Data is not knowledge: Hayek’s epistemological challenge to artificial intelligence

Abstract Artificial intelligence is routinely described as a knowledge technology. This paper argues that the description misplaces a boundary Hayek drew in 1945, between “scientific knowledge”—codifiable, statistical, articulable—and “knowledge of the particular circumstances of time and place.” Kiesling (2015) reconstructs part of the latter as the contextual knowledge problem: knowledge that does not exist outside the interaction producing it, as against the complexity problem of aggregating knowledge that exists dispersed but can in principle be codified. The reconstruction is interpretive and identifies a generative subset of Hayek’s category rather than its whole content; much of the remainder is knowledge agents already possess, and AI dissolves a good deal of it. The argument is not that machine systems cannot discover novel alternatives, nor that their search spaces are closed; both are conceded. It is that discovering means does not originate the ultimate ends by reference to which alternatives acquire economic significance, and that current systems receive those ends rather than constituting them. Under a Misesian account of evaluative agency—a premise the paper declares rather than derives—better aggregation, prediction and optimisation therefore cannot by themselves resolve the generative subset, so neither Polanyian tacit knowledge in large language models nor microstructure price discovery by high-frequency traders reaches it. The taxonomy is applied at three levels—the data pipeline, the tacit-knowledge debate, and the institutional conditions of price generation—as three applications of one classification rather than three independent demonstrations. Its products are a set of operational tests, none of which has been run and which are here read against a published Q-learning pricing architecture, and a criterion distinguishing context-preserving from context-suppressing deployment at the level of the market.