Brazil: Proposed AI regulation
Artificial intelligence ('AI') and the regulation thereof is a highly contentious topic, and in Brazil this is no different. Simone Mendes Santinato, Director - Privacy and Data Protection Business Unit at Novared Brasil, discusses this in relation to recent proposed legislation and the controversies that this has stirred.
Just like practically every other country in the world, Brazil is concerned about how to regulate the use of AI. AI has been widely implemented in many different aspects of life, from everyday scenarios such as simple movie or music recommendations on streaming platforms, to new discoveries for the treatment of serious illnesses, financial risk management, among other applications that can more profoundly affect people's lives.
Thus, Brazil is rushing to accelerate the expansion of AI use in its territory, while also addressing its side effects. A major concern is that of 'machine bias', considering that the data sources and quality of the data could directly influence the outcome of how AI will produce responses. For example, there is a 'geographic' factor that could potentially produce bias, as the majority of the applications are developed outside Latin America, thus failing to take into account the cultural aspects of Brazil, such as facial recognition patterns, regional languages, and accents. As the vast majority of these algorithms collect their data from North America and Europe, its learning inherits a regional bias. If implemented without proper testing and approval in Latin America, there are indeed risks of damage in the short term as well as on a large scale.
Therefore, the complexity of an AI project goes far beyond the technical issues, involving ethics, culture, justice, governance, accountability, and other aspects that require a broad debate by the academic community and society at large. In July 2021, the Ministry of Science, Technology and Innovation published its 'Brazilian Strategy for Artificial Intelligence'1 ('the Strategy'), aiming to stimulate research, innovation, and development of AI solutions.
The Strategy outlines its objectives as to:
- contribute to the elaboration of ethical principles;
- promote sustained investments in research;
- remove barriers in innovation;
- train and prepare professionals;
- stimulate innovation and development; and
- promote an environment of cooperation between public and private entities.
The Strategy (which is merely a proposal and not law) proffers a diagnosis of the current situation in regards to AI and details challenges and visions for the future. The Strategy was created based on advise of a specialised consultancy, which complemented a benchmarking and public consultation process. As such, the way in which Strategy was developed creates an expectation in Brazilian society that the legislative process would follow the same line (i.e. technical and collaborative, through public consultations).
Thus, it was somewhat surprising that, on September 2021, the House of Representatives approved a law that seeks to establish foundations and principles for the development of applications using AI ('the Law'). The Law has yet to be approved by the Senate, but it has been the subject of much criticism.
The first point of content has been the 'urgency process' chosen by the legislature. The complexity of the topic is indisputably incompatible with such an urgent approval, this procedure being problematic also because it ignores the need for collaboration. The lack of communication with leading experts in the field, the technology community, and technical institutions has been a constant critique after news of the Law's approval.
Specialists, universities, associations, jurists, and society in general would benefit from participating more actively in the legislative process, through public consultations and a format that is more open to discussions. This would also be favourable to legislation that offers greater legal certainty to the community.
Items such as technological development, respect for human rights, democratic values, equality, sustainability, and the reduction of inequalities are included in the text without, however, clarifying how to achieve these goals (which are undoubtedly desirable). The Law is composed of 16 Articles and proposes some principles, duties, rights, and governance instruments for AI to be followed by public and private initiatives. However, the Law fails to clarify how these instruments operate.
The text of the Law refers to the need to create complementary laws that effectively regulate some topics. Even the definition of AI by the Brazilian legislator has been a matter of discussion by the scientific community. The Law as proposed defines AI as 'a system based on a computational process that can, for a certain set of goals defined by man, make predictions and recommendations, or make decisions that influence real or virtual environments.' Members of the academic and scientific community have criticised the definition, claiming it does not make sense.
Another criticism of the Law is that some Articles may represent an impediment to technological growth (contrary to what should actually be one of its basic principles). An example is contained in Item IV of Article 9, which, when dealing with the duties of agents, determines that the system must be closed if human control is no longer possible. This proposition goes against the very purpose of AI, which is to make decisions instead of relying it to a human being. A well-trained AI system may be able to recognise patterns that are barely visible to a human being, even if they are an expert in a particular area. Therefore, this 'shutdown' of systems may not make any practical sense.
Since some Articles contradict the understanding of the research community and as the Law does not provide mechanisms to ensure compliance with the proposed rights and duties, many feel that it will be impractical and ineffective in application. Some members of the Representative House, however, defend its contents, stating that the text should largely be concerned with proposing more general principles and that such principles should be complemented by specific legislation, seeing as the matter is constantly being updated due to the speed of technological development. Advocates of the text say that seeking to regulate specific issues could render the Law obsolete in a short period of time. At the time of writing, we are currently involved in a debate process regarding the regulation of AI activity in Brazil.
In the last week of October 2021, following increasing tension over the Law's text, a group of civilians signed an 'Open Letter' to the Senate. This contains strong criticism of the content of Article 6, Item VI, which establishes subjective civil liability as a standard regime for damages caused by AI systems.
This Article contradicts the common understanding built by doctrinal, national, and international studies on the matter. Moreover, there is a real possibility that victims under this regime will be prevented from obtaining compensation for damage suffered, which is contrary to the Brazilian Federal Constitution. In cases where the production of evidence by the victims is not feasible, such regime would create a scenario where the agents behind the AI system would not be held responsible.
It is likely that the Senate effectively understands that there is no viable urgent procedure for a matter as complex as this one. The matter must be left for further debate and contributions from the technical and academic community through public consultations.
The situation described above is in stark contrast to the approval of the Law No. 13.709 of 14 August 2018, General Personal Data Protection Law (as amended by Law No. 13.853 of 8 July 2019). In the latter's case, there were a lot of public consultations, as well as a lot of participation from the technical, academic, and legal communities, which allowed for the approval of a more effective legal text. The expectation is that the debate will also be open in relation to the theme of AI which, due to its complexity, does not allow for hasty approval.
Simone Mendes Santinato Director - Privacy and Data Protection Business Unit
Novared Brasil, São Paulo