Think Future | Informative Report
On 29th September, the Brazilian House of Representatives approved Bill n. 21/2020, also known as the “Legal Framework for Artificial Intelligence”. The proposed regulation had 413 votes in favor and 15 against, and the approved version was a substitute project presented by Congresswoman Luísa Canziani. Now, the Bill must go through the Federal Senate legislative process, where it might be approved or not.
The Bill was first proposed by Congressman Eduardo Bismarck (PDT-CE) to be the first regulation regarding the development and application of Artificial Intelligence in Brazil; its current version, like the first one, proposes only basic principles and abstract rules. Therefore, it presents a series of fundamentals, objectives and guidelines for the Public Administration regarding the matter, while it also predicts that such guidance will be complemented by future specific regulations. It is expected that future proposals shall cover, amongst other topics, issues related to the responsibility of AI actors.
About this topic, however, the proposed Bill already determines that the rules that may regulate it should be based on non-strict liability, hence considering the effective participation of agents, the damages that one wishes to avoid or remedy and the adaptation of such agents to good practices and international standards.
The draft bill defines Artificial Intelligence systems as those based on computational processes that, from a set of human defined objectives, can, through data and information processing, be carried out independently of human action, learning to perceive, interpret and interact with the external environment, making predictions, recommendations or classifications.
The principles that guide the development of AI-based technological systems in the country are also noteworthy, especially the principle of transparency, which determines that, unless otherwise provided by law, and observing commercial and industrial secrets, people have the right to be informed in a clear, accessible and accurate manner regarding the use of artificial intelligence solutions in the following situations:
- To state that a person is interacting directly with artificial intelligence systems, such as through conversation-trained robots that provide personalized online services (chatbots);
- To provide the identity of who operates the system autonomously or individually – or about the legal entity responsible for the operation of artificial intelligence systems, and
- The AI-based system operating guidelines, safeguarding commercial and industrial secrets, whenever there is a relevant and potential violation of fundamental rights.
The Bill, under the approved terms, also determines that governments shall act in sectorial perspectives when regulating AI systems. Such activity must occur by the appropriate agency or entity, considering the context and regulatory framework of each sector or industry, and always analyzing its regulatory impact.
In early versions of the Bill, there was a provision that allowed Federal, State and City public power to request impact reports, but the current proposal was approved without such mechanism. To clarify, impact reports were a set of documents that would contain the life cycle of the Artificial Intelligence system in question, as well as measures, safeguards and risk-mitigation strategies related to each phase of the system, including security and privacy. It is still not clear if the Senate will reintroduce this matter.
Nevertheless, the approved Bill establishes that sectorial agencies and entities must monitor the risk management of AI-based systems in concrete situations, evaluating the application contingencies and measurement procedures in their respective area of practice. In addition, it also determines that because ‘artificial intelligence’ systems are technological representations developed in the field of data processing and computer science, they shall only be regulated by federal norms.
During the voting process in the House of Representatives, two amendments were proposed (both rejected): the need of an impact report for high-risk AI activities and the implementation of diversity actions by the public administration.
It is also noteworthy that Bills n. 4120/2020, (about the use of algorithms by digital platforms on the internet), 240/2020 (same matter), and 872/2021, that is currently running before the Senate, were all detached from the approved version of Bill n. 21/2020. Now they shall begin their own procedures.
Bill n. 21/2020 was sent to the Senate and further developments are expected in the following months.