Senate approves project to regulate AI in Brazil

Senate approves project to regulate AI in Brazil

The proposal includes risk levels for artificial intelligence systems, remuneration for copyright and user rights. The text goes to the Chamber of Deputies The Senate approved this Tuesday (10/12), in a symbolic way, the regulatory framework on Artificial Intelligence. The project establishes the rules for the use of artificial intelligence systems in Brazil, as well as defining the levels of technological risk. The text will still have to be evaluated by the Chamber of Deputies before being subjected to sanctions.




The proposal presented by the president of the Senate, Rodrigo Pacheco, in 2023, had as its main objective to establish the rights of people affected by the technology and define the parameters for the supervision and control of the activity. The text was submitted to a special commission created in the Chamber to evaluate the proposal and the final version was approved only last Thursday by the college, after intense negotiations with the opposition and companies operating in the sector.

The project defines risk levels for artificial intelligence systems, to which different regulations apply. The category is defined based on the impact a given system has on human life and fundamental rights. According to the text, technologies that fall into the “excessive risk” range cannot be developed.

If approved by the House, the rule will cover various applications, from those used to identify people, to more complex ones used in industry, traffic and energy supply. Tools integrated into search and content generation systems, widely used by users on social networks, also need to adapt.

The fine for failure to comply with the rules can reach R$50 million or 2% of the company’s gross revenue. In some cases, system suspension may also be applied, following a model similar to that used by courts to freeze X’s assets in Brazil. The possibility remains that the company will be held civilly liable if it is demonstrated that the system has caused damage to a person or a group of people. Administrative sanctions would be the responsibility of the National Data Protection Authority (ANPD).

Risk categories

Projects with “excessive risk” and which would therefore be banned include so-called autonomous weapons systems, the use of artificial intelligence to exploit human vulnerabilities that lead to behaviors harmful to health and safety, and the creation of that violate people’s rights. children and adolescents.

The assessment of personality traits and social behavior to predict the commission of crimes would also be prohibited.

The law allows, however, the use of cameras to identify people in real time and in public places, as long as the attention is paid to the search for victims, the finding of missing persons, the capture of fugitives, the ongoing repression of crimes whose punishment exceeds two years. in prison or execute warrants.

Among the “high risk” systems are autonomous vehicles, the selection of students for access to education, the recruitment and dismissal of workers and the management of priorities in emergency services. The list is “open-explanatory”, which allows for the inclusion of new technologies that fall into the category and influence citizens’ relevant decisions.

Part of the dispute in the Senate centered on the role that tools used by big tech like Meta would play in the new rules. In the text approved by parliamentarians, the content distribution algorithms on social networks, used by these companies, were excluded from the “high risk” category, which makes the monitoring activity more flexible.

Assessment

In high-risk systems, an algorithmic impact assessment must be carried out and technology providers must ensure the security of the system. Discriminatory biases also need to be mitigated.

The provision that placed “risk to the integrity of information, freedom of expression, the democratic process and political pluralism” as a criterion for defining the level of vulnerability of a technology was not accepted by the Commission at the request of the opposition and was never evaluated by the plenary.

Generative and generic AIs like ChatGPT will have specific rules. Among these, your synthetically produced content, such as text, video, or image, must contain a label, even if visually hidden, that shows it was created by AI.

Copyright

The final text retained a much-debated provision guaranteeing copyright on the works. The proposal allows you to search for and use protected content to improve artificial intelligence. However, the material must be obtained legitimately and cannot be used for commercial purposes. This is the case of research institutes, journalism, museums, archives, libraries and educational projects.

When there are commercial purposes, the owner of the technology must negotiate and compensate the copyright holder of certain contents. It can also deny that a work, for example a song, is accessed to train a certain system. Even the generative use of the image and voice must require prior consent and follow the Civil Code.

User Rights

The project also outlines rights for users of AI systems, which include the right to be informed in advance that they are interacting with Artificial Intelligence, the right not to be discriminated against and to use simple and clear language. Companies must also provide explanations for decisions made by the machine, as well as allow those decisions to be reviewed by humans when there is an associated risk.

gq (Agência Senado, Agência Brasil, DW, ots)

Source: Terra

You may also like