WELFARIST MORAL GROUNDING FOR TRANSPARENT ARTIFICIAL INTELLIGENCE

Abstract

Popular calls to make AI systems ‘transparent’ appeal to a wide range of goals that transparency ostensibly helps us pursue. Yet, as I’ll discuss, there remains considerable ambiguity about why these goals matter morally, and whether transparency is sufficient (or even necessary) for achieving them. Sceptics, moreover, have raised important challenges against the principle, which ought to be better incorporated into our understanding of when and how to pursue transparency. This thesis argues that a satisfactory treatment of these underlying moral considerations may be found by grounding calls for AI transparency in the demands of welfare-consequentialism. By shifting the focus away from the mere technical act of making an AI system transparent towards the benefits and harms that transparency might bring about, welfarism helps us understand transparency as a broader moral and political ideal about how we should relate to powerful technologies that make decisions about us.