2. May 2021

In this primer Prof. Thomas Burri and Dr. Fredrick von Bothmer 1) discuss the EU Commission’s proposal of 21 April 2021 of new Union legislation to regulate artificial intelligence (AI) 2).

“The proposal for a Union regulation only marks the beginning of the legislative process. The Council and the European Parliament will most likely modify the proposal substantially.

Overall, the Commission’s proposal is clever, well written and balanced. Its scope is not overly broad, for it focuses on the most problematic uses of AI and contains only minimal obligations with regard to other AI. The proposal undoubtedly marks a defining moment in the history of AI, since it will ultimately lead to the first comprehensive legislative measure globally containing binding rules on AI. The proposal controversially bans certain uses of AI, but it also contains a whole host of new duties for those who put into circulation what the proposal designates as “high- risk AI”. The proposal carefully attempts to avoid posing an undue administrative burden on market actors. Obviously, some of the lessons from the General Data Protection Regulation have been learned. Nevertheless, the proposal also marks the advent of the regulatory state (and the regulatory Union) in AI and the end of unbridled freedom in all things AI.”

“The proposed regulation distinguishes three categories of AI, namely certain uses of AI that it bans, high-risk AI which it regulates in detail, and low-risk AI which it addresses to a limited extent. Since the first two categories are relatively narrowly circumscribed, the vast majority of existing intelligent algorithms falls into the third category where the proposed regulation essentially requires AI to be flagged.

The proposed regulation prohibits certain uses of AI. It bans the use of AI: a) to materially distort a person’s behaviour; b) to exploit the vulnerabilities of a specific group of persons; c) public social scoring and d) for real time remote biometric identification in public places.3 The ban is limited to the use of AI (“AI practice”). It does not preclude the development of AI which could potentially be used in cases covered by the ban. Given the general-purpose character of AI, this limitation is essential.

The material distortion of behaviour (a) and the exploitation of vulnerabilities (b) are only banned, if they causally affect a human person’s physical or psychological harm, or if they are likely to do so.4 This condition qualifies the ban and makes it narrower. But the ban is still rather vague and broad with regard to these two categories. The terms used in (a) and (b) are not standard legal terms. No case law exists that could indicate how they are to be understood. Some ill intent, maybe fraudulent or deceptive intent, seems implicitly required. Potentially, the prohibition would cover widely used tools, such as social media bots masking as regular users with a view to influencing voters, especially if the psychological harm caused were to be construed widely. It is doubtful whether such tools, whilst of ill repute, need necessarily be banned.”

The full text of the Primer can be found here: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3831424 – version of 21 April 2021 (it may be updated). And here

Prof. Burri and Daniel Trusilo also wrote this article: Ethical Artificial Intelligence: An Approach to Evaluating Disembodied Autonomous Systems available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3816687

1 Thomas Burri, Professor of International Law and European Law at the University of St. Gallen (Switzerland), Dr. iur. (Zurich), LL.M. (College of Europe, Bruges), admitted to the bar in Switzerland (corresponding author – contact: thomas.burri@unisg.ch). Fredrik von Bothmer, Dr. iur. (St. Gallen), LL.M. (Fletcher), Manager Human Rights in Social Compliance at Daimler AG (contact: fbothmer@icloud.com); this text was written in a purely personal capacity and does not represent the view of Daimler AG or engage it in any way.
2 Proposal for a Regulation of the European Parliament and of the Council Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act) and Amending Certain Union Legislative Acts, COM(2021) 206 final, 21 April 2021, available at: https://ec.europa.eu/transparency/regdoc/rep/1/2021/EN/COM-2021-206-F1-EN-MAIN-PART-1.PDF