Testing the Transition from AI Advantage to Practical Dependence A Minimal Knowledge-Work Pilot Protocol
Testing the Transition from AI Advantage to Practical Dependence: A Minimal Knowledge-Work Pilot Protocol presents a preregistration-ready experimental design for testing two early mechanisms through which a local generative-AI advantage might contribute to practical dependence in knowledge work. The protocol operationalizes a limited segment of the conceptual framework developed in When Refusal Remains Formal: Infrastructural Non-Refusability in AI-Mediated Society. It does not attempt to demonstrate society-wide technological lock-in, generalized cognitive decline, or irreversible AI dependence. Instead, it separates two empirically testable questions: whether repeated AI assistance changes a worker’s independent performance, reasoning reconstruction, source verification, or error detection after the system is withdrawn; and whether exposure to high-quality AI-assisted work changes the standard by which identical unassisted human work is evaluated. The proposed study contains two linked but analytically independent experiments. The worker experiment compares three conditions: a control group working without AI, a continuously assisted group, and a withdrawal group that receives AI assistance during an exposure phase and subsequently completes a comparable task without it. The evaluator experiment tests criterion shift by exposing evaluators to either ordinary human reference outputs or high-quality AI-assisted reference outputs before they assess the same set of unassisted human work. Primary outcomes include output quality, completion time, source fidelity, reasoning reconstruction, seeded-error detection, post-withdrawal performance, professional acceptability, and expected completion time. The protocol specifies task construction, parallel task packs, randomization, blinding, preregistration, mixed-effects analysis, ethical safeguards, data structure, feasibility checks, and explicit interpretation rules. The design deliberately does not presume deskilling. Post-withdrawal decline, return to baseline, and learning transfer are all treated as theoretically meaningful outcomes. It also distinguishes individual recoverability from process and institutional recoverability, preventing participant-level findings from being overextended into claims about organizations or society. This document is intended as an open methodological artifact that can be implemented, criticized, adapted, or falsified by independent researchers. Protocol DOI: 10.5281/zenodo.21596479 Associated conceptual article:Karel Hrubec, When Refusal Remains Formal: Infrastructural Non-Refusability in AI-Mediated SocietyDOI: 10.5281/zenodo.21595452