Re-aligning value alignment: A metaethical perspective on AI ethics

AI systems are, and will continue to be, designed for “bounded moral contexts” (Wallach and Wallach 2020), which risks creating a persistent gap between value alignment and the regulatory ambition of human-centric, ethical AI. This gap challenges the adequacy of prevailing ethics and governance approaches, as exemplified by the EU AI Act (2024), which remain largely anchored in applied ethics and market-driven rationales, offering little response to deeper questions of moral integration. By contrast, a metaethical perspective interrogates the ontological and epistemic foundations of moral reasoning and value formalisation in AI. Upholding this perspective, this chapter proposes a multidimensional framework that integrates empathic metaethics, emancipatory approach, democratic participation and the dismantling of structural and unjust dominant views. Taken together, these dimensions would support not only a richer, pluralistic and context-sensitive approach for AI design but also development of a holistic regulatory metaethics. The chapter concludes such a deeper and structural engagement is indispensable to advancing genuinely human-centric AI, ensuring systems are morally robust, socially responsive and conducive to human flourishing.