MEPs on the European Parliament’s Committee on the Internal Market and Consumer Protection (IMCO) and Civil Liberties, Justice and Home Affairs Committee (LIBE) on Thursday 11 May adopted the report by Dragoș Tudorache (Renew Europe, Romanian) and Brando Benifei (S&D, Italian) on the artificial intelligence act (‘AI Act’) (84 votes in favour, 7 against, 12 abstentions) (see EUROPE 13179/1).
“Given the profound impact that AI will have on our societies and economies, the AI Act is very likely the most important piece of legislation of this term” commented Tudorache after the vote.
The issue was debated at length during the negotiations, and in the final version of the text, MEPs supported the extension of the classification of high-risk sectors to health, security, the environment and, as requested by the S&D group, fundamental rights.
The providers of these AI systems should, moreover, carry out an impact assessment on them.
“It is essential to build citizens’ confidence in the development of AI, to define the European way forward to deal with the extraordinary changes that are already taking place as well as to shape the policy debate on AI at global level. We are convinced that our text strikes a balance between the protection of fundamental rights and the need to provide legal certainty for businesses and to stimulate innovation in Europe”, said Mr Benifei.
In addition, AI systems that may influence voters in political campaigns fall under the scope of the text, as do recommendation systems used by very large online platforms, as defined under the Digital Services Act (DSA) (see EUROPE 13176/33). This provision for platforms with more than 45 million users per month had been added during the last weeks of negotiations between the political groups.
Strengthened provisions for generative AI systems
The final version of the text also confirms the work done by MEPs in recent weeks on general purpose AI systems. Providers of foundation models, which can be adapted to a wide range of tasks and purposes, should document their work, increase transparency and assess risks in order to limit them. In case of non-compliance, foundation model providers could be fined up to €10 million or 2% of the annual worldwide turnover generated.
Generative AI systems, such as ChatGPT, would be subject to additional requirements. Providers should therefore inform users that content has been generated by AI and ensure, when designing the system, that it could not generate illegal content. The summary of the copyrighted data used to train the generative AI system should also be published.
EPP outvoted on biometric identification
Following the logic of ranking the risk of an AI system, MEPs drew up a list of ‘unacceptable’ uses. In the version of the text adopted in the European Parliament committees, they completed the list.
In addition to systems that exploit people’s vulnerabilities or those used for social ‘scoring’ purposes, MEPs also added real-time remote biometric identification systems in publicly accessible spaces. Its use after the fact would also be prohibited, except in cases of serious crime and after judicial authorisation.
The EPP group opposed the ban on biometric identification. A separate vote was held on this issue. The rapporteurs had told EUROPE that the outcome of this vote could be close. However, the confirmation of the ban on biometric identification received a more comfortable majority than expected (58 votes in favour, 10 against, 36 abstentions).
“I believe that the banning of real-time facial recognition systems, social ‘scoring’ evaluation systems or the establishment of clear rules for technology to monitor students during online exams are the main outcomes of the compromise negotiated on AI legislation. Under the new rules, artificial intelligence systems will not be able to discriminate against people or violate fundamental human rights”, was the reaction of Marcel Kolaja (Greens/EFA, Czech).
Plenary vote planned for June
MEPs also decided on the role of the European AI Office, which will be responsible for monitoring the implementation of the AI regulation. Finally, the report also encourages the use of ‘sand boxes’, which should allow suppliers to test their AI system for a defined period of time without necessarily having to comply with the regulatory framework.
Only The Left group did not support the text in the final vote in the European Parliament committees. The Identity and Democracy and ECR groups were divided on the issue.
The report will be put to the vote in the European Parliament’s plenary session starting Monday 12 June. Inter-institutional negotiations can then begin, as the Council of the EU had already adopted its position on 18 November 2022 (see EUROPE 13067/11). The aim will be to reach an agreement before the end of 2023.
“I am convinced that we can reach an agreement by the end of the year. [...] This must be done before the end of this mandate. We cannot afford to start from scratch in 2024” Dragoș Tudorache (see EUROPE 13179/2) explained to EUROPE. (Original version in French by Thomas Mangin)