The Upward Invisibility of Capability: Why We Can Tell More Than They Can Hear

The standard account of why expertise resists transfer locates the barrier on the senderside: experts know more than they can tell (Polanyi, 1966). This paper argues that the morefundamental barrier lies on the receiver side: we can tell more than they can hear. We definecapability invisibility as the structural phenomenon whereby those who lack a givencapability cannot perceive its absence — not through inattention or motivated ignorance, butbecause their cognitive system lacks the representational categories required to process therelevant information. Drawing on five converging lines of evidence — functional brainreorganization in expertise acquisition, structural limits of metacognitive monitoring, the freeenergy principle and active inference, bounded awareness research, and skill acquisitionmodels — we demonstrate that capability invisibility is not a correctable bias but anemergent property of cognitive architecture. Neuroimaging evidence shows that expertiseproduces qualitatively different neural pathways, not merely quantitative improvements.Metacognitive monitoring cannot detect gaps in representations it has not constructed.Under the free energy principle, each capability level constitutes a local free energyminimum from which the cognitive system has no gradient information pointing towardhigher levels, making invisibility self-reinforcing through active inference. The character ofinvisibility changes across the capability continuum: at smaller gaps, higher capability isperceived as quantitative superiority; at larger gaps, it is actively miscategorized; at thelargest gaps, it is processed as noise. The institutional consequences are direct: evaluationsystems that place individuals at capability level N in judgment over individuals at level N+2produce systematic miscategorization, not merely undervaluation. Specification alone —documentation, explanation, instruction — is insufficient to address receiver-side frameworkabsence; sustained shared experience generating structured prediction errors is required.