Beyond AI Use Rates: The Human-AI Agency Replacement Ratio and the Six-Axis Structure of Dependency Exposure

Abstract

Assisted performance does not identify what remains when AI access is withdrawn. Assisted performance is produced jointly by a human and an AI contribution, while withdrawal performance rests on the human contribution alone, so one assisted record is consistent with different human-side states. In one randomized trial, two AI conditions produced large assisted gains and neither produced an unaided result above control, one falling below it. AI-use rates likewise do not identify which human functions have shifted to AI. This article revises the Dependency Exposure framework by separating pre-severance exposure, scenario-specific severance risk, realized Six-Axis Loss, and secondary cascade. It introduces the Human-AI Agency Replacement Ratio (HARR) as a four-domain profile across Authorship, Behavior, Cognition, and Dependence. Ratio refers to domain-specific allocation or residual quantities rather than a composite scalar. Authorship, Behavior, and Cognition are identified from pre-specified generative, execution, and cognitive operations; Dependence is observed through residual capacity and recovery after withdrawal. Equal AI-use rates, and even similar AI-assisted performance, can coexist with different HARR profiles and post-withdrawal trajectories. Peer-reviewed evidence also shows durable gains after AI removal, and HARR treats these as competing trajectories to be distinguished prospectively. Severance risk is decomposed into scenario likelihood and impact. Current A/B/C allocations identify which AI-supported contributions a scenario removes, while D records human residual and recovery under stated measurement conditions; substitute coverage and function criticality further shape impact. Post-severance consequences remain organized through economic, psychological, cognitive, ontological, emergentive, and sovereign loss. The framework characterizes dependency while AI remains available and supports fallback testing before involuntary severance, at individual, organizational, and sovereign scales.