Philosophy as Cognitive Assay: Measuring the Delegation Legitimacy Boundary in AI-Assisted Knowledge Work
Citation. Tomita, K. (2026). Philosophy as cognitive assay: Measuring the delegation legitimacy boundary in AI-assisted knowledge work [Preprint]. SocArXiv. https://doi.org/10.31235/osf.io/e9qw5_v2
Summary. This conceptual preprint proposes a “cognitive assay” for designing AI-assisted knowledge work around a delegation legitimacy boundary: AI may reduce the cost of turning expertise into formal outputs, but it cannot itself supply the legitimate warrant for value-laden judgments. The framework separates factual (Sein-type) and value-laden (Sollen-type) components of specification, using ratio profiles rather than a single automation score. Its central design implication is that accelerating codification and combination work shifts attention toward the human processes that develop, contest, and renew evaluative judgment.
Key claims
- The paper distinguishes externalization cost from specification: AI can compress the technical work of converting expertise into code, reports, or designs without replacing the situated judgments that define what should be built, excluded, or treated as quality.
- Its proposed assay treats factual (Sein-type) components of a documented feature decision as provisionally AI-assessable and value-laden (Sollen-type) components as requiring domain-expert evaluation; hybrid decisions should be represented as a profile, not collapsed into a single score.
- The five-step protocol asks AI to extract and classify documented decisions and score factual adequacy, while an expert identifies omitted or implicit value judgments, evaluates normative fit, and interprets the resulting profile.
- Re-reading the SECI knowledge-creation model through this decomposition leads the author to argue that AI acceleration of Externalization and Combination leaves Socialization and Internalization as human-limited bottlenecks for renewing evaluative capacity and absorbing knowledge into practice.
Evidence & limitations
- Design: conceptual framework grounded in philosophical argument and a case-study-derived decomposition of recorded design decisions; the paper demonstrates an analytic protocol rather than testing it across organizations or outcomes.
- Limitations: the proposed scoring values and AI/expert division of labor are theoretical and have not been empirically calibrated, validated for reliability across domains, or shown to improve design quality, legitimacy, learning, or organizational performance; the value/fact distinction can be hybrid and contested.
- Provenance: the Drive PDF is an exact SHA-256 match to the public SocArXiv v2 file identified by the OSF API; Crossref records the v2 DOI and a 2026-06-11 posted date.