Artificial Intelligence: What about the EU’s AI Act?

The regulation of artificial intelligence has been occupying the European Union for some time. This year, the debate on the “Artificial Intelligence Act” (AI Act) became a central digital policy topic after the EU Council Presidency presented its own proposals for amendments last year in November 2021.

In April 2021, the EU Commission had already published a far-reaching draft regulation on AI regulation. The aim of the Artificial Intelligence Act is to make  the EU internal market the global pioneer for the development of secure, trustworthy and innovative AI systems through binding rules. Among other things, the Commission’s draft  provides for a tiered system that divides AI applications into different risk classes with certain conditions.

“The higher the risk of a specific use of AI, the stricter the rules,” said EU Competition Commissioner Margrethe Vestager at the time.

What is the EU Commission planning for A.I.?

E.U. flag

The majority of existing AI applications, such as search algorithms, spam filters or video games, will remain largely unregulated due to their minimal or low risk. It should help to maintain the EU as an innovation-friendly economy. On the other hand, “high-risk AI systems” in the draft are systems that are used, for example, as security components in critical infrastructure. Or that are  intended to assess the suitability of people – for example for loans or jobs.

According to the draft regulation, the providers of such systems will have to subject them to a risk analysis in  the future. Or they shall develop their own risk management system for this purpose.  In particular, operators of critical infrastructures (Telecom,…) and providers of AI systems fear double regulation, in which instruments of IT security regulation and AI regulation overlap. It could lead to a significant increase in bureaucracy. What Europe and bureaucracy? No way…

Europa is not China

In addition, the Commission plans to ban AI applications that manipulate humans to their detriment. Or usage that carry out unprovoked surveillance. Facial recognition software is also to be largely banned. State agencies are prohibited from so-called social scoring, which is practiced in China, for example, against the population. According to the draft, anyone who violates these rules must expect a fine of up to 30 million euros or 6 percent of global annual sales – whichever is higher. Sounds like you should start to put money in Panama 🙂

Raising voices

These proposals have been criticized by individuals, companies and industry associations specializing in  AI systems, while civil society organizations are calling for even stricter rules. This line of conflict was partly reflected in the EU Parliament, which is why the discussion of the draft has progressed only slowly since its presentation in April. In the meantime, it has at least been possible to agree that further parliamentary work on the bill will take place under the joint leadership of the Committee on Civil Liberties, Justice and Home Affairs (LIBE) and the Committee on the Internal Market and Consumer Protection (IMCO). If you have some spare time, you can as well look at this contribution from Stanford University

The proposals of the EU Council for A.I.

The amendments proposed by the Slovenian Presidency at the end of  November 2021 also brought some movement into the matter. This is a compromise proposal that emerged from the negotiations between the governments of the EU states – i.e. also with the participation of the old Federal Government. For example, Member States want to exclude AI systems developed exclusively for military or scientific purposes from the scope of the AI Act. Sounds like a wild card for Zone 51.

Other key points concern a stricter approach to the legal approval of biometric recording systems. The extension of the social scoring will be banned from public authorities to private institutions. And AI systems that are used to calculate insurance premiums are now to be among the high-risk systems. In addition, the EU Council wants to oblige the Commission to review the list of high-risk systems and the list of AI technologies covered by the AI Act every two years. Surprise each time.

Much still needs to be clarified on the A.I. act

The Presidency’s compromise proposal was accompanied by a progress report by the EU Council on the AI Regulation. Member States warn that the requirements for high-risk systems are still too vague and  require practical guidance for companies. This is where usual certification actors should act. EU member states have also  warned of an excessive administrative burden for small and medium-sized enterprises as a result of the regulation. In general, it is considered a central challenge of the planned regulation  to create a balance between safe products and clear legal remedies.  All this without compromising the desired and required innovation in the field of artificial intelligence. Europe does not want to lose each innovation wars…

Social scoring in A.I.

In addition, the extension of the ban on social scoring to private companies and institutions desired by the EU Council also raises questions. On the one hand, this change could have far-reaching implications for the financial sector, where interest rates on loans are currently calculated on the basis of repayment probabilities. On the other hand, it is still unclear how this ban will be effectively examined in the commercial sector.

What is IT and what is A.I.?

Overall, the AI regulation is a rather complex matter, which, in addition to the political debate on individual issues, also contributes to the sluggish progress of legislation. Finally, the AI Act interacts with various other EU laws, from data protection and law enforcement to product safety and IT security and other sectoral legislation. It makes it difficult for policymakers to foresee the impact of all the provisions. Let alone for business…

Perspective

It is therefore difficult to estimate how long it will take for MEPs and parliament and EU states to reach an agreement on a finished law. However, companies that develop or use AI systems should therefore not sit back, but better prepare for implementation. Because sometimes legislation in the EU goes faster than you think.

 

Leave a Comment