We’re only human after all: a critique of human-centred AI

Abstract

AbstractThe use of a ‘human-centred’ artificial intelligence approach (HCAI) has substantially increased over the past few years in academic texts (1600 +); institutions (27 Universities have HCAI labs, such as Stanford, Sydney, Berkeley, and Chicago); in tech companies (e.g., Microsoft, IBM, and Google); in politics (e.g., G7, G20, UN, EU, and EC); and major institutional bodies (e.g., World Bank, World Economic Forum, UNESCO, and OECD). Intuitively, it sounds very appealing: placing human concerns at the centre of AI development and use. However, this paper will use insights from the works of Michel Foucault (mostly The Order of Things) to argue that the HCAI approach is deeply problematic in its assumptions. In particular, this paper will criticise four main assumptions commonly found within HCAI: human–AI hybridisation is desirable and unproblematic; humans are not currently at the centre of the AI universe; we should use humans as a way to guide AI development; AI is the next step in a continuous path of human progress; and increasing human control over AI will reduce harmful bias. This paper will contribute to the field of philosophy of technology by using Foucault's analysis to examine assumptions found in HCAI [it provides a Foucauldian conceptual analysis of a current approach (human-centredness) that aims to influence the design and development of a transformative technology (AI)], it will contribute to AI ethics debates by offering a critique of human-centredness in AI (by choosing Foucault, it provides a bridge between older ideas with contemporary issues), and it will also contribute to Foucault studies (by using his work to engage in contemporary debates, such as AI).