Main provisions and principles
Congressional amendments
What steps remain before the bill can be signed into law?
On 29 September 2021, Brazil's House of Representatives approved the Brazilian Artificial Intelligence Bill (Bill No. 21/2020), passing it on to the Federal Senate for further analysis. Should it eventually be signed into law, the bill will establish principles, duties and guidelines for developing and applying artificial intelligence (AI) in Brazil. The bill will also bring harmonisation with other important Brazilian legislation, such as the Brazilian Data Protection Law (LGPD) and the Brazilian Consumer Protection Code (CDC).
Main provisions and principles
The bill defines an "AI system" as one based on a computational process that can learn, perceive and interpret the external environment by processing data and information from a set of human-defined objectives. The system must also be able to interact with the external environment, making predictions, recommendations, classifications or decisions, by applying techniques such as:
- machine-learning systems, including supervised, unsupervised and reinforcement learning;
- knowledge or logic-based systems;
- statistical approaches, Bayesian inference; and
- search and optimisation methods.
Based on this definition, a system's ability to learn how to perceive, interpret and interact with the external environment and the information it receives appears crucial to understanding whether the system would fall within the scope of the bill's provisions. As such, the bill's provisions would not apply to regular automation processes, as they are exclusively guided by predefined programming parameters that do not include such abilities.
From the outset, the bill establishes that AI systems developed or implemented in Brazil must aim to advance scientific and technological development. Among other goals, these systems should also promote:
- sustainable, inclusive economic growth and the welfare of society;
- increased competitiveness and productivity within Brazil's economy;
- research and development to enhance innovation in productive sectors; and
- environmental protection and preservation.
These goals must align with certain fundamental principles for developing AI in Brazil, including:
- scientific and technological development and innovation;
- free enterprise and competition; and
- respect for:
- ethics;
- human rights;
- democratic values; and
- regional diversity and inclusion.
Further, the bill emphasises the need to promote industry self-regulation for developing and implementing AI systems, and adopting codes of conduct and best practice guides to assist with industry compliance. It also encourages adopting certain regulatory instruments that enhance innovation, such as regulatory sandboxes and regulatory impact analysis for AI systems.
After extensive debate in the House of Representatives, the bill includes the following principles, which would have to be observed when developing or implementing AI systems in Brazil:
- beneficial purpose – the pursuit of beneficial outcomes for humanity;
- human-centric AI – respect for human dignity, privacy, personal data protection and fundamental rights;
- non-discrimination – AI systems must not be used for discriminatory, unlawful or abusive purposes;
- neutrality – AI systems must be structured, developed, implemented and used in a way that reduces the possibility of unlawful bias;
- transparency – the right of individuals to be informed in a clear, accessible and accurate manner about the use of AI solutions – unless they are subject to trade and industry secrecy;
- security and prevention – the management and mitigation of security risks arising from AI systems' operations throughout their entire life cycle and continued operation;
- responsible innovation – agents involved in AI systems' development and operations must document their internal management processes and would be held liable for any issues involving such systems; and
- data availability – non-infringement of copyright arising from the use of data, databases and texts protected by it for training purposes of AI models.
Based on the bill's provisions, public authorities must observe the specific risks that AI systems pose when regulating the use of AI. In other words, the respective degree of intervention should always be proportional to the risks each AI system poses. Moreover, the potential social and economic benefits a given AI system offers must be continuously assessed against the risks posed by similar systems that do not involve AI. Public authorities that seek to adopt rules impacting the development and operation of artificial intelligence systems must first conduct a regulatory impact analysis.
The bill may hold agents involved in AI systems' development and operations liable for breaches of the law (whether intentional or due to negligence) unless otherwise provided for by specific legislation, as is the case with the CDC. There is a strict liability framework for consumer claims under the CDC, meaning suppliers may be liable regardless of any actual fault or intent – consumers only need to prove the relationship between the supplier's actions (or lack thereof) and the damage they were subject to.
The main situations that may trigger liability under Brazil's consumer laws relate to erroneous or defective products and services, a failure to provide clear information to consumers on the risks and limitations of products and services and misleading advertising. There is an interesting legal debate about whether an AI system failure could be considered an erroneous or defective product or service in the context of a consumer relationship, and in which circumstances.
What steps remain before the bill can be signed into law?
Having been approved in the House of Representatives, the bill has proceeded to the Federal Senate for joint analysis with two other proposals that also address the development of AI in Brazil (Bills No. 5,051/2019 and No. 872/2021). A special committee of legal experts has also been established, which is in charge of drafting a new text for the bill. However, there are still important points that these bills have not addressed, such as the parameters for enforcing the law and corresponding penalties.
Moreover, considering the LGPD is now in effect – which significantly impacts several aspects of using AI systems – and the potential for this technology to be adopted on a large scale in Brazil, the National Congress and the special committee must discuss solutions to specific problems concerning AI systems, in order to increase the potential benefits AI may offer Brazilians.
For further information on this topic please contact Fabio Ferreira Kujawski, Paulo Marcos Rodrigues Brancher, Thiago Luís Sombra or Maíra Scala at Mattos Filho by telephone (+55 11 3147 7600) or email ([email protected], [email protected], [email protected] or [email protected]). The Mattos Filho website can be found at www.mattosfilho.com.br.
Ingrid França Moraes Soares assisted in the preparation of this article. An earlier version of this article was first published on Único.