The Missing Human-State Variable in Autonomous Materials Science: Toward an Operational Neurosemantic Interface

This position paper argues that autonomous materials systems contain a structural omission: material state and algorithmic state are increasingly modeled continuously, while the human supervisor is usually represented only through discrete interventions, responses, or self-report. It proposes treating human supervisory state as an explicit time-varying control variable, Ht≠constH_t \\neq \\mathrm{const}, and outlines an architecture in which longitudinal neural measurements contribute to an individualized, machine-readable human-state representation used to allocate supervision within autonomous materials workflows. The paper makes a strict distinction between recurrent neural structure and semantic status. Recurrent EEG patterns are treated only as candidate structures until they acquire independently validated functional grounding. EEG is not assumed to be superior to behavioral, ocular, cardiovascular, electrodermal, or other non-neural measurements; its value must be demonstrated against a strong multimodal baseline. The human-state estimate is intended to influence when and how human supervision is requested, not to authorize physical experimental actions directly. A contingent compositional extension is discussed separately from the core architecture. A companion paper provides the formal operational definitions, compression criterion, promotion rule, staged validation program, multiplicity control, feasibility analysis, and materials-control experiment required to test the proposed architecture.