services. All thirteen copyright directives, as well as the three regulations in the field have Article 114 TFEU as a legal basis.
Importantly, copyright law is equally about culture. However, the Union’s cultural competences, which can be found in Article 167 TFEU, are solely coordinative. Thus, culture cannot be relied on to pass harmonising measures, which is what a potential expansion for purely AI-generated works would seek to do. From a practical perspective, in terms of EU copyright law-making, this renders the culture legal basis borderline useless. On the flipside, the internal market goal’s flexible mechanics have allowed the EU legislator to present (and pass) numerous copyright measures.
Safeguarding the balanced internal market
When legislating with the internal market goal in mind, the EU has not achieved complete homogeneity of rules in many policy fields, including copyright law. Perhaps such absolute harmonisation was not always a desired end goal of the EU legislator. It would not genuinely guarantee a level-playing field for all players in all Member States in a specific market. The EU legislature must consider the overall competitive environment in each Member State and assess whether there are indeed any genuine obstacles to free movement for the internal market. Thus, as Gareth Davies has argued, balance between diversity and harmonisation is key. In achieving this balance, the EU legislator resorts to several tools: the Better Regulation Agenda, the subsidiarity and the proportionality principles.
The EU has committed itself to designing policies and laws with a greater level of transparency and evidence, backed up with the views of citizens and stakeholders. This author argues that, at this stage of economic and socio-cultural research, the assumption maintained by the supporters of positive legislation is borderline speculation. Despite the vast and constantly growing literature on the intersection between copyright and AI, not a single EU-wide impact assessment has been carried out to evaluate whether European copyright law requires harmonisation at an EU level with regard to machine learning and computational creativity. Moreover, policies should not be imposed, but prepared inclusively, listening to the views of those affected by the legislation. This pertains to all stakeholders, not only those with the loudest lobby voice in Brussels. Copyright law is a public issue and as such it requires the input of the public. Such a consultation recently took place in the UK (here and here), but no such efforts have been made on an EU level. As a result, potential legislation in this field risks not only a one-sided representation of the interests of only certain stakeholders, but could also generate excessive costs (legislative, compliance, licensing, among others).
Subsidiarity and proportionality
Since the internal market is a shared competence, both the Union and the Member States may legislate and adopt legally binding acts. The limits of the Union competences in that respect are governed by the principles of subsidiarity and proportionality.
The central idea behind subsidiarity is that in areas which do not fall within the Union’s exclusive competence (so, the internal market and, hence, copyright law), the Union shall act only if and in so far as the objectives of the proposed action cannot be sufficiently achieved by the MS, but can rather, by reason of the scale or effects of the proposed action, be better achieved at a Union level. Subsidiarity may be seen as highly political and potentially ineffective. In fact, it has created no difficulties for copyright legislation.
Proportionality instead may act as a major barrier for any potential legislation in this field. It requires that whatever measure is proposed at an EU level must be proportionate to the interest pursued. In other words, let us not kill a fly with an elephant gun. Generally, it entails three steps:
- assessment of the suitability of the measure for the attainment of the objective (the appropriateness principle);
- evaluation of the necessity of the measure (are there other, equally suitable, less restrictive measures capable of attaining the same objective); and
- balancing the negative impact of the restrictions imposed against the added value (proportionality stricto sensu).
Applying this to the AI/copyright scenario, the suitability test requires that copyright law be the most appropriate measure to attain the objective at stake. Thus, copyright protection would be suitable if there is an existing or imminent obstacle to trade in the context of AI-generated works and if left in the public domain, the functioning of the internal market would be disturbed. As above, at this stage of research, there is not enough evidence to support this assertion, so it is questionable whether the suitability test will be met. Nonetheless, even if such evidence emerges, the necessity test, namely the second factor, is what could present more serious obstacles to pass legislation of this kind. Copyright protection must be the least restrictive measure to achieve the said objective. Here, potentially significant challenges emerge with respect to copyright duration, which is particularly long, and its scope, whereby economic rights have traditionally been interpreted broadly. AI processes can generate a large number of literary, musical and artistic works in the span of several seconds. In light of the term of protection, if these works are automatically covered by copyright law, then the public domain will inevitably be jeopardised, and for a very long time. This brings the discussion to the third factor – proportionality stricto sensu. It is essential to consider and respect the interests of stakeholders other than the AI creation and dissemination teams. An open and inclusive public discussion on copyright and AI via public consultations is essential and it appears that here it is absent.
In sum, should copyright law be extended to protect AI-generated works, the proportionality principle must necessarily step in and ensure that the EU measure does not lead to over-protection, an eventual “tragedy of anticommons” and overexploitation of authorial rights. Unfortunately, in practice, it is questionable whether and to what extent these important constitutional safeguards would have a real effect. Subsidiarity and proportionality have often been criticised for being mere methods of window dressing. Like Stephen Weatherill argues with reference to the proportionality principle, “only legislative choices that verge on the absurd are likely to be condemned as manifestly inappropriate”.
This blogpost is based on a forthcoming article by the author which is accepted for publication in the European Law Review and will be available in April 2023. For further information, please contact the author directly.