The End of Interpretive Sovereignty — A Stratigraphy of Tacit Knowledge in the Age of the Universal Decoder (Subtítulo/traducción ES en la descripción: «El fin de la soberanía interpretativa. Una estratigrafía del conocimiento tácito ante el decodificador universal»)

Almost all of the literature measures labour's exposure to AI as a property of AI. This article reframes it as a property of knowledge and of its legibility by the technique of each era… (pegar el Abstract completo del PDF EN). Theoretical companion (Paper B) to the ITEA empirical programme. Preprint / working paper v2.3 — not peer reviewed. Spanish and English versions included in this record. Abstract Almost all of the literature measures labour’s exposure to AI as a property of AI. This article reframes it as a property of knowledge and of its legibility by the technique of each era. We begin from an objection that in fact strengthens the framework: converting tacit knowledge into explicit knowledge is not an invention of AI but the very mechanism of technical progress (Cowan, David and Foray, 2000; Mokyr, 2002) —and it was always a process with five discernible phases (elicitation · translation · fixation · transaction · interpretation), repeated from orality to writing, from the alphabet to Nebrija’s grammar, from mathematical notation to programming languages. The institutions that governed codification for two centuries (patent, contract, trade secret, employee-invention law) are anchored to the phases of that process. Hence a decomposition: effective expropriation = exposure × CI* × κ_t, where CI* is the intrinsic articulability of knowledge (invariant; what actually calibrates the instrument) and κ_t is the capture capacity of the state of the art (variable; dated). Each new code, moreover, created a decoding class whose interpretive exclusivity constituted the only form of property that knowledge —non-rival, non-excludable once expressed— admits: interpretive sovereignty, exercised as control over the elicitation gate of the process. The discontinuity of generative AI is not that it codifies, nor that it codifies faster: it is that it suppresses the process. Its input language is, for the first time, the very natural language in which the knowledge worker already expresses their knowledge: the five phases collapse simultaneously, the translator class (inventor, engineer, programmer) that fixed the act and the transaction disappears, and where there was a process there is now ingestion without process —with no moment to which to anchor title, price or consent. The asymmetry of decoding is not abolished: it is recentralized in the model’s weights, the most illegible code in history. Generative AI is, strictly, the first universal decoder —a single interpreter that reads every substrate (language, code, demonstration, trace) and dissolves the plurality of decoding classes—: universal in the extension of codes but bounded in the depth of strata, so that the gradient turns but stops, and with it the decoding rent in which value resided collapses all at once. This supplies a mechanism for the turn of the gradient and for the prior-title problem (the worker’s interpretive monopoly, and with it the value of the knowledge —not its possession— dissolves), and explains why the viable remedies are ex post on the stock (θ·V) rather than ex ante on transactions. We propose a stratigraphy of tacit knowledge in five strata (T1 by omission · T2 procedural-demonstrable · T3 somatic · T4 collective-situated · T5 deontic) anchored in Collins (2010) and Soler-Zwart (2013), and a philology of substrates (pairs ⟨language, interpreter⟩) that dates which strata become legible and which class retains sovereignty. κ_t ceases to be a scalar and becomes a vector of legibility per stratum. We derive an empirical programme on O*NET: a Tacit Residuality Index (the photographic negative of exposure) and a falsifiable prediction —AI penetration decreases monotonically with stratum T within each qualification level— with the evolution of κ_t measured via patents, scientific publication and regulatory churn. Keywords: tacit knowledge; codification; codification process; substrate-languages; interpretive sovereignty; universal decoder; value of the decoder; Polanyi’s paradox; stratigraphy; tacit residuality; generative AI. · JEL: O33, J24, D83, K20.