ChatGPT drafts law to regulate AI in Costa Rica, experts divided

The members of Congress told the chatbot to “think like a lawyer” and draft a bill according to the Constitution. They then sent the resulting text verbatim to the legislature.

“We have had many positive reactions and many people who thought it to be very risky,” Congresswoman Vanessa Castro, who led the introduction of the bill, told the Thomson Reuters Foundation.

ChatGPT recommended Costa Rica create an institution to regulate AI systems governed by the principles of accountability, explainability, bias prevention and protecting human rights.

The bill was introduced in May, but is now being discussed in public forums before it goes to the parliamentary commission for amendments and further debate in Congress.

“We learned that artificial intelligence is just another legislative tool that still needs the human hand,” Castro said.

Discover the stories of your interest

Costa Rica is the eighth country in Latin America to discuss or approve a law regulating AI in the past year. Costa Rican Congresswoman Johana Obando said she supported AI regulation, but opposed the bill because ChatGPT simply made up statistics and articles from the Costa Rican constitution.

But her main objection was that she said the bill was a mere “list of good wishes” without much bite – a common concern about AI legislation being discussed across Latin America.

ChatGPT said “we should regulate based on fundamental rights and international conventions,” Obando said. “But what are those rights and conventions? The bill does not mention them.”

Latin American lawmakers are pushing for regulation motivated by the European Union’s AI Act, which includes rules banning the use of the technology in biometric surveillance and for it to be clear what content is AI-generated.

In Mexico, a bill introduced in March encourages the creation of an ethical framework for the development of AI, based on the protection of human rights and personal data.

The ethical framework itself, though, was not described in the draft, which is now also being discussed in public forums with experts and lawmakers.

In June, the Peruvian Congress approved the first law in the region to regulate AI, which only awaits the president’s signature to come into effect.

The law designates a national authority to supervise the development of AI, based on the principles of digital security and ethics.

Peruvian Congressman Jose Cueto, who led the introduction of the bill, said the legislation was only a small part of a much-needed national strategy for cybersecurity and data protection.

“The heart of the law is … to create an environment in which we can make an ethical, transparent and sustainable use of AI,” said Cueto, a cybersecurity expert and former admiral.

Racism and discrimination

Brazil has been engaged in an intense debate about AI regulation for the past four years, with three bills pending in its Congress.

One AI legal framework, approved by the House of Representatives in 2021, but blocked by the Senate, focused mostly on principles and lacked enforcement mechanisms, said Tarzicio Silva, a researcher on algorithmic bias and fellow at the Mozilla Foundation, a U.S. non-profit organisation.

This bill led to the creation of a Senate commission that published a 900-page report proposing a risk-based regulation, in which AI systems that may harm people or target marginalised populations are considered “extremely risky” and are banned.

For example, the report recommends banning the use of AI systems for social scoring, which conditions access to public services based on a score people receive for their behaviour.

Anti-racism advocates like Silva, however, are concerned that the debate has excluded the points of view of minorities.

“This commission was comprised of 18 jurists, 80 experts, and not a single one of them were part of the racial minorities in Brazil,” said Silva. “They didn’t consider Black and Indigenous people.”

AI regulation is a hot issue in Brazil as a new bill – based on the three pending in Congress – was introduced to the Senate in May and will soon be discussed by a parliamentary commission.

For Silva, an important element of the debate is preventing the use of facial recognition systems that could enable the disproportionate arrest of marginalised populations, and automated hiring systems that discriminate against racial minorities.

“We are advocating for the right to review algorithmic decisions and discussing what could be the rights to reparation,” said Silva, referring to possible compensation for people harmed by AI systems.

Lawmakers in the region agree that fighting bias and discrimination in AI systems should be at the heart of new regulations, but much of the proposed legislation is vague about how to prevent, investigate and penalise it.

AI colonialism

A common element in the regional discussions of AI regulation is the need to create an environment that promotes local experimentation, allowing competition with multinational corporations like Microsoft and Google.

In Brazil, one of the bills proposes governmental authorisation to create a “regulatory sandbox” – a framework that allows local businesses to experiment with AI technology in a controlled environment.

“We are currently colonised by products of a few American multinationals,” said Francisco Garijo, president of the Ibero-American Society of Artificial Intelligence, which brings together experts from around the region.

“The best way to deal with this colonialism is to promote the development of local products that can compete with them,” he said.

AI systems also need to be created specifically by and for Latin Americans, taking into account their languages and cultures, said experts who gathered in March at a regional AI summit in the Uruguayan capital.

“In the creation of AI technologies for the region (it is necessary to) value their (Latin American) participation in research and development, and not only as mere producers of raw data and manual annotation,” said experts who signed the Montevideo Declaration on Artificial Intelligence.