Elaborating a Human Rights friendly Copyright Framework for Generative AI

As works are increasingly produced by machines using artificial intelligence (AI) systems, with a result that is often difficult to distinguish from that of a human creator, the question of what should be the appropriate response of the legal system and, in particular, of the copyright system has become central. If the generator of copyright protection has traditionally been the author’s creative input, AI forces us to reassess what in the creative process is special in human creativity and where the creative input lies in AI-generated works. But it also poses more fundamental questions on what the copyright system should achieve and who/what it should protect. In particular, since many human authors will potentially face the competition of these AI machines on the market, new ways of remunerating creators will have to be imagined while making sure that the copyright system does not stand in the way of these important technological developments. This contribution analyses the copyright issues related to so-called “generative AI” systems and reviews the arguments currently being advanced to change the copyright regime for AI-generated works. To do so, the underlying human rights framing intellectual property laws are used as the starting point from which a balanced copyright framework for generative AI could (and even should) be derived. It follows from the applicable human rights framework for copyright, but also from the anthropocentric approach of human rights, that the protection of creators and human creativity must be considered the point of reference when assessing future reforms with regard to copyright and generative AI systems. This approach establishes generative AI systems as an instrument of the human creator – and not as a substitute. It also reinforces the notion that copyright should be a tool to protect creativity and creators, not a legal mechanism to secure the amortization of economic investments in AI technology. As a consequence, it is argued that the copyrightability of AI-generated outputs should be considered with utmost care and only when AI is used as a technical tool for creators in their creation process – in other words, when they can serve a human author. At the same time, AI systems are here to stay, and their development should not be inhibited, as they can have many beneficial aspects (including for creators) if appropriately regulated. The right to train generative AI systems via machine learning technology can be derived from the right to science and culture and freedom of (artistic) expression (Arts. 19 and 27(1) Universal Declaration of Human Rights (UDHR); Art. 15(1)(a) and (b) International Covenant on Economic, Social and Cultural Rights (ICESCR); Arts. 11 and 13 EU Charter of Fundamental Rights (EUCFR)), as AI can lead to useful advances in science and the arts; moreover, it is important for human creators to be able to use outputs produced by generative AI in their creative process. This grounding is even stronger when the training is conducted for research purposes, as the training process can then also benefit from the fundamental right-to-research justification. However, since a large quantity of copyrighted works is required for the training of generative AI systems, a remuneration obligation for these uses arises from a human rights perspective, in particular when AI systems have a commercial purpose. It follows from the right to the protection of the creator’s moral and material interests (Arts. 27(2) and 17 UDHR, 15(1)(c) ICESCR; 17(2) EUCFR, 1 Protocol No. 1, 8 European Convention on Human Rights (ECHR)) that authors must be adequately remunerated for the commercial use of their works unless there is a strong justification legitimizing the use. For this reason, it is proposed that the machine learning process using copyright-protected works to train the AI gives rise to a limitation-based remuneration right to the benefit of human creators. The article also briefly explores if and when the moral interest of creators deriving from human rights protection could justify their opposition to the use of their work for the purpose of training AI systems. It is argued that the weaker the fundamental rights claim to train the AI is, the stronger the moral rights claim could be. For example, training an AI to produce works for discriminatory or racist purposes will benefit from a weaker (if any) fundamental rights protection, but will potentially raise important moral concerns of the author of the work used for training purposes. More generally, the article concludes that in order to secure a vibrant space for culture and creativity, (finally) cherishing and putting the Human Author at the center of the copyright system is necessary (and not only to erect fences to the benefit of copyright industries, which could be the unfortunate result of the recent first broad regulatory intervention on AI by the EU, the so-called “Artificial Intelligence Act”). In doing so, it might be possible in the future to have AI-systems that serve creators and creativity, and not the other way around.