Layered agency as a basis for allocating accountability for AI

The increasing use of artificial intelligence-driven systems (AI) in organizations challenges the traditional concept of agency in organization theory as well as the traditional allocations of accountability. Recognizing that these issues are closely related, and drawing from several different conceptualizations of agency across disciplines, this paper develops a layered framework of agency in the context of AI. The framework distinguishes three analytically separable but interacting layers. The first layer conceptualizes agency in functional terms, focusing on the causal role. The second layer examines interpretive agency, where agency emerges through social practices and human interpretation. The third layer adopts a composite or sociotechnical aspect of agency, in which agency emerges through entanglement of sociotechnical assemblages. On this basis, the paper proceeds to discuss how a layered understanding of agency can be used to inform the allocation of accountability. Having discussed perspectives that focus exclusively on human actors or AI systems, and recognizing that any accountability analysis must be firmly rooted in the circumstances at hand, a sociotechnical approach to accountability for composite agents is outlined.