Embodiment, Measurement, and the Limits of Disembodied Intelligence: Why the Body Is Not an Implementation Detail

Human meaning, understood as the structure of what matters, what threatens, and what restores, emerges from viability regulation: the ongoing physiological process by which organisms maintain themselves within the bounds of continued functioning. This paper develops that claim into a precise epistemological argument with immediate consequences for psychological practice, cognitive modeling, and AI system design. Stress regulatory systems are characterized by hysteresis, the tendency to persist in dysregulated states long after the original stressor has been removed, and by a structural measurement asymmetry: the internal variables that determine which regulatory regime a person currently occupies are latent, person-specific, and recoverable only partially from behavioral or linguistic observation. The epistemic consequence is that two people can be observationally equivalent from the outside over days or weeks, while inhabiting opposite internal states that require opposite actions for recovery and viability. Any intelligence operating exclusively on external signals, including large language models and AI-assisted wellness systems, will be systematically under-calibrated in exactly the contexts where calibration matters most: near stress thresholds, during post-stressor recovery, and in the early stages of burnout or trauma response. This paper offers a non-metaphysical account of why embodiment is cognitively relevant, identifies a specific and bounded failure mode for disembodied AI in psychological and clinical contexts, and reframes the common inferential error "you seem fine, therefore you are fine" as a measurement problem with concrete implications for clinical practice, occupational health, and AI system design.