infoJustice

Christophe Geiger frames the copyright issues surrounding generative AI within a human rights framework, highlighting the classic tension between fostering new culture and protecting authors’ interests. He argues that this framework supports the right to train and develop generative AI to explore new artistic and scientific expressions, which aligns with the right to science, culture, and research. However, this must be balanced with the author’s right to benefit from their work and the protection of their moral and material interests. Geiger emphasises that human rights provisions, such as Article 27 of the Universal Declaration on Human Rights, advocate for securing authors’ interests to achieve cultural and scientific development, but do not mandate that this protection must take the form of exclusive intellectual property rights. This flexibility allows for alternative solutions, like remuneration systems, that are not based solely on exclusivity.

Geiger critiques the current legal approaches to training AI models in both the US and the EU. He notes the growing uncertainty in the US as to whether training AI qualifies as “fair use,” which is an unpaid use, with recent court developments and a US Copyright Office report suggesting a more nuanced view than previously held. In the European Union, he points to the “bizarre” text and data mining (TDM) exception that permits commercial use unless rightsholders opt out. He argues this opt-out system leads back to licensing schemes where authors are unlikely to benefit due to unequal bargaining power with producers and publishers, who are the primary rightsholders controlling the licenses. Essentially, Geiger contends that neither the all-or-nothing “fair use” model nor the EU’s opt-out system effectively ensures authors receive fair compensation for the use of their works in commercial AI systems.

As a solution, Geiger proposes a statutory remuneration system for the commercial use of copyrighted works in training generative AI, with payments going directly to authors. This model is based on the concept of “remunerated exceptions,” which are well-established in Europe and have proven effective at generating significant revenue that can be distributed favourably to authors, sometimes in a 50/50 split with producers. Specifically, he suggests replacing the EU’s current opt-out provision with a mandatory statutory remuneration scheme for commercial TDM activities. This approach, he argues, aligns with a human rights-compliant reading of copyright law by securing authors’ right to fair remuneration for the commercial exploitation of their works, thereby incentivising creativity while still permitting the development and use of AI technology.