Michael Polanyi

In the [YOU] on AI Field Guide

Polanyi is the epistemological engine beneath the cycle's central anxiety. The exhilaration Segal documents in [YOU] on AI—the twenty-fold productivity multiplier, the imagination-to-artifact ratio collapsing, the coding assistant that produces in an hour what a team spent a year building—operates entirely in the explicit dimension. The large language model processes text. It processes text with extraordinary sophistication. But the domain of text is not the domain of human knowledge; it is the portion of human knowledge that made it into language, that was written down, that entered the training corpus. The tacit knowledge that lives in the body and in the perceptual systems calibrated through years of material engagement is not in the training data. It cannot be in the training data. It is the knowledge that resists the medium of language.

The senior software architect in the cycle who could “feel a codebase the way a doctor feels a pulse” is Polanyi's concept embodied. His twenty-five years of debugging have deposited layer after layer of subsidiary awareness beneath the surface of conscious attention. He does not think about the codebase; he thinks through it, attending from accumulated understanding to the focal judgment that something is wrong. That judgment precedes articulation. The wrongness registers as a felt quality before the explicit diagnosis begins. This is what Polanyi's Paradox names: the tacit exceeds the explicit, and any system trained exclusively on the explicit is trained on a systematically incomplete representation of expertise.

Segal's honest account of nearly publishing a fabricated Deleuze passage is the cycle's paradigmatic case of Polanyi's indwelling gone wrong. When a tool is indwelt—absorbed so completely into the user's perceptual apparatus that the user attends through it rather than to it—the critical evaluation that indwelling requires is suspended. A cane transmits the texture of the pavement faithfully. A large language model hallucinates knowledge with the same fluency with which it produces it. The from-to structure that enables skilled tool use is the same structure that makes the hallucinations invisible until they are scrutinized. The Deleuze passage read like insight because it was smooth, and the smoothness had suppressed exactly the scrutiny that would have caught it.

Polanyi would not have expected the market to notice the erosion. The market evaluates the focal product—the brief, the code, the analysis—and cannot distinguish the brief produced by deep subsidiary awareness from the brief produced by shallow tool dependency. Both are competent by every explicit standard. The distinction exists in a dimension the market does not measure and will not measure until the moment of consequential failure: the system deployed under conditions no specification anticipated, the patient whose symptoms departed from the training data, the infrastructure tested by a load the test suite never simulated. At that moment, the tacit ground that would have detected the problem—and that was not deposited because the friction that deposits it was handled by the tool—reveals its absence.

Origin

Born in Budapest in 1891 to a prosperous Jewish family, Polanyi trained as a physician before pivoting to physical chemistry and earning a distinguished reputation for his work on adsorption and reaction kinetics at the Kaiser Wilhelm Institute in Berlin. When the Nazis came to power in 1933, he accepted a position at the University of Manchester, where he spent the rest of his scientific career before turning, in the 1940s, to the philosophy of knowledge. The turn was not a retirement from science but a response to it: Polanyi had watched the Soviet Union attempt to direct scientific research by central planning, and the attempt had convinced him that the planning body could not know what it needed to know to direct science—that the tacit, distributed, personally committed character of scientific judgment is precisely what makes science work and what makes it resistant to the administrative rationalization that destroyed it in Lysenko's Russia.

His major works accumulated across three decades: Science, Faith and Society (1946), Personal Knowledge (1958), and The Tacit Dimension (1966). The argument developed steadily from the specific case of scientific judgment to a general epistemology: all knowing has a tacit dimension, all knowledge is personal in the sense that it requires a knower who commits to it, and the positivist ideal of impersonal, fully articulable, objective knowledge is not a rigorous aspiration but an incoherent one. The formal occasion of the Manchester Debate in 1949—the seminar where Polanyi posed his challenge to Turing directly—gave the tacit-knowledge framework its sharpest AI application, though neither participant could have foreseen how precisely the question would anticipate the next seventy years of the field.

Polanyi died in 1976, four years before the personal computer and three decades before the large language model. But his framework anticipated both with a structural precision that makes him, alongside Hubert Dreyfus—who drew on Merleau-Ponty's phenomenology to make similar arguments from a different tradition—the philosopher whose work the AI moment most urgently requires.

Key Ideas

Tacit Knowledge. Tacit knowledge is the vast, inarticulate substrate of understanding that operates beneath conscious awareness and cannot be captured in specification. It is not a limitation to be overcome but a structural feature of all knowing. Every explicit statement presupposes a tacit framework within which the statement makes sense. The training data is the explicit. The tacit is the ground from which the explicit emerged, and it cannot be reconstructed from the explicit alone.

Indwelling. Indwelling is the process by which a tool becomes phenomenologically transparent—absorbed so completely into the user's perceptual apparatus that she attends through it rather than to it. The pianist does not attend to the keys; she attends through the keys to the music. The blind person does not attend to the cane; she attends through the cane to the pavement. When a tool is indwelt, critical evaluation of the tool's outputs is suspended—which is safe with a mechanically faithful tool and dangerous with a tool capable of confident hallucination.

From-To Structure. The from-to structure is the universal architecture of all knowing: consciousness attends from subsidiary elements to focal meanings. The subsidiary elements must remain subsidiary—the moment they become focal, the skill dissolves. AI restructures this architecture for millions of practitioners simultaneously, changing the content of the subsidiary awareness from which they attend and thereby changing the quality of the focal judgments they can make.

Personal Knowledge and Connoisseurship. Personal knowledge is knowledge that requires a knower who commits to it—who stakes her reputation, her judgment, her identity on the claim she makes. AI produces outputs that lack any such commitment: no one stands behind them. Connoisseurship—the cultivated capacity to distinguish quality from adequacy through tacit standards that resist specification—is the most personal form of knowing and the form most systematically threatened by indwelling an unreliable tool.

Polanyi's Paradox. Polanyi's Paradox—formalized by economist David Autor in 2014—is the observation that humans routinely outperform explicit instruction: we can do far more than we can say how to do. This paradox explains why the tasks most resistant to automation are precisely the tasks that require tacit knowledge, and why deep learning, which extracts statistical patterns from explicit outputs, captures the shadow that tacit knowing casts across a corpus without capturing the knowing itself.