Cognitive Resource Depletion

In the AI Story

The resource being depleted is not knowledge of facts (which AI possesses in greater breadth than any human) or knowledge of procedures (which AI can execute more reliably than most practitioners). The resource is tacit knowledge — the embodied, pattern-recognitive, judgment-based capacity that decades of friction-rich practice deposit in a practitioner's nervous system. A senior software architect's ability to sense that a system design 'feels wrong' is the product of thousands of hours of debugging, troubleshooting, and recovering from failure; it cannot be articulated, transmitted through instruction, or acquired through shortcuts.

The mechanism of depletion is specific. AI tools, by removing the friction through which tacit knowledge forms, interrupt the deposition process. A junior developer who uses Claude Code to generate working software from natural language descriptions is producing output without undergoing the experience that would build the judgment to evaluate that output. Each individual instance is trivial — one developer, one project, one thin layer of tacit knowledge that did not form. But the instances are systemic, simultaneous, and compounding. Across the technology industry, across every knowledge profession AI is penetrating, the substitution is occurring at a scale and speed that has no historical precedent.

The depletion is self-reinforcing through the mentorship pipeline. Tacit knowledge is transmitted primarily through apprenticeship — close, sustained interaction between experienced and developing practitioners. When AI tools reduce the need for this interaction (the junior developer gets faster solutions from Claude than from a senior colleague), the mentorship interaction becomes harder to justify economically. The interaction that would have transmitted tacit knowledge from one generation to the next is optimized away, and the loss is invisible because output metrics improve.

The Easter Island analogy is precise. Each tree felled was individually rational. The collective accumulation destroyed the resource. And when the forest was gone, it did not regenerate — because the conditions for regeneration (seed stock, soil stability, protection from erosion) had been destroyed along with the trees. Each generation of practitioners trained in an AI-augmented environment has access to more powerful tools and less deep understanding than the generation before. The tools compensate for the missing understanding — until the problems exceed the tools' capability, at which point the compensation fails and the absence of tacit knowledge becomes consequential.

Origin

The concept emerged from the synthesis of Diamond's resource-depletion framework with contemporary research on expertise formation — notably K. Anders Ericsson's work on deliberate practice, Harry Collins's work on tacit knowledge, and Gary Klein's research on naturalistic decision-making. These traditions converge on the finding that expert judgment is built through friction-rich experience that cannot be shortcut.

The specific application to AI-driven expertise depletion is not Diamond's own but follows directly from his framework when applied to the documented mechanisms of expertise formation. The empirical grounding includes the Berkeley study of AI-augmented work (2026), the emerging literature on cognitive offloading and automation dependence, and the specific accounts of expertise atrophy emerging from AI-adopting organizations.

Key Ideas

Tacit knowledge is a resource that can be depleted. Like topsoil, forests, or aquifers, the stock of deep professional expertise in a society is finite, depletable, and requires active replenishment.

Replenishment requires friction. The specific developmental experiences that build tacit knowledge — debugging, troubleshooting, patient deduction — cannot be replaced by efficient alternatives; the friction is the mechanism.

Depletion is invisible at any single moment. No individual instance of AI-assisted shortcut constitutes depletion; the cumulative effect operates below the threshold of perception, as Diamond's creeping normalcy framework predicts.

The mentorship pipeline is particularly vulnerable. The intergenerational transmission of tacit knowledge depends on specific mentorship interactions that AI tools make economically harder to justify.

Threshold effects are probable. The system may appear to function normally until the routine breaks — the novel problem, the unprecedented situation — at which point the missing tacit knowledge becomes suddenly and consequentially absent.

Debates & Critiques

The concept is contested on multiple fronts. Skeptics argue that it is empirically unproven — that the decline in traditional expertise metrics (debugging hours, memorized algorithmic fluency) reflects obsolete skills being replaced by new ones rather than cumulative depletion. Defenders argue that the specific capacity at stake (judgment under novel conditions) is exactly the capacity that conventional metrics miss, and that the depletion will become visible only when conditions outside the training distribution arise. The debate mirrors Diamond's original methodological challenge: resource depletion is invisible until threshold effects occur, at which point corrective action may be too late. The appropriate institutional response depends on judgments under uncertainty about future conditions.