Regulating high-risk artificial intelligence systems : a critical analysis of the EU ‘Artificial Intelligence Act’ proposal

Abstract

In April 2021, the European Commission unveiled a proposal for an EU regulatory framework on artificial intelligence; a proposal which is essentially the first ever attempt to enact a horizontal regulation of AI within the EU. This proposal generated a lot of interest; AI technologies are expected to bring forth a wide array of economic and societal benefits to a wide range of sectors. However, despite its numerous advantages and possible applications, AI is not flawless. In fact, it is envisaged that AI will give rise to various legal challenges and implications. The proposed Regulation adopts a risk-based approach, introducing different requirements according to the perceived level of risk imposed by the AI system. This dissertation conducts a critical analysis of the EU Commission’s proposed AI Act, in relation to high-risk AI systems and any obligations emanating or relating thereto, highlights multiple shortcomings and proposes solutions thereto. In the proposed Regulation, no definition of ‘high risk AI systems’ is provided; nonetheless, by taking a look at Articles 6, 7 and Annex III of the proposed law, the rationale behind introducing numerous obligations and requirements for high risk AI systems becomes evidently clear. High risk AI systems are meant to be used in areas which essentially form the foundations of society and fundamental human rights, and embody the quintessential elements of the EU’s vision – justice, fairness and self-empowerment. There is no denying that uncontrolled use of AI in these sectors may lead to devastating outcomes, irrespective of whether it is the result of a rogue AI or a criminal’s malicious intent. Despite the Commission’s laudable intentions, the proposed regulation has received various criticism from academics. This dissertation highlights some of the major weaknesses of the proposed regulation, such as the imposition of too many obligations on economic operators, which could potentially end up pushing AI developers away from the internal market. This seems to defeat the scope of promoting innovation. Following a thorough analysis of the provisions relating to high risk AI systems envisaged in the proposed law, this study concludes that the proposed regulation has some shortcomings which should be addressed before the law is enacted and enforced.