Polanyi is perhaps the single figure in the cycle whose work was most specifically and precisely vindicated by the AI era, and the vindication is a double one that cuts in directions he did not expect. He was right that tacit knowledge cannot be reduced to explicit rules—the failure of symbolic AI is his proof. He was wrong that the tacit required the committed, embodied, personal knower—the machines acquired the competence without the commitment. That gap, between knowing and meaning, is where the cycle’s deepest inquiry lives. We have built something that knows more than it can tell and means nothing by any of it, and that strange fact reveals, by its absence, exactly what made our own knowing matter.
His concept of commitment gives the clearest available diagnosis of what large language models conspicuously lack. A model produces fluent, often accurate assertions about the world. It does not, in any sense Polanyi would recognize, commit to them. It has no stake in their truth, bears no responsibility for their falsity, reaches for no reality beyond the pattern of its training text. It will assert a fact and, prompted differently, assert its negation, with the same untroubled fluency, because there is no committed knower behind the assertions for whom the contradiction is a crisis. This is the precise technical and philosophical content of what the field calls hallucination, seen from Polanyi’s angle. The model is not lying; lying requires a commitment to truth that one then violates. It is doing something stranger: generating assertions with no one home to mean them.
The apprenticeship concern Polanyi’s framework generates is the most practically urgent thing the cycle takes from him. Tacit knowledge can only be transmitted from person to person by apprenticeship, never by instruction alone; the tradition is a relay race in which the baton cannot be described, only handed. If the machine performs the task, the human apprenticeship that used to transmit the underlying tacit knowledge may not happen. The junior doctor who lets the machine diagnose does not build the tacit diagnostic feel that comes only from doing it. AI is the first technology aimed squarely at the tacit itself—at diagnosis, judgment, writing, the competences that previous tools left to humans. When a tool absorbs the explicit, the tacit apprenticeship continues; when the tool absorbs the tacit, the apprenticeship is exactly what is at risk.
His alignment diagnosis is the contribution the cycle most directly extends technically. The things we most deeply want from AI systems—that an output be wise, that a decision be just, that a life go well—are tacit in Polanyi’s precise sense. We can recognize wisdom when we encounter it; we cannot fully specify it as a metric without remainder. The moment you make the value explicit as an objective function, you have done the thing Polanyi warned against: dragged the subsidiary particulars into focal view and destroyed the integration. Specification gaming and reward hacking are not bugs to be patched; they are structural features of optimizing for the tacit, exploiting the permanent gap between the explicit proxy and the tacit value the proxy was supposed to capture. Goodhart’s Law has a Polanyian foundation.
Born in Budapest in 1891 into a vivid intellectual household (his father built railways; his mother kept a salon), Polanyi trained first in medicine at the University of Budapest, then turned to physical chemistry. He published more than two hundred scientific papers over three decades, contributing to reaction kinetics (the Eyring-Polanyi equation), X-ray diffraction, and the theory of adsorption. His name appears in the foundational equations of the field alongside those of Henry Eyring and R.A. Evans. His students and colleagues included Eugene Wigner and Melvin Calvin, both future Nobel laureates; his own son John would win the Nobel Prize in Chemistry in 1986. This is the résumé of a man at the center of hard, formal, rule-governed science.
Yet it was from inside that science that Polanyi began to notice what the rules left out. A chemist does not become expert by memorizing equations. He becomes expert by acquiring a feel—for when an apparatus is behaving, for which anomalies are signal and which are dirt on the glassware, for which lines of attack are promising before any argument can say why. When Hitler came to power in 1933, Polanyi, a Jew working in Berlin, accepted a chair at the University of Manchester and left Germany. In England, increasingly preoccupied with the threat that totalitarianism posed to free inquiry, he drifted from the laboratory toward the questions underneath it. In 1948 Manchester created a chair in social science specifically so he could leave chemistry and think full time about knowledge, freedom, and the nature of science. Science, Faith and Society (1946), the massive Personal Knowledge (1958), and the compact Tacit Dimension (1966) followed. He had become a philosopher, but one with chemicals under his fingernails.
He must not be confused with his brother, the economic historian Karl Polanyi, author of The Great Transformation. The brothers held nearly opposite views on spontaneous order and markets. Michael’s “Republic of Science” essay defends a self-governing spontaneous order of free inquiry; Karl’s work is a critique of self-regulating markets. They are constantly conflated. This volume concerns Michael only.
We Can Know More Than We Can Tell. The foundational claim of Polanyi’s philosophy: the deepest kind of knowing runs ahead of our capacity to articulate it, and this is not a marginal phenomenon but the foundation on which all knowing rests, including the explicit knowing of science. The from–to structure of skill—attending from subsidiary particulars (the feel of the hammer, the muscles of the arm) to a focal achievement (the nail)—means that the subsidiaries are transparent, integrated below the level of anything stated. Try to make them focal and the performance collapses. The tacit is not tacit because we have been lazy about writing it down; it is structurally incompatible with being written down because its mode of existence is the integration, not any list of parts.
Personal Knowledge and the Committed Knower. All knowledge is personal for Polanyi: it requires a committed person staking herself on a claim about reality, in the responsible expectation that reality will bear her out. This is not subjectivism; it is the fusion of personal participation and universal intent that he called the fiduciary character of knowledge. A machine that produces assertions with no stake in their truth, no responsibility for their falsity, no reaching for a reality it cares about, is not a knower in Polanyi’s sense. The machine has the competence; the commitment is absent; and that gap—between competence-without-commitment and knowledge proper—is where the deepest questions of the AI era live.
Indwelling and the Body. Knowing is rooted in the body. When you use a tool well, you stop feeling the tool as an object pressing on your hand and start feeling through it, at the world on its far end. You dwell in the tool, incorporating it into your bodily being. All skill is a form of indwelling, and all knowing reaches from the embodied subsidiary awareness of a situated body out toward a focal object. Disembodied systems have no body to attend from—what they have is the statistical residue of embodiment, the vast sediment of artifacts produced by billions of embodied human acts, mined at one remove. This explains both the successes and the characteristic brittleness: the machine succeeds where the human residue is rich enough to encode the regularity, fails in the brittle and uncanny ways we recognize where success would require actual bodily attunement to a real, present situation.
Against the Ideal of Wholly Explicit Knowledge. The ideal that animates symbolic AI—that perfect knowledge is fully specified, drained of the personal, reduced to rules that can be mechanically applied—is incoherent, Polanyi argues. To use explicit rules, one must judge how they apply to a given case. That judgment is not in the rules; it is a tacit act. Any attempt to make the judgment explicit generates a further rule needing a further judgment, producing an infinite regress. Explicit knowledge is not self-sufficient; it always rests on a tacit foundation that cannot be absorbed without generating another layer. The brittleness of rule-based AI was the regress made flesh.
The Republic of Science. Science cannot be centrally planned, because the judgment of which problem is worth pursuing is precisely a tacit judgment—the scientist’s personal foreknowledge of an as-yet-unknown reality—that cannot be made explicit or handed to a planner. The spontaneous order of free inquiry is required by the nature of the knowledge, not merely by political preference. Translated to AI: a technology whose competence cannot be fully specified cannot be fully governed by specification, and the parts of its governance that matter most must live in the tacit judgment of responsible practitioners.