The Limits of Ethical Artificial Intelligence: Moral Agency, Non-delegable Responsibility and Sociotechnical Governance

Abstract

This article critically examines the conceptual legitimacy of the expression “ethical artificial intelligence” by asking whether artificial systems can be regarded as moral agents in any robust normative sense. Adopting a conceptual and theoretically argumentative design, and drawing on a critical narrative review of the literature on AI ethics, moral agency, algorithmic formalisation and sociotechnical governance, the article argues that AI systems may generate normatively relevant outputs, simulate morally acceptable decisions and support human deliberation, but they cannot assume moral responsibility, provide normative justification for their decisions or answer to those affected by their operation. The analysis distinguishes between ethics as a property of moral agents and ethics as a quality of sociotechnical systems designed, deployed, and governed by human actors. From a responsibility-centred, discourse-ethical and pluralist perspective, the article contends that the impossibility of fully ethical machines does not invalidate AI ethics. Rather, it redefines its scope by relocating ethical responsibility from artificial systems themselves to the human and institutional arrangements that design, authorise, supervise, and contest their use. The article concludes that the central challenge is not to construct morally responsible machines, but to prevent automation from producing zones of opacity, normative displacement, and human irresponsibility in socially consequential decision-making.