The European Parliament has voted to confirm its negotiating mandate for the AI Law, reaching an important milestone that unlocks the next stage of negotiations towards a pan-European rulebook for artificial intelligence.
MPs backed a modified version of the Commission proposal that expands the rulebook in a way they say aims to ensure that AI being developed and used in Europe is "fully in line with EU rights and values." , including human supervision, security, privacy and transparency, non-discrimination, social and environmental well-being”.
Among the changes that MEPs have backed is a complete ban on remote biometric surveillance and predictive policing. They also added a ban on "undirected capture of facial images from the Internet or CCTV images to create facial recognition databases", so it's basically a hard ban on Clearview AI and the like.
The proposed ban on remote biometric surveillance would apply to both real-time or subsequent (after-the-fact) applications of technologies such as facial recognition, except, in the latter case, for law enforcement for the prosecution of serious crimes. with legal suspension.
MEPs also added a ban on the use of emotion recognition technology used by law enforcement, border agencies, workplaces and educational institutions.
Parliamentarians also expanded the classification of high-risk AI systems to include those that pose significant harm to people's health, safety, fundamental rights or the environment, as well as AI systems used to influence voters and the election results.
MEPs also added larger social media platforms that use algorithms to recommend content to the high-risk list.
The plenary vote follows the committee's endorsement of the amended proposal last month after MEPs from different political groups discussed how they wanted to amend the Commission text, including adding obligations on makers of so-called general-purpose AI.
In response to rapid developments in generative AI, MEPs have supported imposing a set of obligations on basic/general-purpose AI models, such as the technology behind OpenAI's AI chatbot ChatGPT, which requires such systems identify and mitigate risks before placing them on the market, as well as apply transparency disclosures to AI-generated content and implement safeguards against the generation of illegal content.
General-purpose AI makers are also required to publish "detailed summaries" of the copyrighted information used to train their models as proposed by MEPs.
During a tour of European capitals to meet lawmakers last month, OpenAI CEO Sam Altman criticized this aspect of the EU proposal. He suggested the company might have to withdraw service in the region if it couldn't comply, telling reporters he was hopeful the obligations would be reversed.
In any case, today's plenary vote shows overwhelming support among MPs for the amended version of the bill, including the proposed obligations for general-purpose AIs, with 499 votes in favor and just 28 against (plus 93 abstentions). ).
The vote to approve the mandate means that discussions between the parliament and the governments of the EU member states can now begin, with the first tripartite dialogue scheduled for tonight.
Commenting in a statement after the vote, co-rapporteur Brando Benifei said:
All eyes are on us today. As big tech companies sound the alarm about their own creations, Europe has stepped forward and proposed a concrete response to the risks that AI is beginning to pose. We want to see the positive potential of AI for creativity and productivity harnessed, but we will also fight to protect our position and counter the dangers to our democracies and freedoms during negotiations with the Council.
In another supporting statement, co-rapporteur Dragos Tudorache added:
The AI Act will set the standard worldwide in the development and governance of artificial intelligence, ensuring that this technology, destined to radically transform our societies through the enormous benefits it can offer, evolves and is used in accordance with European values of democracy, fundamental rights and the rule of law.
The version of the AI Law that MEPs have endorsed today also adds exemptions for research activities and AI components provided under open source licences, which MEPs suggest will ensure support for innovation, along with the regulatory sandboxes for test systems to be established in the framework.
The MEPs' proposal also adds a set of consumer rights over AI decision-making, including the ability for consumers to seek collective redress if an AI system has caused them harm.
The European consumer organization, BEUC, welcomed these changes, but criticized the parliament for not backing a full ban on the use of emotional recognition AI (as the proposal does not limit commercial use on that slippery slope).
He also believes that MEPs have given developers too much discretion to decide whether or not their systems fall into the high-risk category, which he says could undermine the effectiveness of the risk-based framework.
That can be a source of contention during tripartite discussions that need to find a compromise between the position of the Council of the EU, which is the body made up of the governments of the member states, and the legislators in parliament to achieve the necessary political agreement on a final text and seal the file.
In general, the Council of the EU takes a more pro-industry line, while the parliament tends to be more concerned with fundamental rights. It remains to be seen where the two parties will meet to regulate the AI.
If they cannot agree, the EU law-making process may stall, or even fail. But there is a push in Brussels to get past this file given the amount of global attention now focused on AI regulation. (Being first to strike with a democratic rulebook for AI presents opportunities for the bloc to exert influence beyond its borders as other jurisdictions scramble to figure out their own approaches to regulating a complex field of fast-developing technology. ).
The Council adopted a position on the previous file in December. At the time, Member States largely favored deferring what to do with general purpose AI, to additional implementing legislation. But, given what has happened in the meantime, with generative AI tools like ChatGPT shooting into the center of the technology discussion and raising multiple red flags to regulation (and many tech industry typologies), it will be interesting to see what states will agree on. members and MEPs on the need to add obligations for this class of AI systems to the text of the Law on AI itself.
The original proposal for the risk-based framework for AI was submitted by the EU executive in April 2021. While that first draft text from the Commission did not address the issue of general-purpose AI as broadly, it did propose provisions transparency for chatbots and technology deepfake. Already at that time, EU legislators were of the opinion that consumers should be informed that they are interacting with machine-generated content.
While the Commission remains hopeful that the tripartite talks on the AI Law shelving will provide a political agreement by the end of this year, there will still be an implementation period, so the legislation is likely not to be implemented before 2026.
That's why the EU is also working on several voluntary initiatives that aim to put pressure on AI companies to self-regulate security in the meantime.
