Emerging as a pacesetter in artificial intelligence (AI) regulation, the European Commission published on 21 April 2021 its much-anticipated proposal for a Regulation on a European approach for AI that focuses on building trust and managing the opportunities and threats of AI. As part of the package, it also published a proposal to replace the current Machinery Directive with a Machinery Regulation to capture the new features of machinery products.
As the first legal framework of its kind, the proposed AI regulation would have far-reaching implications for tech companies. AI systems will be limited from entering the European Union if they do not meet the set standards and breaking the rules could lead to fines of up to 6% of global turnover. The proposal covers a lot of ground, from AI application in robot-assisted surgery to state use of facial recognition.
The draft regulation follows a risk-based approach: the higher the risk, the stricter the rule. There would be no restrictions if the risk is minimal and only specific transparency requirements if the risks are limited. However, high-risk AI systems will be subject to strict obligations before they can be put on the market such as setting up a risk management system, high-quality data sets and appropriate human oversight. The rules currently list eight categories of high-risk AI systems including for example self-driving cars, recruitment algorithms or systems that evaluate credit worthiness.
Taking a stance to protect fundamental rights, the Commission also proposes to ban AI systems considered a clear threat to people’s safety, livelihoods and rights. This includes AI systems or applications that manipulate human behaviour to circumvent users’ free will (e.g. toys using voice assistance encouraging dangerous behaviour of minors) and systems that allow social scoring by governments. Real-time remote biometric identification systems in public spaces like facial recognition would be in principle prohibited. However, the devil is in the details as several exemptions have been carved out.
Now the European Parliament and the Member States will consider the Commission’s proposal then try to find an agreement on a final text. The text is likely to change in the process. The Parliament has already been very proactive on the topic with recommendations on aspects such as ethics, intellectual property and civil liability for AI. Other resolutions are currently being prepared. In a letter to the Commission dated 8 March 2021, some Members of the European Parliament contested the limitation of mandatory rules to a narrow list of high-risk applications. Heated discussions are to be expected on what constitutes high risks, exceptions to list of banned AI applications and the degree of self-regulation for conformity assessments.
The legislation is anticipated to become law in 2023. One should also keep an eye out for the upcoming review of the General Product Safety Directive, the Radio Equipment Directive and the New Legislative Framework that will aim to address a few gaps in the current product safety regulation.
Please find below an interactive timeline for a snapshot of the key milestones for AI legislation.
Any questions? Contact our tech team at firstname.lastname@example.org.