The rush to regulate artificial intelligence could hamper advances

The idea of ​​establishing a law for artificial intelligence (AI) in Brazil for the 2024 elections, raised by members of the Legislature and Judiciary in recent weeks, is hasty and, in some cases, is arising from ignorance and politicization of the subject. . Experts consulted by People’s Gazette They fear that this will harm a technical debate on the topic.

For them, the discussion needs to be guided by a deeper knowledge of the new tools, and the use of artificial intelligence has not even matured enough for rules to be formalized for the field. The rush, according to experts, could hinder the arrival of technological advances in Brazil and generate a law destined to quickly become obsolete.

In recent weeks, the subject has become an obsession for some parliamentarians and members of the Judiciary. On January 19, the president of the Senate, Rodrigo Pacheco (PSD-MG), stated at an event organized in Zurich, Switzerland, by the Lide Group, that he wants artificial intelligence to be regulated by April this year.

The rush for Congress to regulate the issue was also expressed in December by Alexandre de Moraes, president of the Superior Electoral Court (TSE). Both Moraes and Pacheco claim that regulating AIs is necessary to ensure the integrity of this year’s municipal elections.

In speeches by parliamentarians and magistrates, the main concern from an electoral point of view has been the deepfakesthat is, videos and audios created with artificial intelligence, simulating with great verisimilitude the characteristics of a person, such as their physiognomy, mannerisms or voice.

You deepfakes They were, for example, the main target of a draft released on the 25th by the TSE to establish rules related to AIs for the 2024 elections and of a hearing with representatives of civil society, the academic community, companies and political parties. The court signaled that it wants to force users to “explicitly report the use of fabricated or manipulated content in any form of electoral propaganda”.

In the Senate, PL 2,338/2023, authored by Pacheco himself, on the use of artificial intelligence is being processed. This is a proposal prepared by a commission of 18 jurists, with 45 articles. Pacheco’s desire is to approve the law in the first half of this year, so that it can be applied during the municipal election.

Rushing into broad, generic regulation on artificial intelligence is a mistake, experts say

For lawyer Rodrigo Ferreira, a researcher in artificial intelligence, it is hasty to think about general legislation on AI in Brazil, especially taking into account the variety of phenomena that fall under the umbrella of artificial intelligence. He criticizes, in the Senate’s initiative, the lack of technical experts in the debates.

“The project was the result of a commission of jurists. What we observe is that there has been a lot of debate mainly among professionals in the area of ​​social sciences and humanities, with many jurists involved and some economists, in addition to the people who deal more with the of human rights activism. But we see few technicians at the table, few people who actually have experience in the development of artificial intelligence systems participating in the debate”, he says.

Ferreira explains that AI cannot be considered a single technology; Therefore, a generic regulation on the subject, such as the one being processed in the Senate, is not the best solution. Even in the European Union, where the debate for generic regulation has been ongoing for much longer, there is still strong controversy about the need for a legal framework of this type.

“Artificial intelligence is a huge set of general-purpose technologies. There are components like self-driving car, medical diagnostic device, facial recognition camera… There are a myriad of applications that are different technologies under a gigantic umbrella called ‘intelligence’. artificial’. The idea that the debate is already ripe for generic regulation of technology or a deeper understanding of risks in all segments is not sustainable”, he observes.

For Paulo Lino, member of the Council of the International Association of Chiefs of Police for training in Artificial Intelligence and Cybercrime, the idea of ​​rushing to regulate artificial intelligence could end up resulting in a law with an expiration date and a restriction of technological advances.

“It is not the right time. We are not mature. The evolution of artificial intelligence is so fast, so dynamic and versatile in its applications, that, if we create a law today, in two days it could be obsolete and no longer applicable “, it says. “The law is watertight. In a very short time, things will be completely different, and we cannot make new regulations every day”, he adds.

Experts prefer more flexible solutions

For Lino, the ideal way to avoid abuses in the application of artificial intelligence would be to attribute the role of regulation to an autarchy, as a regulatory agency, “ideologically independent and flexible”. “Nothing like anything we have today,” she says.

For now, proposals for regulating AI in Brazil are based mainly on European discussions, where members of the European Parliament (MEPs), the Council of the European Union and the European Commission released in December a provisional agreement for the regulation of artificial intelligence on the continent. .

One of the focuses of the European text is to prevent large technology companies from using sensitive information to influence people’s behaviors and manipulate the way they think or act. The document talks, for example, about curbing the use of AI to categorize people based on personal characteristics or to create facial recognition databases without a specific objective.

Rodrigo Ferreira assesses that creating rigid and generic legislation on artificial intelligence, as Europe is trying to do, is not the best path for Brazil in the current context. “In my reading, Brazil will gain a lot if, instead of mirroring the European model, it looks at the model of the United States and other countries that are opting for a sectoral rather than centralized approach,” he says.

“Instead of establishing a standard intending to regulate artificial intelligence – and treating an autonomous car, a medical device, facial recognition technology in the same way – there would be a sectoral regulator to establish specific standards for each case. If I have different risks For medical devices, I call a sectoral regulator – Anvisa, for example – to issue rules on the approval or not of a certain type of device”, adds the specialist.

For Ferreira, sectoral regulators would have an easier time updating regulations as technology advances and the risks related to it change. “Imagining that a parliamentarian who doesn’t have the slightest familiarity with the subject will produce a law in this regard, and that this law will remain current and relevant in three, four, five years, is a dream. That won’t happen, especially because technology advances very quickly.”

The Argentine experience aggravated the fear of deepfakes

One of the reasons why Brazilian parliamentarians and members of the Judiciary have wanted to accelerate the discussion on artificial intelligence is the phenomenon of deepfakesespecially after the episode in the Argentine elections in which the candidate Sergio Massa, a Kirchnerist candidate competing with Javier Milei, was falsely portrayed using cocaine using artificial intelligence.

Experts do not underestimate the concern about the phenomenon of deepfakes and its potential for creating false information and manipulating electoral preferences, but they do not think this is a justification for approving generalist legislation on AIs in Brazil this year.

Paulo Lino also considers the excessive focus on the independent user as the main concern for the electoral period related to AIs to be mistaken. For him, there are much more relevant problems that have not received the same attention. “If we really wanted to protect voters and provide more fairness and transparency to the electoral process, we would take other steps, such as controlling access to data that technology companies have on voters. They practically manage public opinion,” he says.

To exemplify the scale of other problems related to AIs, Lino reports that, in the United States, government agencies used artificial intelligence software capable of automatically crossing information from different networks with the aim of locating potential perpetrators of crimes such as pedophilia or shootings in pasta. “If this is misused, I can start locating, for example, ideological enemies. And that cannot be allowed,” he says.

Because of this dilemma, the tool was banned in the USA, despite its proven usefulness for the police. For Lino, this is an example of how hasty legislation on artificial intelligence could end up treating very complex issues from an ethical point of view lightly.

The government’s objective, according to him, should be to regulate abuse, not monitor the use of artificial intelligence. “It’s a pretty big difference. When we create legislation, we have to be very careful to maintain a technical and consumer protection bias, and not an ideological bias. The objective cannot be to monitor. The objective must be to prevent abuse. And These abuses are already happening today, and they don’t come mainly from independent developers. They are happening, today, through the actions of large companies”, he states.