In 2000, the United States needed to refurbish its W76 nuclear warheads. This required producing a material called Fogbank — an aerogel-like substance used in the weapon's two-stage detonation process. The original facility had closed in the early 1990s. Nearly all the personnel with direct production knowledge had retired or left. Few records had been kept.
Engineers spent five years failing to reproduce it. The stumbling block was invisible to them: a specific impurity in the original material that was structurally necessary. Modern quality-control procedures were removing it, because it looked like a contaminant. The original producers had not known the impurity was essential — they were making Fogbank correctly without understanding why their process worked. That understanding had never been written down, because it had never been made conscious.
Recovery cost approximately $92 million. The knowledge had been lost not in spite of documentation, but partly because assumptions about quality had led engineers to improve the process in ways that destroyed it.
This is the problem Michael Polanyi described in 1966 in a single sentence: "We can know more than we can tell."
The Structure of Knowing
Polanyi was a physical chemist who turned to philosophy after World War II. He was arguing against two things simultaneously: Soviet central planning, which assumed all knowledge could be made explicit and centrally managed, and the Western positivist tradition, which held that genuine knowledge must be expressible in propositions. His counter: both were wrong about the nature of knowledge itself.
His framework begins with something mundane. You can recognize a friend's face in a crowd of thousands. You cannot describe how — not with sufficient precision for someone else to recognize that face from your description. The knowledge exists. Its articulation does not. And critically: if you try to articulate it while doing it, focusing consciously on each feature, you disrupt the recognition. Making the subsidiary explicit destroys the focal whole.
This is what Polanyi called the from-to structure of tacit knowing. We attend from subsidiary particulars to a comprehensive meaning. The particulars are used, not examined. The moment you examine them, the meaning dissolves.
A pianist who starts thinking about each finger movement mid-performance loses the music. A surgeon using a probe feels at the tip, not in the palm — the hand has become part of the instrument. Dreyfus argued that a chess grandmaster, when forced to explain each move as they play, is pulled away from the tacit pattern recognition that expertise actually runs on — replaced by deliberation that expertise had already surpassed. The explicit knowledge is real; it is also, in the moment of performance, counterproductive.
What This Means for Transmission
If knowledge cannot be fully articulated, it cannot be fully transmitted through text. It requires direct, sustained, embodied contact between someone who holds the knowledge and someone who is developing it.
Harry Collins, a sociologist of science at Cardiff University, extended Polanyi's framework into three distinct types. Relational tacit knowledge is tacit for contingent social reasons — someone hasn't thought to write it down, or assumes the other person knows. This could in principle be made explicit. Somatic tacit knowledge is embedded in the body: the body knows things the conscious mind cannot access. And then there is the deepest type — collective tacit knowledge — which is not located in any individual body at all. It is a property of a practice community: the distributed, implicit standards of people who have spent years doing something together. What counts as a rigorous proof. What a "natural" movement looks like in a craft. These standards exist nowhere in writing. They live in the shared practice of a community, and they cannot be transmitted by proximity to any single master. They require immersion in the community itself.
Most knowledge management systems address the first type, acknowledge the second, and completely ignore the third. The third is the most important.
The Transmission Infrastructure
Every traditional civilization built systems specifically designed to transmit what cannot be written down. The guild apprenticeship system was not primarily economic or a quality-control mechanism. It was epistemological infrastructure — built to move knowledge from body to body across time.
The seven-year minimum apprenticeship was not bureaucratic excess. It was the minimum time required to build subsidiary awareness to a depth at which genuine mastery becomes possible. Stuart and Hubert Dreyfus, studying pilots and chess players in the 1980s, mapped this as a five-stage progression from novice to expert. Novices follow context-free rules. Experts no longer rely on rules at all — they perceive situations as calling for a response, without conscious deliberation. Expertise is precisely the replacement of rules by tacit pattern recognition. You cannot skip the years. There is no shortcut.
The masterpiece, the physical object evaluated by existing masters, was not just a competency test. It was a knowledge audit conducted by people who could recognize tacit competence through its visible products, because they themselves held the same knowledge. You cannot evaluate a masterpiece from a rubric. You can only evaluate it if you know what it is to have made something equivalent.
When the French Revolution abolished guilds in 1791 and industrialization made craft production economically marginal, what was lost was not primarily the output of craft knowledge. It was the transmission infrastructure. The explicit knowledge — recipes, dimensions, procedures — survived in texts. The tacit knowledge was in people, in communities of practice, in the sustained contact between masters and apprentices. When those communities stopped, the knowledge stopped.
Roman concrete is the extreme case. The material stood for 2,000 years. MIT researchers confirmed its self-healing mechanism in 2023 — lime clasts that recrystallize over centuries, sealing cracks. The mechanism was present the entire time. The knowledge of how to produce the material, never fully codified, died with the practice. What we have now is chemical analysis of surviving structures, not the knowledge of how to build with the material.
The Same Loss, Five Domains
The pattern repeats across industries, with different technical details and the same underlying structure.
In traditional medicine, pulse diagnosis requires reading approximately 29 distinct qualities across three positions on each wrist, each at three depths. A 2021 review in Complementary Therapies in Medicine concluded that while some elements can be partially quantified using instruments, the integrative clinical judgment cannot be replicated by current measurement approaches. Students learn by placing their fingers on the master's wrist while being diagnosed simultaneously — direct somatic transmission. There is no other way. In Western medicine, the parallel loss is the erosion of continuity of care: the GP who has followed a patient for a decade carries knowledge that no chart contains, and that no specialist encounter can develop.
In agriculture, traditional farmers read soil health by hand — feel, smell, visual appearance. Experienced practitioners can assess drainage, biological activity, and approximate fertility in ways that anticipate problems before laboratory tests would even be ordered. Varietal selection required reading plants across a full growing season. The Mesoamerican milpa system — corn, beans, and squash interplanted in a specific configuration — is well documented as a sophisticated ecological system. The names are known. The practice — the timing, spacing, variety selection, and reading of the growing system in a specific place — required generations to develop and, once abandoned, requires something close to those generations to redevelop.
In software, tacit knowledge loss has a specific name: the bus factor. A 2024 paper on "knowledge islands" found that 10 to 15 percent of developers typically hold disproportionate knowledge of critical system components. What is irreplaceable is not code — code can be read. What is irreplaceable is architectural intent: why the system was designed this way, what trade-offs were made, what was tried and rejected. The collapse of junior developer pipelines after 2021 broke the primary transmission mechanism: osmosis through physical proximity to senior developers. When junior developers were present in the same space as seniors, they overheard debugging sessions, participated in informal reviews, absorbed tacit knowledge about how a system behaves under pressure. That transmission no longer reliably happens.
In nuclear weapons manufacturing, the Fogbank case is not isolated. GAO reports found that approximately 35-37 percent of critically-skilled nuclear production personnel were at or near retirement age — a standing institutional crisis, not a one-time event. Boeing in 2017 was forced to rehire hundreds of retirees to maintain commercial jet production — workers who held tacit knowledge of specific manufacturing processes that had not been successfully transferred to successors. The knowledge had walked out with the retirements.
The Political Dimension
Ivan Illich observed in 1973 what happens when tacit knowledge becomes professionalized. Two things occur simultaneously. The credential system takes over the transmission function — replacing tacit-knowledge-transmitting apprenticeship with explicit-knowledge-testing examinations. And citizens are deskilled: they lose the ability to do for themselves what they now depend on credentialed professionals to do.
This is not just a complaint about quality. It is structural. Because tacit knowledge requires practice to maintain, any system that removes practice from citizens and gives it to professionals necessarily degrades citizens' tacit knowledge. There is no workaround. The loss is proportional to the degree of professionalization. The doctor does not just know more than the patient — the patient has been removed from the context in which health knowledge is built. The farmer does not just lack specific technical knowledge — they have been separated from the practice through which ecological knowledge accumulates across generations.
Illich called the end state counterproductivity: institutions past a certain scale produce the opposite of their stated purpose. Medicine produces iatrogenic harm. Schools produce less genuine learning than self-directed practice would. The mechanism each time is the same: growth requires deskilling the population being served. Otherwise the population doesn't need the institution.
What Recovery Requires
Japan's Living National Treasures program designates specific individuals as holders of irreplaceable traditional techniques — lacquerwork, ceramics, textile dyeing, wooden joinery — and pays them 2 million yen per year specifically to maintain active transmission relationships. The payment is not for the product. It is for the teaching. The explicit acknowledgment: this knowledge is in people, not in documents, and the people must be preserved in active practice.
Traditional Japanese joinery contains more than 200 distinct kumiki joint types. UNESCO added traditional Japanese architectural craftsmanship to its Representative List of Intangible Cultural Heritage of Humanity in 2020. The crisis is not that the specifications have been lost. It is that the practitioners who can execute them, and teach their execution, are dying without successors.
The Dreyfus framework is unambiguous on what recovery requires. You cannot rebuild expertise by writing better documentation, running better workshops, or developing more sophisticated onboarding programs. These are novice-to-advanced-beginner tools. They move people from context-free rule-following to beginning pattern recognition. They do not produce proficiency. Proficiency requires years of practice in the presence of practitioners who already hold it — feedback loops from people who can recognize quality through its products, because they themselves hold the tacit knowledge required to produce quality.
Toyota understood this. Nonaka and Takeuchi's study of Japanese corporate knowledge creation found that Toyota's production system could not be transferred by giving another company Toyota's documentation. It was transferred through long-term physical presence and apprenticeship-like supplier relationships — not document transfer. American competitors read Toyota's publications and still could not replicate what they described. The knowledge was not in the publications.
What Polanyi identified in 1966 was not a temporary problem to be solved by better knowledge management. It is structural. Explicit knowledge is not self-sufficient — it is always parasitic on tacit knowledge for its application. Every rule requires an interpretation, and that interpretation requires practice, context, and embodied understanding to apply correctly.
We built systems on the assumption that what matters can be written down. The failures are everywhere now: nuclear weapons we temporarily couldn't refurbish, concrete formulas we can analyze but not replicate, farming systems we can describe but not rebuild, buildings we can photograph but not feel. Each case tells the same story.
The knowledge was never in the documents. It was in the practice. When the practice stopped, the knowledge went with it.