MEPs held their first political debate on the artificial intelligence (AI) regulation on Wednesday (5 October). The discussion progressed to sensitive topics such as the highly controversial issue of biometric recognition.
The AI Regulation is landmark European legislation intended to regulate artificial intelligence by introducing a series of obligations proportionate to the potential harm of applications of these technologies.
So far, the European Parliament’s co-rapporteurs, the social democrat Brando Benifei and the liberal Dragoș Tudorache, have confined the discussion to the more technical aspects, hoping to create momentum before tackling the more political difficulties.
This approach was not without success since the file progressed in several parts. During the meeting, MEPs formally accepted the first two sets of compromises on administrative procedures, conformity assessment, standards and certificates.
Some important parts of the third set of compromises on obligations for high-risk systems were also approved.
However, most of Wednesday’s political meeting in Strasbourg was devoted to very sensitive topics such as the scope of the AI regulation and restrictions on the use of biometric recognition, a highly controversial technology. that identifies a person by face or other personal characteristics.
As expected, the most heated part of the debate focused on biometric recognition systems. Ahead of the meeting, the co-rapporteur shared a programme, consulted by EURACTIV, including a proposal to reformulate the article on prohibited practices.
In the initial draft, the European Commission proposed to ban subliminal techniques, the exploitation of vulnerabilities, social rating and real-time biometric identification systems.
However, this latest prohibition had some notable exceptions, including when it came to identifying kidnapping victims, warning of imminent threats such as terrorist attacks, and spotting criminal suspects.
Progressive MEPs and civil society organizations have harshly criticized this approach, saying such exceptions could lead to more widespread surveillance. The MEPs in charge of the file have now formulated this prohibition clearly, while leaving the other prohibited practices unchanged.
The article now prohibits “the placing on the market or making available on the market, putting into service or use of remote biometric identification systems which are or may be used in spaces accessible to the public or to the private sector, both offline than online. »
Notably, the text removes the reference to real time, extending the ban to a posteriori identification, affecting some of the most controversial cases, such as the facial recognition company Clearview AI.
Another important modification concerns private spaces, specifying that the prohibition also applies to the virtual domain.
While the aim of the preliminary discussion was precisely to bring out the views of the political groups, two European Parliament officials told EURACTIV there appeared to be a clear majority in favor of the ban.
The complete ban would go in line with a resolution adopted last October, when the centre-right European People’s Party (EPP) was largely isolated in its desire to give law enforcement some leeway in the use of these tools.
The scope of AI rules is another sensitive topic.
The co-rapporteurs proposed an exemption for public authorities of third countries and international organizations that use AI in the framework of international cooperation or judicial cooperation agreements and if they are covered by a data adequacy decision or an agreement on fundamental rights.
This compromise sits midway between progressive MEPs who call for stronger fundamental rights safeguards — which are not necessarily covered by data adequacy rulings — and centre-right lawmakers who favor a broader exemption for third countries.
Furthermore, the co-rapporteurs suggested that the regulation would not affect EU data protection rules, including those related to law enforcement, the EU consumer protection and data security regime. products, national labor law and actual research and development (R&D) activities.
The R&D exemption was particularly debated. According to one official, the problem is more with the wording, because no one wants to create a legal vacuum.
For a second parliamentary official, regarding the question of scope, it is generally accepted that each group must give up one element. Therefore, the text must be considered as a whole.
The Commission’s initial draft provided for the obligation for the suppliers of AI systems most likely to cause harm to be registered in a European database.
The compromise amendments proposed to extend this obligation to users who are public authorities and to anyone who makes a substantial change to the system. The EPP and the Greens disagree with such wording, but each of these two groups for opposite reasons.
This article has already been discussed at a technical meeting at the end of August. MEPs failed to address the new wording at Wednesday’s meeting, with the other two topics monopolizing the debate.
According to information from EURACTIV, topics such as the definition of AI, the rules for general-purpose AI and the list of high-risk systems in Annex III will also be discussed at Politics.