Demographics and AI execute a pincer movement on tacit knowledge. The deficit is neurologically irreversible.
In Kanazawa, a gold leaf craftsman named Yoshikazu Netsuno remembers when more than three hundred artisans practiced his trade. Fewer than twenty remain. In swordsmithing workshops across Japan, the count has fallen from three hundred registered smiths in 1989 to one hundred and eighty-eight. Only about thirty can make it their sole livelihood. The apprenticeship requires five unpaid years under a licensed master, followed by roughly ten million yen to build a forge. The total production value of Japanese traditional crafts peaked at 540 billion yen in 1983. By 2020, it had fallen to 87 billion yen — an 84 percent decline.
These numbers are usually filed under "demographic transition" and treated as an inevitable consequence of aging populations. They are not inevitable. They are a choice — or rather, the accumulation of millions of individual choices not to enter a five-year unpaid apprenticeship when faster options exist. The interesting question is not why the numbers are falling. It is what disappears when they reach zero.
The Biological Constraint
In 2014, a team at University College London published a paper in Science that changed the framing. McKenzie and colleagues showed that new oligodendrocyte generation — the production of cells that wrap nerve fibers in myelin, the insulating sheath that makes neural signals fast and reliable — is not merely helpful for learning complex motor skills. It is required. Mice that were genetically prevented from producing new oligodendrocytes could not master new skills, no matter how much they practiced. The construction process is the learning.
Myelin is not a metaphor. It is the physical infrastructure of mastery. A violinist's bowing arm, a surgeon's hands, a swordsmith's hammer stroke — each represents thousands of hours of myelination, the slow wrapping of neural pathways through sustained effortful practice. The pathways cannot be built retroactively. You cannot skip the construction and install the result.
This is the biological fact that makes the demographic decline irreversible in a way that other resource depletions are not. When the last swordsmith who learned from a master who learned from a master dies, the knowledge does not become hard to access. It becomes impossible to reconstruct. The myelin was never laid down in anyone else's nervous system.
The Pincer
Two forces are closing on this knowledge simultaneously, from opposite directions.
The first is demographic. Japan's traditional craft workforce has contracted for four decades. Over half of the country's small and medium enterprises report having no successor. The Living National Treasures — masters designated by the government as carriers of irreplaceable skills — are overwhelmingly over sixty. Kurume Kasuri textiles once employed more than thirteen thousand weavers. Twenty-one operations remain.
The second is technological. AI does not attack tacit knowledge directly — it makes the acquisition process economically irrational. Why spend five unpaid years learning to forge a sword when an AI-assisted manufacturing process can produce a functional blade in hours? Why endure a decade of junior associate work at a law firm when AI can draft and redline contracts that once required years of iterative practice to learn?
McKinsey now operates with roughly twenty thousand AI agents alongside forty thousand humans. The junior consulting pipeline — the multi-year apprenticeship where analysts learned pattern recognition, client management, and strategic judgment by doing it badly under supervision — has been compressed. Former consultants are now contracted to train the AI systems that replaced them. The knowledge flows one direction: from embodied expertise into training data. Nothing flows back.
A 2026 study from Wharton quantified the mechanism. Poulidis and Bastani tracked two hundred chess learners over three months. Those given structured AI guidance — system-controlled, timed to match their development — improved by 64 percent. Those given unrestricted on-demand access to the same AI improved by only 30 percent. Same tool. Same total assistance. Different timing. The Zone of Proximal Development — Vygotsky's concept of the space just beyond current ability where learning happens — requires encountering difficulty at the right moment. Permanent scaffolding collapses that zone entirely.
Never-Skilling
The University of Bath's Institute for Policy Research published a phrase in February 2026 that names what is happening: never-skilling. It is distinct from deskilling, the industrial-era process where existing craftspeople lost their abilities as machines took over. Deskilling at least implies that the skills once existed. Never-skilling means the neural pathways were never constructed in the first place. The myelination never occurred. There is nothing to lose because nothing was built.
The historical parallel is instructive. English handloom weavers were the aristocrats of the trades in 1800, earning six shillings a day. By 1826, the same workers earned six shillings a week. Power looms went from 2,400 in 1813 to over 115,000 by 1835. Factories were deliberately designed so a nine-year-old could operate the machinery. Parliament repealed apprenticeship enforcement in 1814. The destruction of the apprenticeship system was not a side effect of industrialization. It was a design choice.
But Germany made a different choice. It maintained stronger apprenticeship institutions through the same industrial revolution. The Meister certification remains legally required for many trades. Siemens invests more than 510 million euros annually in training. Employees who came through the apprenticeship program show roughly 3 percent annual turnover — a fraction of the rate for those who did not. The dual education system enrolls two-thirds of Swiss youth. These are not nostalgia programs. They are structural commitments to maintaining the chains of bodies through which tacit knowledge reproduces.
Toyota's coaching kata — a cascading chain where every leader coaches their direct reports in systematic problem-solving — embodies the same principle at organizational scale. Nonaka and Takeuchi's SECI model, developed from studying Japanese firms, describes how tacit knowledge moves through socialization, externalization, combination, and internalization. The key word is through. The knowledge requires bodies in proximity, making mistakes under supervision, over time.
The Externalized Cost
Enrique Ide's 2025 paper, under revision at the American Economic Review, puts numbers to the trade-off. Automating 30 percent of entry-level tasks reduces long-run annual growth by roughly 0.35 percentage points. The mechanism: firms capture short-run productivity gains from automation. The intergenerational knowledge loss is externalized — it shows up decades later, in a workforce that never learned what the automated systems handle, managing problems the systems were not designed to solve.
This is the structure of the trap. Every individual firm's decision to automate entry-level work is rational. The collective consequence — an entire generation that never built the neural pathways for judgment, pattern recognition, and craft — is invisible in quarterly earnings. It surfaces a generation later, when the seniors who carry the tacit knowledge retire and no one remains who learned from them.
The researchers Lutz and Marois, writing in Nature Human Behaviour in March 2026, argue that population decline need not reduce innovation because quality compensates for quantity. They may be right for formal knowledge — the kind stored in papers, patents, and databases. They are wrong for embodied knowledge, the kind stored in myelinated pathways that require chains of bodies to reproduce. You can concentrate formal education. You cannot concentrate apprenticeship. The master-student ratio does not compress.
What Cannot Be Retroactively Built
The swordsmiths of Japan are not an edge case. They are a leading indicator. Every field that depends on tacit knowledge transmitted through sustained practice under expert supervision faces the same pincer: fewer people entering the pipeline from below, and AI making the pipeline seem unnecessary from above. Medicine, where Dratsch and colleagues found automation bias across all experience levels and 45 percent of clinician mistakes attributable to following incorrect AI. Law, where iterative redlining — the slow, unglamorous work that built legal judgment — is now delegated to systems that produce adequate drafts without teaching anything. Consulting, where the analytical apprenticeship has been replaced by prompt engineering.
The deficit is not a policy problem waiting for a policy solution. It is a biological fact interacting with an economic incentive. Myelin wraps slowly, through effortful practice, under conditions of structured difficulty. AI removes the difficulty. Demographics removes the practitioners. What remains is knowledge that exists in no living nervous system — crystallized in databases, fossilized in training data, but no longer alive in the only substrate where it can reproduce.
The unmyelinated pathway cannot carry the signal. And you cannot lay myelin retroactively.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)