Moral Judgment Without Moral Agency: Why Judgment Accuracy Does Not Ground Moral Agency in the Age of AI

As large language models increasingly match human performance on moral-judgment tasks, two bodies of literature have developed in parallel. Empirical moral psychology treats moral judgment as an output and benchmarks artificial systems against human responses. The philosophy of artificial moral agency denies agency to such systems mainly on the ground of sentience, an empirically inaccessible criterion that has left the debate stalled. This paper argues that human–AI convergence in moral judgment conceals a categorical asymmetry that neither literature has theorized. Drawing on Spinoza’s conatus and Damasio’s account of homeostasis and affect, it argues that moral agency presupposes affective-homeostatic self-maintenance as its constitutive ground: the organization from which valuation first arises. This condition is necessary but not sufficient, and the claim is constitutive rather than normative, so it does not derive an “ought” from an “is.” Sentience is repositioned as the experiential manifestation of this organization rather than the explanatory terminus. Contemporary AI systems, including those trained with human feedback, are therefore moral judges without moral agency. The paper closes with a behavioral research program, including concrete experimental designs, on the conditions under which people attribute moral agency to human and artificial agents.