Pacheco proposes regulatory framework for AI written by experts

The president of the Senate, Rodrigo Pacheco (PSD-MG), filed last night (5) a bill to establish a regulatory framework on the use of artificial intelligence (AI) technology in the country. The text was prepared by jurists and professors specializing in civil and digital law. One of the main points of the text is the prediction of liability of the supplier or operator of the artificial intelligence system, to the extent of its participation in any damage caused. In the case of AI whose application involves high risk, this responsibility will be objective, that is: it will not depend on proof of guilt or intent (conscious intent). In the case of a system with a risk below high, the supplier or operator of the system will be presumed guilty, and it is up to him to prove that he is not guilty, with the victim of the damage exempted from proving such guilt. The text does not provide for criminal sanctions, but it describes the obligations of the competent authority to monitor compliance with the rules, among which is to apply administrative sanctions for breaching the rules, including a fine of up to R$ 50 million for individuals and of up to 2% of billing in the case of legal entities. Risk gradation The entire project is based on the idea of ​​risk gradation in the application of AI – from low and moderate to high and extreme. In these last two cases, stricter governance rules are established for the use of technology, including the elaboration of human intervention and deactivation mechanisms. In addition to the use in autonomous vehicles and the operation of strategic infrastructure – such as an energy transmission system –, among the high-risk activities in the application of AI, the text lists the recruitment and evaluation of employees, the evaluation of criteria for access to public and private services and credit assessment, healthcare applications and biometric identification systems. The understanding is that, in these cases, the use of AI systems may lead to direct, indirect, illegal or abusive discrimination, due to personal information such as skin color, sexual orientation or geographic origin. “In addition to adopting definitions on direct and indirect discrimination – thus incorporating definitions from the Inter-American Convention against Racism, enacted in 2022 –, the text focuses on (hyper) vulnerable groups both for the qualification of what may be a high-risk system and for the reinforcement of certain rights”, says the justification for the project. The text places limits on the use of AI systems in public safety for biometric identification in public spaces, for example, such as facial recognition cameras. In this case, the use of technology is considered to be of extreme risk and will only be permitted upon approval of a specific federal law and court authorization in three cases: Search for suspects of crimes with a maximum sentence of more than two years in prison; Search for victims of crimes and missing persons; and Crimes in the act. The entirety of the bill, which replaces three other previous initiatives, can be accessed on the Chamber of Deputies website. Elaboration The text was elaborated by a committee of 17 people, including lawyers, professors of law and technology, as well as a criminal expert from the Federal Police and a legislative consultant, in addition to the support of several civil servants. The work was coordinated by Justice Ricardo Villas Bôas Cuevas, from the Superior Court of Justice (STJ). Throughout 2022, the group promoted four public hearings and an international seminar, in addition to 12 thematic panels, and said it had heard more than 60 specialists from different areas on different aspects of the theme. A study was also commissioned on regulations in 30 countries that already adopt some type of rule. The text must now be analyzed by the thematic committees of the Senate.

Agência Brasil