Copyright is not averse to new technologies. Its history is intrinsically linked to technological development. At each stage, revisions, adjustments, and adaptations to the existing organizational and legal structure are necessary and ideally preceded by wide-ranging and informed debate. The same holds true for Artificial Intelligence (AI) technologies. And the interaction between copyright and AI makes open public discussion even more important.
Overall, the existing legal framework in Latin America is not adequately structured to address uses that are relevant to contemporary research or the development of AI systems, such as Text and Data Mining (TDM) practices, since, for example, most countries of the region lack a general exception for research and/or TDM activities. However, some countries, like Brazil, are actively discussing provisions to regulate issues at the intersection of AI and copyright.
Brazil began taking its first steps toward AI regulation around 2018. Copyright made its appearance in the proposed Bills in 2020, with the focus on TDM. However, starting in April 2024, the latest revisions have expanded to include critical copyright-related issues that are now being debated globally – such as transparency, opt-out, and remuneration.
Within a period of around two months, several official versions were made public by the Senate, where the Bill is under scrutiny. Based on the latest version – the ‘Rapporteur’s vote supplement’ – we consider the following topics particularly relevant for the AI/copyright discussion: (i) transparency and information on the use of protected works in AI system training (art. 60); (ii) copyright limitation for TDM for research purposes (art. 61); (iii) possibility of opt-out (art. 62); (iv) remuneration for copyright holders for use of their works in AI system development (art. 64); and (v) protection of image and voice (art. 66).
In addition to harmonizing and better systematizing terms and concepts, we see it as highly relevant to distinguish more precisely between the various situations affected by the proposed norms. This is because each situation presents unique characteristics that require tailored regulation, and treating different scenarios in the same way increases the risk of injustices and unforeseeable negative effects, which may actually undermine the effectiveness of the legislation itself.
Among the necessary distinctions, we highlight the following: (i) differentiating authors and artists from intermediaries that often hold the economic rights in their works, both as qualified subjects for the exercise of rights and as privileged recipients of any remuneration; (ii) research activities in general from those centered on the development of AI systems, especially regarding the conditions for using TDM, which are essential for any data-intensive research; and (iii) the development (and offering) of AI systems in general from generative AI systems.
Broadly speaking, authors and artists who are hoping that their well-founded fears about the future of their professional activities will be addressed are unlikely to be effectively and equitably remunerated. Data-intensive research, even when not linked to the development of AI systems, could be hampered by the restrictive conditions for the use of TDM techniques. Furthermore, the development of national AI systems, especially those that are not generative, could be slowed down by the increase in requirements and high barriers for development and entry into the market, which a few large companies have so far dominated.
As mentioned before in relation to research rights, the provisions in Bill 2338/23 deal with these from various angles in different parts, and some of the issues involving research activities concern the (non-)application of the rules to certain research activities (art. 1, §1, c); principles, foundations, and promotion of research (art. 2, X; art. 57, II and III); access to data for research (art. 48, IX); (iv) text and data mining (art. 4, XIII; art. 61); and differentiated treatment of open and free standards and formats (art. 1, §2).
Although the text reflects strong recognition of the relevance of research, as the Brazilian Senate Bill stands today, it does not reasonably deliver on its promises, as it overly restricts all types of research by, in practice, deeming licensing necessary with the requirement of legal access for TDM.
In addition, it does not, as we believe it should, distinguish and regulate differently (i) research in general, without any direct relation to AI systems; (ii) research with AI, which only uses AI systems as instruments to achieve its goals; (iii) research on AI, which is related to the ongoing development of the field of Computer Science and others; and (iv) research on AI systems, which is essential to address problematic issues that are inherent to the development, training and use of these systems.
Regulating technologies, especially those for which we have yet to fully grasp the scale of their impact, will never be easy, complete, or flawless. For this reason, the initiative and those working on it deserve recognition. At the Brazilian Copyright Institute (IBDAutoral), we addressed the copyright-related issues raised in the most recent version of the Brazilian AI Bill (the ‘Rapporteur’s vote supplement’ linked above). The study was a contribution to the public debate not only on the issues briefly introduced above but also on others further developed in the text of the study, which is fully available in Portuguese, with executive summary in English and Spanish.
The debate is likely to intensify and new political arrangements around the issue are constantly developing. The existing regulatory models do not seem to fit the needs and circumstances of developing countries and the global south in general. Envisioning a satisfactory solution may require stepping out of the copyright box and some degree of legal innovation and detachment from the well known but probably inadequate existing structures. Only time will tell.
