EU makes landmark decision on Artificial Intelligence agreement
After three days of intense discussions between member states and the European Parliament, the European Union has concluded an unparalleled global agreement known as the AI Act, which regulates the development of artificial intelligence.
"We've reached an agreement!" European Union Commissioner for Internal Market and Services, Thierry Breton, announced via the X service, formerly known as Twitter. This is a landmark decision that positions the European Union as the first continent globally to establish specific guidelines for using artificial intelligence. Breton stressed that it is a "political agreement" designed to stimulate the innovation of advanced technologies in Europe while limiting potential misuse.
The AFP agency underlines the considerable influence of the ChatGPT system, an American text generator developed by OpenAI, in initiating these negotiations. This system, which can produce essays, poems, and translations in just seconds, has provided a glimpse into the vast capacity of artificial intelligence while also shedding light on possible risks, such as the circulation of fake images on social media platforms.
The phenomenon of generative artificial intelligence was a key consideration during the negotiations. European Parliament members highlighted the necessity for stringent governance over this technology. They specifically urged for greater transparency about algorithms and vast databases.
Before the start of the negotiations, EU member states expressed concern that overly restrictive regulations could impede the progress of promising leaders in the industry, such as Aleph Alpha in Germany and Mistral AI in France.
The AFP reports that the agreement reached on the preceding Friday evening will now be strengthened with technical work targeting the finalization of the text for the AI Act.
The compromise accounts for two areas: the continuing development of artificial intelligence and the imposition of constraints to prevent abuse. Regulations ensuring the quality of data used in algorithm creation and adherence to copyright laws will apply to both aspects. Producers will also have to verify that generated sounds, images, and texts are signposted.
The central component of the act lists the only rules applicable to systems deemed "high risk." These primarily encompass systems used in sensitive sectors such as critical infrastructure, education, human resources, and law enforcement. These systems will be obligated to meet several requirements, such as maintaining human control over the machine, presenting technical documentation, and incorporating a risk management system.
The regulations encompass a provision for special supervision over artificial intelligence systems that interface with humans. These systems must inform users that they are interacting with a machine.
The AI Act also anticipates the enforcement of prohibitions, mainly concerning applications that contravene European values. China, for instance, with its citizen assessment systems, mass surveillance, and remote biometric identification of individuals in public places. However, EU nations will maintain some latitude regarding certain law enforcement actions, such as counterterrorism measures.
The agreement further stipulates that European legislation will implement control measures and penalties. A European Office for Artificial Intelligence is set to be established within the European Commission. Monetary fines for severe infringements could be up to 7% of turnover, with a minimum figure of 35 million euros (approximately $39 million).
