Confronting the European regulatory proposal with the ethical incidents of AI.

Summary In April 2021, the European Commission released a proposal to regulate the use of artificial intelligence (AI) in the European market. The approach of this report is to put the provisions of the proposal into perspective based on an empirical knowledge of AI incidents identified in recent years. To do so, we will rely on a database of 567 ethical incidents related to AI that took place between 2010 and 2021. This database is based on the work of Charles Pownall and the Artificial Intelligence Incident Database (AIID), to which several metrics have been added to refine the analysis of the European text. By comparing the provisions of the text with empirically observed cases in Europe and elsewhere in the world, the goal is to determine whether the proposal is in line with the real practices of AI actors and to what extent they can be improved.