Learning by Transacting: A General Theory of Bidirectional Knowledge Flows in Market Exchange

Arrow (1962b) identified a fundamental paradox in markets for information: a buyer cannot value knowledge without possessing it, and a seller cannot demonstrate knowledge without surrendering it. Six decades of institutional design, from patent law to trade secret doctrine, answer that problem. This paper argues that Arrow’s paradox is a limiting case of a more general phenomenon. Every market transaction occurs across an interaction surface through which knowledge flows in both directions, and the classical analysis holds only under four restrictive assumptions: that knowledge is an artifact, that disclosure is one-shot, that it runs from seller to buyer alone, and that absorption is automatic. Relaxing these assumptions yields a general theory of epistemic flux in which the direction of payment and the direction of net knowledge flow are independent. The framework recovers the classical paradox, the economics of consulting, the theory of the firm’s knowledge boundary, and the platform economy as parameter settings of a single model, and it is closed with an equilibrium: buyers choose usage and absorption, a hub prices access, and welfare is evaluated against a planner. Three results follow for markets in machine intelligence. First, an inverse information paradox: consumption of intelligence requires disclosure of context, and the discloser cannot value what she surrenders, because its value is realized only in aggregation with disclosures she cannot observe. Second, a two-sided welfare theorem: where the positional harm of rival access dominates, decentralized adoption over-discloses and under-absorbs relative to the planner; where the level benefit of better models dominates, it under-discloses, so the externality’s sign is an empirical object rather than an assumption. Third, an epistemic subsidy: a hub that values aggregate flux prices inference below marginal cost, which reinterprets free tiers and zero-retention premiums as the subsidy returned. The paper unifies Arrow’s two 1962 contributions, learning by doing and the information paradox, into one apparatus; explains why vendor competition cannot dissipate the asymmetry, since money can rebate a subsidy but cannot restore relative position; and specifies the institutions the new regime will require.