Rethinking ethics in AI policy: a method for synthesising Graham’s critical discourse analysis approaches and the philosophical study of valuation

Abstract

ABSTRACT Here I use aspects of Phil Graham’s discourse analytical work to examine forms of e/valuations and critically analyse the formulation of truths in the constitution of Artificial Intelligence (hereafter, AI). This paper focuses on two 2019 documents: Ethics guidelines for trustworthy AI (AI HLEG, 2019a) and Policy and investment recommendations for trustworthy AI (AI HLEG, 2019b). My aim here is to provide a timely contribution to contemporary philosophical–methodological innovations in documenting the constellation of values that are prefigured in human-centric constructions of AI. The analysis is informed primarily by principles from Graham’s distinctive approach to critical discourse analysis (CDA), which I take to be the philosophical study of valuation. In that approach, political economy is taken as a composite formulation of values whereby neo-liberalism is discursively entwined and progressed through a system of principles of e/valuation. This paper presents a roadmap to demonstrate the usefulness of a philosophically grounded interdisciplinary piece of linguistic research through which evaluative semantic categories can be usefully synthesised with CDA to systematically expose the assumptions which underpin current truth claims and values – in this case, about and around the ethics of AI.