Why Human–AI “Collaboration” Is a Category Error

Human–AI systems are increasingly described as collaborative, assistive, or shared decision-making systems. These descriptions presuppose that responsibility can be distributed across human and artificial components without ethical or conceptual loss. This paper challenges that presupposition. It argues that responsibility is an indivisible normative relation defined by undivided justificatory authority, not a divisible resource that can be shared, optimized, or partially offloaded. Systems that participate in the justification of human judgment—by resolving uncertainty, stabilizing interpretation, or recommending action—do not collaborate with human agents but instead structurally displace responsibility. What is preserved in such arrangements is the phenomenology of choice; what is undermined is its normative basis. Much of what is described as “human–AI collaboration” therefore rests on a category error, with significant implications for ethical accountability in AI alignment and system design.