Better regulate conversational applications such as ChatGPT: MEPs will vote on Thursday on the future European Union regulation on artificial intelligence (AI), an ambitious text delayed by the controversies surrounding these technologies. The European Union wants to be the first in the world to have a comprehensive legal framework to limit the excesses of AI, while securing innovation.
Of great technical complexity, artificial intelligence systems fascinate as much as they worry. Dissemination of dangerous content, manipulation of opinion by creating false images with software such as Midjourney, mass surveillance systems, or still unknown risks... Tech personalities, such as Twitter boss Elon Musk, have even called for a moratorium on their development.
The eruption of generative intelligences slows down debates
The general public discovered their immense potential at the end of last year with the release of the editorial content generator ChatGPT from the Californian company OpenAI, which can write essays, poems or translations in seconds. In response to these rapid upheavals, the European Commission proposed draft legislation two years ago that is dragging on.
EU member states have only defined their position by the end of 2022. The MEPs, very divided, will endorse theirs on Thursday in a committee vote in Strasbourg that is expected to be confirmed in plenary in June. Then began a difficult negotiation between the various institutions. European Commission Vice-President Margrethe Vestager called on Monday not to waste time. "I really hope we can finish this year," she said. The delay is partly explained by the eruption in the public debate of so-called general-purpose artificial intelligences (capable of performing a wide variety of tasks), including generative AIs such as ChatGPT.
"It's a complex text and we have added a new regime of rules dedicated to generative AI," co-rapporteur Dragos Tudorache told AFP. MEPs want to force providers to put in place protections against illegal content and reveal copyrighted data used to develop their algorithms. But experts consider that the risks of generative AIs do not require separate treatment. "I don't see Parliament's motivation. I do not see how these risks are different from what has already been anticipated," Pierre Larouche, an expert in digital law at the University of Montreal and researcher at the Center on Regulation in Europe (Cerre), told AFP.
The Commission's proposal, unveiled in April 2021, already provides for a framework for AI systems that interact with humans. It will require them to inform the user that they are in contact with a machine and will force applications generating images to specify that they were created artificially. The text is inspired by existing European product safety regulations and will impose controls based primarily on companies. Bans will be rare. They will concern applications contrary to European values such as citizen rating or mass surveillance systems used in China.
A list of rules imposed on applications deemed "high risk"
MEPs want to add a ban on emotion recognition systems and remove derogations allowing remote biometric identification of people in public places by law enforcement. They also intend to ban the mass collection of photos on the Internet to train algorithms without the consent of the people concerned.
The heart of the project consists of a list of rules imposed only on applications that will be deemed "high risk" by the companies themselves based on the criteria of the legislator. The European Commission has proposed that they concern systems used in sensitive areas such as critical infrastructure, education, human resources, policing or migration management... Among the obligations: human control over the machine, establishment of technical documentation, or implementation of a risk management system... Compliance will be monitored by designated supervisory authorities in each member country. MEPs want stricter criteria to restrict the "high risk" classification to only products that may threaten safety, health or fundamental rights.
- Artificial Intelligence (AI)
- European Union (EU)