부산대학교 도서관

The importance of any conversation on the ethics of Artificial Intelligence (AI) cannot be overstated as such advanced technologies now influence almost every aspect of our lives. To date, dominant approaches to AI ethics have traditionally focused on the agency of the organisations creating such AI projects (e.g. the Google AI principles), but more recent approaches focus on wider structures and frameworks that may potentially support responsible AI design (e.g. the Ethically Aligned Design framework by IEEE). In this paper, we argue that such frameworks are both incomplete and insufficient as these approaches give primacy to either agency or structure. Moreover, in both approaches, the emancipation of those for whom AI is designed is almost completely ignored. To remove these lacunae, we suggest an emancipatory approach to AI ethics based on the key tenets of the philosophy and ontology of critical realism. The stratified ontology of critical realism suggests that structural conditions may be modified to activate the mechanisms supporting any desired outcome (e.g., bias-free AI). In addition, such an emancipatory approach works towards activating the mechanisms supporting the reflexivity of the stakeholders in any specific AI context. This paper makes an argument for adopting this particular philosophical approach, and we discuss three critical realist-based mechanisms in support of our argument: ethical training of AI professionals, citizen engagement, and freedom of AI information.