From Measuring Mind to Assessing Consciousness: The Measurand Problem in AI Consciousness Research
Year: 2026
Type: preprint
Abstract: Research on artificial intelligence (AI) consciousness increasingly relies on theory-derived indicators and probabilistic assessment. This paper isolates a problem underlying debates about validation, calibration, indicator validity, and extrapolation: the operational identity and cross-domain transportability of phenomenal system consciousness—whether, at a specified time and under specified conditions, there is something it is like to be an appropriately individuated system. Current procedures directly assess proximal cognitive or architectural properties, including reportability, global availability, recurrent processing, higher-order representation, integration, and predictive modeling. Their bearing on phenomenal consciousness depends on a response model linking records to proximal properties and a bridge model linking those properties to consciousness. Cross-domain inference additionally requires distinct response-transport and phenomenal-bridge-transport models. Bridges must specify whether a proximal property is necessary, sufficient, necessary and sufficient, identical with or constitutive of, or probabilistically diagnostic of consciousness, since these relations license different inferences. I formulate the No Unlicensed Transfer principle: evidence that a procedure tracks a proximal property warrants only the consequences licensed by a separately constrained bridge; cross-domain transfer requires separate evidence concerning both relations. Drawing on the philosophy of measurement, construct validation, natural-kind methodology, and transportability research, I distinguish bearer individuation, observational equivalence, target coordination, and measurement invariance. A common intended referent does not itself secure a common operational variable, and model-relative aggregation may aid decision-making without constituting calibrated, theory-neutral measurement. A worked application to the Digital Consciousness Model maps its expert-elicited indicators, latent hierarchy, stance-specific likelihoods, priors, and cross-stance weighting onto these roles. The constructive proposal treats indicators as proxies within explicit models constrained by prediction, intervention, convergent and discriminant testing, negative controls, and licensed cross-domain transport.
Author: Sugio, Hajime
Institution: Sophia University
Language: English
Subfield: Philosophy
Field: Arts and Humanities
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7221238766