The road to regulation of artificial intelligence: the Brazilian experience

​

With the political approval around the EU AI Act on 8 December, Europe’s comprehensive regulatory framework joins other governments initiatives in response to growing concerns about AI safety with a surge in regulatory initiatives. These include the Bletchley Declaration following the UK AI Safety Summit to progress on the EU AI Act, the comprehensive AI governance strategy outlined in Biden's Executive Order, and the G7's statement on the Hiroshima process, endorsing an AI 'Code of Conduct'.

AI regulation in Brazil

Amidst the attention these initiatives have garnered, a significant effort in AI regulation has been unfolding also in Brazil, where an extensive bill on AI has been under discussion in the Congress. The legislative proposal currently under debate stems from a draft prepared by a commission of legal experts tasked with providing input for AI regulation in Brazil. Formed in February 2022, the commission engaged in profound discussions with representatives from civil society and numerous experts, crafting a legislative proposal tailored to the specificities of Brazil. Their task was to design a regulatory framework that addresses the unique challenges and opportunities within Brazil.

The imperative of AI regulation in Brazil cannot be underestimated, particularly given the extensive adoption of technology in both the private and public sectors of one of the world's most unequal countries. The impact of these technologies is disproportionately felt, exacerbating existing disparities in income, race, gender, and territories. For instance, Black women experience the lowest income levels, and police brutality disproportionately affects Black people. Predictive algorithms and facial recognition systems contribute to the problem, leading to numerous cases of wrongful arrests. Shockingly, in 2019, 90% of individuals arrested through facial recognition in Brazil were black.

Against this backdrop, the proposal elaborated by the legal experts was introduced to the Brazilian Congress and is currently under examination by the Senate (Bill No. 2338/2023). The proposal seeks to establish principles, rules, and guidelines to regulate the development and application of AI in the country. The Brazilian bill mirrors the EU AI Act in many aspects, as both texts reflect the OECD's expanded definition of AI systems, encompassing not only decision-making but also model creation and data training. Additionally, they share a risk-based approach, tailoring the regulatory obligations based on potential AI technology risks. This includes a similar list of high-risk applications and the prohibition of applications considered to pose excessive or unacceptable risks, such as those related to social scoring.

Learnings from the Brazilian case

Despite these similarities, the Brazilian bill stands out by adopting a rights-based approach alongside its risk-based framework. Contrary to common perceptions of legal transplant or a mere reflection of the Brussels effect, the Brazilian bill positions the country on par with more developed economies, striving to carve out its unique path in AI legislation. For example, a notable distinction of the Brazilian proposal from the EU AI Act lies in its guarantee of rights to individuals and the provision of judicial and administrative mechanisms to enforce these rights. While establishing fundamental rights for all interactions between machines and humans, such as information and transparency, it imposes additional obligations when the AI system has a significant impact or produces relevant legal effects. This includes the right to contestation and human intervention, ensuring a fair and comprehensive defence – an informational due process – for individuals affected by automated decisions. This rights-based approach also explains the choice to include in the Brazilian bill rules related to civil liability, covering both the supplier and the operator of the AI system.

While the approval of the EU AI Act is expected to fuel momentum for AI regulation in Brazil, the country’s intricate landscape of stakeholder interests suggests a prolonged journey before the Brazilian Congress finalises a text. The Senate’s Temporary Committee, tasked with evaluating AI legislative proposals, has become an arena for political and legal debates, revealing evident disputes. In December 2023, Senator Astronaut Marcos Pontes presented a new text for Bill 2338, signalling a setback by eliminating chapters on the rights of those affected by AI systems, governance measures, and surprisingly, algorithmic impact assessments – the foundation of any risk-based approach. This proposal, which moves the discussion on innovation and regulation away from human rights considerations, contradicts Brazil's need for AI regulation rooted in the integration of constitutional values into decision-making processes.

A new version of Bill 2338 is expected from the rapporteur of the Temporary Committee, Senator Eduardo Gomes, with hopes that it will cover issues not addressed in the original draft, such as rules for generative AI, a more detailed risk classification, and specifications on how rights can be exercised. 

Future changes in the context of generative AI should address transparency for data subjects and deployers, as well as measures to ensure that data management tools can effectively mitigate non-discrimination risks in the training phase. In addition, crucial discussions surrounding how AI interacts with social media content moderation tools and strategies to mitigate disinformation and harmful practices associated with large language models and synthetic media need to be held. Notably, the EU AI Act opted against designating the use of AI in social media companies' recommendation systems as a high-risk application, citing coverage under the systemic risks outlined in the Digital Services Act (DSA). In contrast, Brazil lacks such comprehensive platform regulation, with discussions surrounding the Brazilian bill for platform transparency and accountability (Bill No. 2630/2020) currently at a standstill – making it a critical topic, especially with the upcoming municipal elections in Brazil in 2024.

In conclusion, it is clear that Brazil needs to advance discussions on establishing acceptable levels of risk posed by emerging technologies and the corresponding strategies for addressing them. Decisions of this nature should not be exclusively delegated to corporate entities – as evidenced by the recent OpenAI controversy amid AI safety disputes – but should also undergo thorough public and democratic scrutiny. Achieving this in Brazil necessitates significant progress in essential debates surrounding AI regulation and related fields, including platform regulation.