← Back to feed

A paper posted to arxiv (2607.29380) argues that AI is not just displacing individual workers but disrupting the mechanism by which professional expertise reproduces itself across generations. The authors call the dynamic the Cognitive Commons problem.

The core claim: every individual firm has an incentive to cut entry-level roles, because AI can handle those tasks more cheaply. But the profession-wide pool of deep expertise — the senior practitioners who catch what AI gets wrong — depends on a continuous pipeline of junior workers who develop judgment through doing the work, not just reviewing AI output. When every firm cuts the entry-level jobs simultaneously, nobody refills the pipeline. The expertise pool deteriorates, but no single firm bears the cost alone or has the incentive to fix it unilaterally.

Two Failure Modes

The paper identifies two distinct ways the regeneration pipeline breaks down.

The first is direct elimination: AI replaces entry-level roles outright. Junior analysts, junior lawyers, junior engineers, junior medical coders. The work that used to develop domain judgment no longer requires a human trainee.

The second is more subtle. Junior workers remain employed but produce outputs they cannot fully validate, because AI handles the cognitive generation while the human provides review. If the review does not require genuine domain understanding to perform, the learning that builds expertise does not happen. The output looks fine. The practitioner does not develop the judgment to know when it is not fine.

The paper calls the corrective capacity “the Validation Tether”: the Internalized Mastery required to catch plausible but substantively wrong AI outputs. Without that mastery, AI error propagates through systems staffed by people who are not equipped to detect it.

The Employment Data

Early labour market data fits the concern. In highly AI-exposed occupations, workers aged 22-25 saw a 16% relative employment decline from October 2022 to September 2025. Workers aged 35-49 in the same occupations grew employment by more than 8% over the same period.

The divergence is directionally consistent with a market pruning its junior layer while retaining and promoting senior workers who have already accumulated the judgment AI cannot replicate. It does not prove the Cognitive Commons mechanism — causality is harder to establish than correlation — but the demographic pattern is what the theory predicts.

Why No Market Fix

Standard economic intuition says that if expertise becomes scarce, its price rises, which eventually induces investment in producing more of it. The paper argues this fails for the Cognitive Commons.

The expertise being eroded is a public good: any firm benefits from practising in an industry full of senior people who can evaluate AI output and train one another. But the investment in producing that expertise — keeping junior roles that develop judgment — is borne by individual firms. When every firm cuts simultaneously, the public good is undersupplied, and no individual firm can solve it by acting differently.

Professions may partly fill the gap through formal training programs, apprenticeships, or certification regimes that substitute for on-the-job learning. But the paper notes that most professional expertise historically required the cognitive struggle of doing real work under expert supervision, not just academic instruction. Whether that struggle can be replicated in classrooms and simulations, rather than in offices, is an open empirical question.

The Compounding Risk

The concern compounds over time. The senior practitioners who currently serve as the validation layer were trained in a pre-AI environment where junior roles were abundant. As they retire, the cohort behind them — trained in an environment where AI did the junior work — may lack the Internalized Mastery their predecessors had, even if they have equivalent credentials.

The paper does not prescribe a solution but frames the problem as a coordination failure, which implies the fix, if any, is likely regulatory, professional, or institutional — not something individual firms will arrive at through self-interest alone.