A critique of the ‘as–if’ approach to machine ethics

Abstract

In this paper, I argue that the replication of the effect of ethical decision-making is insufficient for achieving functional morality in artificial moral agents (AMAs). This approach is named the “as–if” approach to machine ethics. I object to this approach on the grounds that the “as if” approach requires one to commit to substantive meta-ethical claims about morality that are at least unwarranted, and perhaps even wrong. To defend this claim, this paper does three things: 1. I explain Heidegger’s Enframing [ Gestell ] and my notion of “Ready-Ethics,” which, in combination, can hopefully provide a plausible account for the motivation behind the “as if” approach; 2. I go over specific examples of Ethical AI projects to show how the “as if” approach commits these projects to versions of moral generalism and moral naturalism. I then explain the flaws of the views that the “as if” approach necessitates, and suggest that they cannot account for the justificatory process crucial to human moral life. I explain how Habermas’ account of the justificatory process could cast doubt on the picture of morality that the meta-ethical views of the “as if” approach proposes; 3. Finally, I defend the relevance of discussing these topics for the purpose of functional morality in AMAs.