The Politics of Standardising Ethics. The IEEE Initiative and the Governance of AI

Abstract

Although scholars have discussed the role of both ethics and technical standards in AI regulation, the recent creation of technical standards on AI ethics has been under-discussed. We explore this gap by asking two related questions: 1) How does ethics change through standardisation? 2) What should be the mutual roles for risk producers, assessors and bearers in the use of ethical standards for governing AI risks? Our case study is the Institute for Electrical and Electronic Engineers (IEEE) 7000-2021, part of the IEEE 7000 series, and its potential interplay with the AI Act requirements (AIA; Art. 40-41). We follow approaches from Science and Technology Studies (STS) and social sciences studies of risk, which allows us to study standards 1) as an exemplary case of the phenomenon of the AI ethification and institutionalisation of ethics, 2) which is increasingly framed in terms of (ethical or value) risks to be minimised and managed. We aim to show that ethics standards, while they promise to offer a more responsible way to govern AI, transform ethics into an engineering requirement modelled on the procedures, language and logic of operationalisation of standardisation. As a result, IEEE ethics standards suffer from traditional challenges related to representation, accountability, and enforcement. We, therefore, argue that there is a need for 1) a division of powers between risk producers and “ethical” assessors of risk who define acceptable “ethical” thresholds; 2) alternative, dissenting voices in standardisation working groups, especially civil society or consumer organisations that are often de facto excluded from these exercises.