The Knowledge That Lives in Struggle

In the AI Story

Declarative knowledge is what AI handles superbly. The capital of France is Paris. The time complexity of a binary search is logarithmic. The mechanism of action of beta-blockers is receptor antagonism. These facts can be retrieved, verified, communicated, and transferred directly between knowing agents. But declarative knowledge is not what separates the master from the competent practitioner. What separates them is the elaborate, flexible, context-sensitive procedural knowledge that allows the master to perceive patterns invisible to others, to anticipate outcomes before they manifest, to make judgments that appear intuitive but are the output of an enormously sophisticated pattern-matching architecture.

The apprenticeship traditions across every culture and craft have converged on the same insight: the apprentice must do the work, not merely watch the master. The watchmaker's apprentice does not learn to repair watches by observing the master's hands; she learns by repairing watches badly and feeling, through thousands of fumbled attempts, the difference between a properly tensioned spring and one slightly off. The medical resident does not learn to diagnose by reading textbooks; she learns by encountering patients whose presentations defy textbook patterns and developing, through hundreds of diagnostic failures, the pattern-recognition capacity that eventually allows detection of diseases textbooks cannot adequately describe.

The AI displacement of this mechanism is precise. When Claude handles the configuration problems, the developer does not encounter the unexpected errors that would have forced understanding of connections between systems. The ten minutes of genuine learning embedded in four hours of tedium disappear along with the three hours and fifty minutes the practitioner is glad to lose. The loss is invisible in the short term because the output is unchanged; it manifests months or years later as a hollowness in judgment whose origin the practitioner cannot reconstruct. This is the pattern Edo Segal documented in his Trivandrum team and the pattern the Hosanagar endoscopist data confirms at clinical scale.

The deeper point concerns variation. Expert mental representations are transferable because they are built through systematic variation — encountering the same deep principles across many surface configurations. The developer who debugs configuration errors across fifty projects builds a representation abstract enough to apply to novel configurations. The developer who describes problems to Claude and receives solutions experiences uniform interactions without the variation that abstraction requires. Even when the outputs are varied, the cognitive engagement is not. The knowledge that would have lived in the struggle does not form, because the struggle has been delegated.

Origin

The claim that some knowledge is available only through personal engagement has deep roots in the philosophical tradition — Michael Polanyi's 'tacit knowing,' Gilbert Ryle's 'knowing how' versus 'knowing that,' the Aristotelian distinction between techne and phronesis. Ericsson's contribution was to give this philosophical insight an empirical foundation in the psychology of expertise, showing how the procedural knowledge distinguishing experts is built through the specific conditions of deliberate practice.

The AI-specific application of the insight is contemporary, emerging as researchers document what happens when the conditions for building this knowledge are removed at scale by tools that handle the difficult work without requiring understanding from the user.

Key Ideas

Non-propositional form. The knowledge is not stored as statements that can be transmitted; it is encoded in cognitive structures that must be built from the inside.

Variation-dependent construction. Transferable knowledge requires systematic variation across surface forms; uniform AI-mediated interactions foreclose the abstraction variation enables.

Apprenticeship convergence. Traditions across unrelated cultures have independently arrived at the insight that doing is necessary, watching is insufficient — convergent evidence that the insight is structural rather than cultural.

Invisible loss. The disappearance of this knowledge produces no visible signs until situations demand it; the output quality of AI-assisted work conceals the developmental deficit.

Not romanticism but mechanism. The claim is not nostalgia for difficulty but a precise specification of how a particular cognitive architecture is constructed — an architecture that matters when situations exceed the tool's competence.