backIn May 2023, the Bill of Law No. 2,338/2023 (“PL 2,338/23″ or “Bill”), which aims to regulate artificial intelligence (“AI”) systems in Brazil, was presented to the Senate for analysis and vote. The text is the result of the work of a committee of jurists that analyzed several proposals related to the subject, in addition to regulations from other countries that already deal with this matter.
Largely inspired by European legislation, the Bill mandates that AI systems available in Brazil must be self-classified by the developer according to their risk, including excessive risk ratings, which AI systems are prohibited, and high-risk classifications, which systems can be made available upon the meeting of specific compliance requirements, such as algorithmic impact assessment, robustness, accuracy, reliability, and coverage tests, as well as human supervision to prevent the occurrence of risks arising from of its use to the rights and freedoms of individuals.
The Bill lists as high-risk tools those used for the following activities:
- Credit rating;
- People identification;
- Administration of justice;
- Implementation of autonomous vehicles;
- Medical diagnoses and procedures;
- Decision-making about access to employment, education, or public and private essential services;
- Assessment of students and workers;
- Management of critical infrastructure, such as traffic control and supply system of water and electricity;
- Individual assessment of the risk of committing crimes, personality traits, and criminal behavior.
The system’s degree of risk determines the civil liability in case of property, moral, individual, or collective damage. Even though the supplier or operator of the AI causing the damage is required to repair it integrally, regardless of the degree of autonomy of the system, they are also held liable for the damages caused, in spite of whether there is intent or fault, to the extent of their participation in the damages, when such systems are high risk or excessive risk. In any case, the fault of the agent causing harm will be presumed by applying the reversal of the burden of proof in favor of the victim.
PL 2,338/2023 also establishes that AI must not use subliminal techniques to induce people to behave in a manner detrimental or dangerous to their health and safety, nor can it exploit vulnerabilities of specific groups, such as those associated with age or disability, to induce harmful behavior. In addition, the Government is prohibited from using artificial intelligence to assess and classify people based on their social behavior or personality, with the intention of determining access to goods and public policies illegally and disproportionately.
Rights of People Affected by AI
Concerning the people affected by artificial intelligence systems, the Bill aims to protect the common citizen, establishing some minimum rights, such as prior information regarding interactions with artificial intelligence systems; explanation of the decision-making; recommendations or predictions made by AI systems; the right to challenge the decisions or predictions of artificial intelligence systems that produce legal effects or significantly impact the interests of the affected person; and the right to determination and human participation in decisions of artificial intelligence systems. The right to privacy and data protection is also reiterated in the document.
The text also emphasizes the right to non-discrimination and the correction of biases, explicitly prohibiting discrimination based on geographic origin, race, color or ethnicity, gender, sexual orientation, socioeconomic class, age, disability, religion, or political opinions. However, the adoption of criteria for the differentiation of individuals or groups is permitted when there is reasonable and legitimate justification “in the light of the right to equality and others fundamental rights”.
Also, as a form of protection, the Bill provides that in cases where the decisions of AI systems have a potentially irreversible impact or could pose a risk to the life or physical integrity of individuals, a high degree of human involvement in the decision-making process of the tool will be required.
In case of serious security incidents, such as threatening persons’ life or physical integrity, interruption of operation or provision of essential services, damage to the natural environment, or violation of fundamental rights, the competent authorities should be notified.
According to the current text, suppliers, and operators of AI tools will be able to adopt programs of governance aligned with the legislation. Although not mandatory, this type of practice can demonstrate the good faith of the accused and, consequently, be considered in cases of application of administrative sanctions, for example.
Copyright, Innovation, and Effectiveness
Still, PL 2,338/2023 deals with the issue of copyrights created by AI. In line with the legislation already in force on copyright, it authorizes, for example, the automated use of works by the systems of artificial intelligence, without the copying or use of such material constituting a copyright infringement, by research institutions, journalism, museums, and libraries, as long as the purpose is not reproduction and distribution of the work and that there is no harm to the economic interests of the copyright holders. This perspective expands the exceptions to copyright protection provided for in the Copyright Act (Law No. 9,610/1998).
Finally, the text also provides that the competent authority may authorize, by request of interested parties, the creation of an experimental regulatory environment called an experimental sandbox to stimulate innovation. This environment allows the temporary operation of products and services innovators without mandatory compliance with all industry rules but with the monitoring of Regulatory Agencies.
The law resulting from the approval of the Bill will take effect one year after its publication.
***
share
