What Latin America can learn from the EU's artificial intelligence legislation

The EU’s artificial intelligence regulation represents the first attempt worldwide to legislate AI and prevent the technology from colliding with fundamental rights and, like the personal data legislation previously enacted, could hold valuable lessons for Latin America.

“The first thing to highlight about European AI regulation has to do with the fact that artificial intelligence can and should be regulated,” Beatriz Busaniche, president of Fundación Vía Libre, an Argentine digital rights NGO, told BNamericas.

The European law aims to protect fundamental rights, democracy, the rule of law and environmental sustainability, while harnessing AI’s innovation potential.

Latin America has some soft law instruments that provide a regulatory AI framework, such as the OECD recommendations, the UNESCO ethics document and the Montevideo and Santiago declarations. In addition, bills to regulate AI are advancing in different stages in Argentina, Brazil, Chile, Colombia, Costa Rica, Peru, Mexico and Uruguay.

However, a proper AI framework has not yet been approved and many countries are lacking personal data protection or copyright regulations.

“Latin America must now begin to develop its own vision of what it wants and what it does not want from AI. Regulation should reflect the realities, opportunities and needs of our citizens, without leaving aside that AI has no borders and global cooperation is essential,” Daniel Rodríguez Maffioli, technology and policy director at law firm Ecija, told BNamericas.

Busaniche criticized the absence of Latin American representatives in the international debate, which is aggravated by a defunding of the scientific and academic systems by regional governments, mainly in Argentina.

One option is to strengthen the application of existing laws or adapt them to the new AI reality. “Appealing to the rules and human rights already recognized seems appropriate, although certainly insufficient,” Maffioli said.

Agreeing on the need to enforce regulations, Busaniche said, “Latin America has to strengthen its capabilities and try to have stronger regulations.”

She highlighted that, for example, Argentina has an enforcement authority protection law that has traditionally been weak. “We must do institutional work of investment and capacity building.”

EUROPEAN LAW

The EU regulation establishes a series of obligations for AI, based on potential risks and impact.

Risk-based regulation avoids putting a brake on innovation by including very restrictive rules for applications that would not, a priori, have a significant impact. But this approach could also have a downside.

“Some sectors have criticized that risk-based regulation presupposes that human rights are applicable only in certain high-risk cases,” said Maffioli.

He added the European law "is extremely extensive, complex to apply in practice, and very onerous for innovators and entrepreneurs."

European law provides clear obligations for high-risk AI systems such as critical infrastructure, vocational education and training, employment, essential public and private services, certain law enforcement systems, migration and customs management, justice and democratic processes. These AI systems must assess and reduce risks, maintain usage records, be transparent and accurate, and have human oversight.

In addition, AI applications that violate the rights of citizens, such as biometric categorization systems based on sensitive characteristics and the indiscriminate capture of facial images from the internet or recordings from surveillance cameras to create facial recognition databases, are prohibited. Emotion recognition in the workplace and schools, citizen rating systems and predictive policing will also be banned under certain parameters.

However, there are some exceptions that would allow security forces to use biometric identification.

“The supposedly prohibited uses actually contain very broad exceptions that would allow public authorities to easily evade such a prohibition. This could compromise the freedom, privacy and dignity of citizens,” said Maffioli.

“They went from a strict ban on facial recognition systems in public spaces to having some spaces for the use of this type of technology,” Busaniche said. “Some differentiated uses are also [allowed] that leave migrant populations, for example, refugee populations, particularly unprotected,” she added.

Busaniche also highlighted that some previous European legislation, some of which protects people from automated decision-making, “sounds powerful on paper but most companies have found a way to circumvent it by putting a person in the middle to continue giving priority to the automated decision.”

European law also raises transparency requirements for general-purpose AI systems and models on which these systems are based. For example, compliance with copyright laws and the need to make public detailed summaries of the content used to train models are foreseen.

More powerful models that could pose systemic risks will need to meet additional requirements, such as performing model assessments, analyzing and mitigating risks, and reporting incidents.

Additionally, artificial or manipulated images, audio or video content must be clearly labeled.