On Accuracy of European AI Law: A Proto Manual for Legal Scholars & Engineers

Artificial intelligence does not necessarily do what we want, rather what we ask. This has to do with accuracy, which is not a singular thing. This book describes the different aspects of accuracy defined in technical literature and which of those are reflected in European AI law.

The work provides a framework to describe different aspects of accuracy and identifies five core technical aspects of accuracy. The manuscript applies those aspects to the AI Act and its GPAI Code of practice, the Medical Device Regulation and the GDPR. It analyses the legislations themselves, the state of the art and its relation to technical standards, caselaw and legal scholarship. Additionally, a deep analysis is done on a medical AI case.

This book serves two ends. It is a proto manual for engineers aiming to implement EU legal frameworks on AI applications, especially medical device software. It also serves as a critique on the state of European AI law, aiming to inform legislators on opportunities to create a more accurate legal AI framework.

Raphaël Weuts is a leading European AI governance expert. He contributed to the development of the EU AI Act and advises multiple government agencies. He was visiting professor of AI in Leuven and taught over three thousand students, covering the fields of engineering, computer science and law. He coordinated several international organisations working on AI governance and did research on AI at multiple institutions in Leuven and Brussels. His research interests include AI law, technical architecture of AI applications in healthcare and the governance of AI agent alignment.