The Knowledge That Cannot Be Written Down

Much of what we treat as knowledge can be stated, stored, and transmitted. Propositions, procedures, datasets, and formal models belong to this explicit realm. They can be written down, indexed, searched, and, increasingly, generated. Large language models and related systems have become extraordinarily fluent in this domain. They can recombine, paraphrase, and extend explicit knowledge at a scale and speed no individual human can match.

Yet a substantial portion of human competence has always lived outside this realm. It is the knowledge that shows itself in skilled performance, in the timing of a decision, in the sense that something is off even when the formal criteria are met. This is tacit knowledge: the understanding that is difficult or impossible to fully articulate, yet is essential to expert judgment and effective action.

The distinction is not new. Michael Polanyi observed decades ago that we know more than we can tell. A physician recognizes a pattern of symptoms before being able to list the diagnostic criteria. A craftsperson feels the resistance of material and adjusts pressure without consulting a rule. A scientist develops a sense for which anomalies are worth pursuing and which are noise. These capacities are real, consequential, and stubbornly resistant to complete verbalization.

Explicit Knowledge and Its New Abundance

The recent expansion of generative systems has made explicit knowledge dramatically more available. Technical documentation, historical records, scientific literature, and practical advice can be summarized, cross-referenced, and applied with minimal friction. The cost of retrieving and recombining stated information has fallen sharply. This is a genuine advance. Many tasks that once required laborious search or specialized training can now be performed more quickly and by a wider range of people.

The danger lies in treating this abundance as comprehensive. When fluent answers are always at hand, it becomes easier to overlook the residual forms of understanding that never fully entered the written record. Expertise begins to look like access to better prompts rather than the slow accumulation of judgment under conditions of incomplete information. The social prestige once attached to hard-won skill can shift toward those who navigate the explicit layer most effectively.

This shift is not inevitable, but it is already visible in how organizations allocate attention and reward. Systems that surface and manipulate explicit knowledge scale more readily than systems that cultivate tacit competence. The latter still requires time, feedback from reality, and often apprenticeship. It does not compress as easily.

Where Tacit Knowledge Continues to Matter

Tacit understanding is not mystical. It arises from repeated engagement with a domain under conditions that force adaptation. The body and the mind together form expectations about how a situation will unfold. When those expectations are violated, attention sharpens. Over time, the practitioner develops a repertoire of responses that feel immediate rather than deliberative. The knowledge is carried in the pattern of attention and the readiness to act, not primarily in a set of propositions that could be listed in advance.

This form of knowledge is especially important in environments marked by ambiguity, novelty, or high stakes. Clinical diagnosis under incomplete data, negotiation under shifting incentives, scientific inquiry at the edge of established theory, and leadership during crisis all rely heavily on capacities that resist full codification. Rules and models remain useful, but they do not substitute for the ability to read a situation and adjust in real time.

Human–machine collaboration therefore requires clarity about what each party contributes. Machines can rapidly supply relevant explicit knowledge, surface patterns in large datasets, and generate candidate options. Humans remain responsible for the integration of that material with contextual understanding that has not been fully formalized, for the weighting of competing considerations that cannot be reduced to a single objective, and for the ownership of consequences. Treating the machine’s fluency as equivalent to the full range of required understanding collapses this division of labor and invites overconfidence.

The Risk of Premature Formalization

There is a recurring temptation to convert tacit knowledge into explicit form as quickly as possible. Capture the expert’s heuristics, encode the decision tree, train the model on the resulting labels. In some domains this works well enough. In others it produces brittle systems that perform adequately within the training distribution and fail when the environment shifts in ways the original experts would have sensed.

The problem is not that formalization is impossible. It is that formalization is almost always incomplete. What gets written down is a projection of the tacit competence under the conditions present at the time of elicitation. The unstated background assumptions, the feel for when a rule should be bent, and the capacity to improvise when the projection fails remain with the human. Organizations that treat the formalization as a complete substitute gradually lose the ability to notice when the formal system is drifting from reality.

This dynamic has implications for how we train people. If the primary educational goal becomes the efficient retrieval and recombination of explicit knowledge, the slower processes that build tacit competence receive less time and fewer resources. The result is a workforce that can operate fluently within existing frameworks yet struggles when frameworks themselves must be revised.

Preserving the Conditions for Tacit Development

Tacit knowledge does not arise automatically from exposure to information. It requires sustained practice in environments that provide feedback, variation, and consequences. Apprenticeship models, deliberate practice with expert coaching, and the willingness to tolerate temporary underperformance while skill consolidates all remain relevant. Digital tools can support these processes—by providing simulations, rapid feedback, or access to remote mentors—but they do not replace the need for real engagement with the domain.

In research and professional settings, this suggests a deliberate division of attention. Use machines to expand the explicit base and to test ideas against large corpora. Reserve human effort for the cultivation of judgment under uncertainty, for the development of taste in a domain, and for the maintenance of shared practices that transmit what cannot be fully written down. The two efforts are complementary rather than competitive, provided their different characters are respected.

There is also a cultural dimension. Societies that treat only the measurable and the explicit as real knowledge tend to undervalue the slower forms of competence. Status, funding, and institutional design follow the measurable. Over time this can erode the very capacities that allow institutions to adapt when measurement frameworks themselves become inadequate.

Conclusion

The expansion of machine fluency in the explicit domain is a powerful development. It removes certain frictions and democratizes access to large bodies of stated knowledge. It does not, however, abolish the residual domain of tacit understanding. That domain continues to underwrite skilled performance, contextual judgment, and the capacity to revise frameworks when they no longer fit.

The practical task is not to deny the power of explicit systems, nor to romanticize the ineffable. It is to maintain institutional and personal conditions under which tacit competence can still form, and to design collaboration so that each form of knowledge is applied where it is strongest. The knowledge that cannot be written down remains indispensable precisely because the world continues to exceed what any formalization can fully capture.