The argument leans on a reconstruction by Kiesling (2015), which splits Hayek’s category into two distinct problems. The complexity problem concerns aggregating knowledge that exists dispersed across society but could in principle be codified—the kind of problem that better data pipelines, prediction and optimisation genuinely address. The contextual knowledge problem concerns knowledge that does not exist outside the interaction generating it, and no amount of aggregation can substitute for it. Wright stresses that this reconstruction is interpretive and captures a generative subset of Hayek’s category rather than its whole content; much of the remainder is knowledge that agents already possess, and AI does dissolve a good deal of that. The boundary being redrawn is therefore partial, not total.

A crucial premise is declared rather than derived: a Misesian account of evaluative agency, under which ends are constituted by acting subjects rather than inferred from data. Given that premise, the paper argues, better aggregation, prediction and optimisation cannot by themselves resolve the generative subset of contextual knowledge. This is why, in Wright’s taxonomy, neither Polanyian tacit knowledge absorbed into large language models nor the microstructure price discovery performed by high-frequency traders reaches the category Hayek identified. The systems may compress and reproduce patterns of past evaluation with remarkable fidelity, but on this account they remain downstream of the evaluative acts that gave those patterns meaning.

The taxonomy is applied at three levels—the data pipeline, the tacit-knowledge debate, and the institutional conditions of price generation—presented as three applications of one classification rather than three independent demonstrations. Its concrete products are a set of operational tests, none of which has yet been run, read here against a published Q-learning pricing architecture. That caveat matters: the paper offers a conceptual framework and proposed tests, not empirical results confirming them. Its final contribution is a criterion for distinguishing context-preserving from context-suppressing deployment at the level of the market—a measure of whether a machine system leaves the situational knowledge that generates prices intact, or quietly replaces it with optimisation over a frozen record of past circumstances. Whether that criterion survives contact with running systems is precisely what the unexecuted tests are designed to determine.

Subject of Research: Hayek's epistemological challenge to artificial intelligence

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

Article References: Wright, C. (2026). Data is not knowledge: Hayek’s epistemological challenge to artificial intelligence. Humanities and Social Sciences Communications. https://doi.org/10.1057/s41599-026-09557-y

Image Credits: AI Generated

DOI: 10.1057/s41599-026-09557-y

Keywords: Hayek, artificial intelligence, epistemology, knowledge, large language models, tacit knowledge, price discovery, Mises, economics, philosophy of AI, Q-learning, contextual knowledge