European Parliament backs amendments to AI bill

Two committees of the European Parliament backed amendments to the AI bill, which would ban a wide range of uses of the technology. Reuters reports.

These include restrictions on facial recognition programs in public spaces and predictive policing algorithms.

The amendments also set transparency requirements for creators of generative models. They will be obliged to mark content as AI-generated, and to disclose information about materials used for training.

“This vote is an important milestone in regulating artificial intelligence. AI should serve people, society and the environment, not the other way around,” said European Parliament member from the Greens Kim van Sparrentak.

In addition, the law requires the creation of a publicly accessible database of high-risk AI systems used by public authorities.

Taken as a whole, the document covers many AI use cases, including facial recognition, generative algorithms and autonomous vehicles. All systems will be divided into three categories: low risk, high risk and unacceptable. The first will escape regulation, the second will be subject to certain rules, and the third will be prohibited.

As an example of “unacceptable” AI, MEPs cited social scoring systems or the use of the technology to manipulate public opinion.

High-risk algorithms include autonomous vehicles, medical devices and other technologies where a failure could endanger a life.

Companies using AI will be required to assess risks in terms of “health, respect for fundamental rights, democracy and the environment”.

The proposal bans:

• Real-time facial recognition in public spaces;
• Biometric data collection for training identification systems from cameras and social networks;
• Use of emotion recognition technologies by police, employers or educational institutions;
• Exploitation of AI systems that categorize users by demographic attributes;
• Use of crime-prediction systems.

In total, lawmakers considered around 3,000 amendments. The document will now be put to a vote in the European Parliament.

After the law comes into force, market participants will have about two years to bring their activities into line with the new rules.

Violators face a penalty of 6% of annual turnover or €30 million.

Earlier in April, a group of European lawmakers announced plans to develop a draft of rules to govern the development of AI systems.