Responsibility Without Gaps: A Relational Theory of Responsibility in AI Systems

This paper synthesizes and systematizes a conceptual framework developed across a series of earlier preprints on artificial intelligence, responsibility, and socio-technical practices. It argues that the so-called “responsibility gap” is not a genuine feature of AI systems, but a category mistake resulting from the mislocation of responsibility at the level of system outputs. Drawing on Kant’s account of judgment and Wittgenstein’s analysis of rule-following, the paper distinguishes between outputs and decisions, showing that artificial systems generate outputs but do not make decisions. Decisions arise only within practices that involve interpretation, contextualization, and application. On this basis, the paper introduces the concept of authority illusion, defined as the misrecognition of distributed human responsibility as if it were located in the system itself. Under conditions where outputs function as decisions, responsibility is displaced from its position within practices and projected onto systems, where it becomes unrecognizable. The paper further develops a relational account of responsibility, according to which responsibility is distributed across socio-technical practices, including design, data selection, model construction, deployment, and interpretation. This framework is situated within a broader social ontology of dependence and antagonism, showing that authority illusion is not merely a conceptual error but a structurally sustained feature of contemporary socio-technical life. This paper synthesizes and systematizes the conceptual framework developed across four earlier preprints: Enser (2026a), Enser (2026b), Enser (2026c), and Enser (2026d).