AI transparency: an emerging principle in the IP ecosystem?
Jacques de Werra · Journal of Intellectual Property Law & Practice · 2023
Transparency is undoubtedly one of the most fundamental pillars of the artificial intelligence (AI) policy ecosystem. The recent large-scale availability and widespread use of generative AI tools—such as ChatGPT—and of other increasingly powerful AI tools make transparency an even more critical issue. This explains why the principle of algorithmic transparency has—unsurprisingly—found its way in most (if not all) leading AI policy regulations. By way of example, the European Union has just published (on 15 December 2022) a ‘European Declaration on Digital Rights and Principle for the Digital Decade’,1 which has the ambition (according to its preamble) to ‘promote a European way for the digital transition, putting people at the centre’. The Declaration includes a chapter on ‘Interactions with algorithms and artificial intelligence systems’, in which the European Parliament, the Council and the Commission (referred to as ‘we’ in the Declaration) commit in particular to ‘ensuring transparency about the use of algorithms and artificial intelligence, and that people are empowered and informed when interacting with them’. This requirement for transparency in the use of AI tools is generally associated with a so-called right to algorithmic transparency. To be sure, this right can have quite diverse implications. It can affect both the developers of AI tools (so that they shall disclose how the AI tool works) and the users of such tools (so that they shall disclose to the public whether they have used such AI tools). In the IP ecosystem too, transparency is becoming of central importance. The concept of transparency is also quite diffuse due to its multiple facets. Sometimes, transparency applies for the benefit of the right holders, in order to protect their (financial) interests and specifically those of the authors. This is what results from the European Directive on Copyright in the Digital Single Market (2019/790), which provides for a right of the authors and of performers to receive on a regular basis ‘up to date, relevant and comprehensive information on the exploitation of their works and performances from the parties to whom they have licensed or transferred their rights, or their successors in title, in particular as regards modes of exploitation, all revenues generated and remuneration due’ (art 19). In the same vein, the Directive states that ‘Member States shall provide that online content-sharing service providers provide rightholders, at their request, with adequate information on the functioning of their practices with regard to the cooperation referred to in paragraph 4 and, where licensing agreements are concluded between service providers and rightholders, information on the use of content covered by the agreements’ (art 17 para 8). In the IP ecosystem, transparency can, however, also be a tool that can be invoked against right holders and also against applicants who have filed for IP protection, to avoid the misappropriation of IP rights and to protect the legitimate interests of third parties. This is what is provided for under patent law in order to identify the source of the genetic resources or traditional knowledge on which an invention would be based. This obligation of transparency could potentially be anchored at the global level in a future International Legal Instrument related to Intellectual Property and Traditional Knowledge/Traditional Cultural Expressions that is presently discussed at the World Intellectual Property Organization in Geneva in the Intergovernmental Committee (IGC) on Intellectual Property and Genetic Resources, Traditional Knowledge and Folklore. Based on the (non-paper) first draft text prepared by the IGC Chairman, Ms Lilyclaire Bellamy (dated 21 February 2023) stated that ‘[i]ntellectual property applications developed using traditional knowledge shall include information on the indigenous [peoples] and local communities or other beneficiaries providing such traditional knowledge’2 and ‘[t]he application shall also state whether free, prior and informed consent or approval and involvement to access and use has been obtained’ (ibid.). Quite interestingly, transparency is also anchored in the very recent Agreement under the United Nations Convention on the Law of the Sea on the conservation and sustainable use of marine biological diversity of areas beyond national jurisdiction (adopted in March 2023).3 Even though this agreement is obviously not an IP treaty, it still remarkably provides for very detailed obligations of transparency in order to ensure the ‘fair and equitable sharing of benefits arising from activities with respect to marine genetic resources and digital sequence information on marine genetic resources of areas beyond national jurisdiction for the conservation and sustainable use of marine biological diversity of areas beyond national jurisdiction’ (art 7 lit a) and does expressly refer to patents in this context (art 10 para 6 lit a). Beyond this specific situation related to patent law and genetic resources, the phenomenal development of generative AI tools pushes the IP ecosystem to include a principle of algorithmic transparency in order to identify the human or non-human origin of creations so that it can be decided whether these creations shall be worthy of protection under IP law, specifically under copyright law. This is what the US Copyright Office has done in its very recent ‘Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence’ (applicable as of 16 March 2023).4 The Guidance provides that ‘[a]pplicants have a duty to disclose the inclusion of AI-generated content in a work submitted for registration and to provide a brief explanation of the human author’s contributions to the work’. It can be anticipated that more precise guidelines will progressively emerge in order to distinguish what shall constitute human creation and what shall not. In any event, this specific obligation of algorithmic transparency that can arise in the copyright framework (leaving aside the discussion about its compatibility with art 5 para 2 of the Berne Convention) resonates with the general obligation of algorithmic transparency (as provided for in the European Declaration on Digital Rights and Principles for the Digital Decade). There is indeed a certain convergence between their respective objectives, namely, to inform humans about the human or non-human source of a communication that appears to originate from a human (knowing that—still imperfect—technologies are also being developed for that very purpose). From this standpoint, the general right to algorithmic transparency, which constitutes one of the digital fundamental rights, could potentially help legitimize the development of a special right to algorithmic transparency under IP law (and—by mirror effect—of a special obligation of transparency that would be imposed on alleged right holders/IP applicants), in order to protect society’s interest in keeping in the public domain what should remain there. The convergence between these two facets of algorithmic transparency confirms—if necessary—the major influence of fundamental rights on the intellectual property ecosystem, an influence that has always existed but which has become even more critical in the digital environment. The future (or ChatGPT?) will tell whether, when and how a right to (algorithmic) transparency could become an emerging component of the global IP ecosystem.