"Editor's note: This article was written before the publication by the EU Commission of its proposal for an artificial intelligence (AI) regulation [29] . In a first and provisional analysis of the proposed regulation, we observe that the proposed regulation incorporates some of the basic principles laid down in our article: it prioritizes fundamental rights and incorporates some human rights principles, such as accountability, and the inclusion of governance through supervisory authorities to implement and enforce the regulation. Nevertheless, we still feel that many of the suggestions present in our article, which would help to operationalize the regulation, are not addressed. One example is the reduced scope of the regulation to a list of “high risk applications,” leaving without a legal framework all other AI applications. We believe that the principles that inspire the regulation should also be applied in “lower risk applications.” Defining only the compliance process for AI developers, but leaving open the specific technical requirements that these high risk applications shall meet leaves untouched the existing gap between legal language and engineering practice. There are no described mechanisms by which all stakeholders (other than developers and implementers) can influence AI development, monitor their performance or claim redress if harmed. These shortcomings and other issues presented in our article leave the door open to loopholes that we hope the European Parliament can fix during the legislative process."

Support: Digital