Regulative but not distributive: a dual-layer model of artificial moral responsibility

Abstract

As autonomous AI systems increasingly cause real-world harm, traditional responsibility frameworks—anchored in free will and the capacity to suffer—have proved inadequate, resulting in what scholars call a "responsibility gap". This study explores the conceptual viability of artificial moral responsibility and delineates its definitional boundaries. Drawing on responsibility theories, functionalist agent frameworks, and punishment theory, alongside psychological findings of retributive intuitions and sociological insights into collective conscience, this paper differentiates regulative responsibility from distributive responsibility and systematically evaluates the extent to which artificial intelligence meets the constitutive criteria of regulative moral responsibility. The analysis shows that AI can legitimately bear corrective responsibility through consequentialist correction mechanisms, yet cannot satisfy victims' retributive demand for desert, which requires a morally responsive human community. The paper concludes by proposing a "dual-layer responsibility model", in which AI assumes corrective responsibility while humans retain residual symbolic responsibility—jointly addressing both the responsibility gap and the retribution gap without fabricating AI minds or abandoning victims' demands for justice.