From the Villani mission to the AI ACT: the emergence of European AI regulation
The regulation of artificial intelligence has progressed considerably in recent months with the finalization of the draft European regulation on artificial intelligence. It would, of course, be inaccurate to assume that France alone is responsible for a text drawing on multiple influences. However, the dual process undertaken within the framework of the Villani mission and the revision of the bioethics law has led to reflections and proposals that have subsequently resonated at the European and international levels. The recognition of the principle of Human Oversight of AI is one example.
The Villani Mission: Giving Meaning to Artificial Intelligence
While the “rise of the regulatory state” has been underway since the beginning of the 20th century, the construction of a legal and normative framework for artificial intelligence has gradually become essential. Since 1940, AI has been developing and emerging in the wake of cybernetics. This immersion profoundly disrupts our habits, our relationship to work (and to others), and requires that we structure it and restore the central role of public policies, whose decisive and crucial function will give meaning to AI.
Before the French government took up this issue, the legislative framework focused primarily on addressing the digital impacts and the right to disconnect, and then, a few months later, on developing a vision for a digital republic with the aim of unleashing innovation and creating a framework of trust. These initial steps facilitated the establishment of working groups on artificial intelligence in 2017, thus initiating a genuine national strategy extending to 2025. This objective is eminently social, as it guarantees the preservation of our humanity and prevents a change of an anthropological nature. It is in this spirit that the mission entrusted to Cédric Villani on artificial intelligence was born.
Abundant Regulatory Attempts Leading to the Completion of the AI Act
As early as 2017, recommendations advocated for the creation of a specific legal personality for the most sophisticated autonomous robots, which could be considered responsible electronic persons, liable for any damage caused to a third party. A White Paper on Artificial Intelligence in 2020, a precursor to regulation on the subject, proposed concrete measures to support the transition to an AI society in France. However, it was only in 2022 that the European Commission proposed a directive presenting harmonized rules on AI and demonstrating a commitment to moving towards a common framework. Seeking to establish a hierarchy of risks among systems equipped with artificial intelligence, the creation of a European legal framework became essential, particularly to limit risks such as “the potential profound transformation of the patient-doctor relationship.” Adopted by the European Parliament on June 15, 2023, the AI Act established a regulatory framework for the development of AI and a commitment to harmonizing its governance by guaranteeing EU Member States “the use of transparent, reliable, non-discriminatory, and environmentally sound systems.”
Recognition of the Principle of Human Oversight of AI: An Illustration of the Continuum Between the Villani Mission, the Bioethics Law, and the AI Act
Proposed in mid-2017 by Ethik-IA, the principle of Human Oversight of AI as a method for the positive regulation of artificial intelligence in healthcare and beyond was recognized for the first time in the conclusions of the Villani Mission. Human Oversight refers to the need to maintain all autonomy of human action and decision-making in the context of the increasing use of AI in healthcare. This principle was subsequently endorsed by the National Consultative Ethics Committee (CCNE) in its Opinion 129 and then incorporated into French law in the bioethics law and the draft European AI Act. In the new bioethics law, the relevant provisions are found in Article 17. The words “Human Guarantee” do not appear in the text itself but are, however, very directly referenced in the explanatory memorandum of the bill and in its impact assessment. Regarding the legislative provisions themselves, we find the dual dimension of the “human guarantee” recommended by CCNE Opinion 129: a new duty to provide information on the use of AI in healthcare is recognized in order to allow, as far as possible, free and informed consent to the treatment protocol incorporating algorithmic processing, and human supervision in the algorithmic design and in its real-world application. Nevertheless, while the general principle is not explicitly stated, Article 17 imposes a series of new obligations that effectively implement elements of human oversight. In the draft European AI Act, the Human Oversight principle is reiterated in Articles 14 and 29. Paragraph 1 of Article 14 stipulates that artificial intelligence solutions must be designed and developed in such a way that they can be supervised by humans. This human oversight must be subject to real-world monitoring of the artificial intelligence. Article 29 of the draft regulation, which is part of Chapter 3, “Obligations of suppliers and users of high-risk AI systems and other parties,” concerns the obligations of users of AI systems. Failure to comply with these provisions will result in penalties equivalent to those of the GDPR. This race towards regulation is thus an investment against the emergence of new risk models, which public authorities must address regularly in the years to come.
[pdf-embedder url=“https://www.sih-solutions.fr/wp-content/uploads/2023/11/014SIHMAG_v02_p08-09.pdf” title=“SIH Solutions Magazine”]