The EU's ethics guidelines for trustworthy AI
On 8 April 2019 the High-Level Expert Group on AI set up by the European Commission published its Ethics Guidelines for Trustworthy AI: four ethical principles, seven requirements for systems, and a self-assessment list for piloting. The December 2018 draft drew feedback from more than 500 contributors.
Why it matters
The document calls itself non-binding, yet its requirements became the basis of binding ones: the 2021 AI Act proposal writes that its minimum requirements for high-risk systems are derived from these guidelines.
The authors are the 52 members of a group the Commission set up in June 2018. Trustworthy AI has three components - lawful, ethical and robust - but the guidelines deliberately give no guidance on the first. The four principles: respect for human autonomy, prevention of harm, fairness, explicability. The seven requirements: human agency and oversight; technical robustness and safety; privacy and data governance; transparency; diversity, non-discrimination and fairness; societal and environmental well-being; accountability. The assessment list in Chapter III was to be piloted and returned to the Commission revised in early 2020. The document creates no obligation: stakeholders may use it voluntarily, and it states that it does not substitute for regulation. It does not meet the criterion of a normative text with articles; the record stands here for the consequence read in the AI Act proposal.