Artificial Intelligence Regulation in Brazil

2023 Aug 05

Artificial Intelligence Regulation in Brazil

This week, the Data Hackers podcast had an episode that I found to be one of the best, which was about Artificial Intelligence Regulation in Brazil with the always excellent Professor Diogo Cortiz.

The discussion, along with Paulo Vasconcellos, Gabriel Lages, and Monique Femme, was extremely educational and informative. Far from the commonplaces of Twitter and hot takes, everyone showed very reasonable and well-founded opinions on the subject.

The Bill with the initial text of the regulation is No. 2338 of 2023, which “Provides for the use of Artificial Intelligence,” authored by Senator Rodrigo Pacheco (PSD/MG) Link.

Following Diogo’s suggestion in the episode, I decided to read the entire bill and make some notes for Machine Learning Articles, but I decided to put my general considerations about the bill in writing here on the blog for SEO purposes, but at some point, the video will be uploaded to the channel.

…

In my personal assessment, the bill overall has a score of 6.8/10 with a satisfactory and reasonable initial text.

Some of my notes on the bill:

The Good:

  • The bill is mostly centered on end-user/consumer applications, with practically no mention of systems based on industrial data;
  • The part about including humans in the loop and explainability shows that specialists managed to convey important aspects to legislators;
  • The bill takes into account the proportionality of regulation according to the economic scale of the agent;
  • It correctly establishes the regulatory sandbox;
  • It starts with more modern thinking, with a balance between government and market, and does not yield to Brazilian prohibitionism;

What is Not Clear:

  • It does not establish a differentiation between editorial and content provider platforms. I think something like the US Section 230 would be a great complementary law that could give more substance to the bill. Reinforcing what is editorial and what is content provision can provide more legal certainty for platforms;
  • The part on jurisdictional conflicts is not very clear (e.g., Brazilian users accessing foreign services, or Brazilian agents using technology with intellectual property not subject to Brazilian jurisdiction);

What Can Be Improved:

  • It makes little reference to existing legislation such as the LGPD and the Brazilian Civil Framework of the Internet, often being redundant;
  • It relies on application via regulatory agencies instead of using the judiciary and its instruments as an avenue for mediation and definition;
  • It does not precisely establish the right to be forgotten like the European General Data Protection Regulation;
  • More powers for the executive branch and agencies, instead of mediation through the judiciary with legislative oversight;
  • The part on individual civil liability of AI operators is very poor and needs adjustments, as there should be joint liability among developers, project managers, and entrepreneurs, or a single responsibility mechanism as in engineering projects;
  • The copyright part could be clearer and will be a problem if the text is not refined. A robots-ai.txt should be included so that data has an opt-in mechanism by default, letting the content creator determine whether they want to participate in data aggregation or not.

My complete notes are available at the link below:

  • Link

  • Permalink: https://flavioclesio.com/assets/files/pl-2338-2023.pdf