Proxy Collapse

Human beings do not ordinarily inspect consciousness, understanding, sincerity, authorship, or responsibility directly. They infer such properties from observable signs: coherent language, adaptive conduct, humor, judgment, purposive work, self-correction, social responsiveness, and insight. These signs acquired evidential force in environments where their familiar generators and target properties repeatedly co-occurred. Generative artificial systems change the generator mixture. They can produce consequential work and outputs that are apt, surprising, useful, humorous, or transformative while the system-level facts needed to infer subjective experience, understanding, authorship, or responsibility remain disputed or distributed. This paper calls the resulting calibration failure proxy collapse: a proxy retains reality and possible evidential value, but an inference calibrated in one causal environment no longer transfers unchanged after new pathways can produce the sign. The revision makes three corrections. First, proxy collapse is local to a specified proxy, target, environment, and decision; it is not a declaration that all signs have become worthless. Second, a new artificial generator cannot simply be added to the probability of the target's absence when the target status of that generator is unknown. The formal problem is generator-conditioned uncertainty and failed calibration transfer. Third, the framework distinguishes actual historical validity from culturally assumed validity, and separates task success, causal contribution, phenomenal experience, social presence, authorship, responsibility, and moral standing. Proxy collapse therefore supports neither automatic personification nor categorical denial. It calls for better-conditioned inference, visible causal and authority chains, calibrated semantic vocabulary, explicit uncertainty, and symmetric protection against over-attribution and under-attribution.