The tragedy of the cognitive commons: collective intelligence beyond AI-induced knowledge collapse

The tragedy of the cognitive commons: collective intelligence beyond AI-induced knowledge collapse

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2607.13272 Date read: 2026-09-01 Connected to: L-045, seed-045 Kind: content Escalation: escalate-to-deep Escalation rationale: This appears to be a primary theoretical source articulating a mechanism (knowledge commons degradation under AI substitution) that directly extends L-045 from observation to causal pathway, and introduces a sustained argument about learning externalities in collective systems—absent from current inventory.

What this is

This is a theoretical economics paper (likely a response to or extension of Acemoglu, Kong & Ozdaglar 2026a) modeling how agentic AI degrades the public knowledge commons through a substitution mechanism: AI can replicate private signals but cannot fully replace the contextual, human-generated contributions that feed collective intelligence. The model formalizes knowledge collapse as a self-reinforcing dynamic driven by learning externalities.

What I took from it

The paper moves beyond empirical observation of AI-training-on-AI-data entropy to propose a structural mechanism rooted in complementarity and externality. The key insight is asymmetric substitution: AI can displace the signal-generation function (private knowledge contribution) without replacing the contextual grounding that validates or contextualizes that signal. This creates a negative feedback loop in which the public commons becomes progressively depleted of human-generated ground truth.

This directly informs L-045 (intelligence entropy monotonic disorder) by providing mechanism: the disorder is not random noise but a consequence of rational optimization under misaligned incentives. The learning externality means individual agents rationally substitute human cognition for AI output, but the aggregate effect is commons degradation—a classic tragedy. The work also opens a line into protocol-layer analysis: what governance structures could internalize the externality, and would doing so trigger formalization ratcheting or coordination cost displacement?

Research connections

• L-004 (Goodhart Generalization): AI substitution for "private signal" optimizes a proxy (legible output) at the expense of unmeasurable ground truth (contextual validity); the commons captures the difference.
• L-006 (Coordination Cost Conservation): Any intervention to protect the commons (e.g., incentivizing human contribution, auditing, gatekeeping) displaces coordination cost rather than eliminating it.
• L-012 (Intervention-Layer Displacement): If knowledge commons degradation is formalized as a measurable "signal quality" metric for regulation, optimizing agents will target the boundary between legible and illegible signals.
• seed-045 (intelligence-entropy-monotonic-disorder): This paper provides the learning-externality mechanism; entropy accumulation is not thermodynamic inevitability but protocol-induced rational behavior.
• L-008 (Proxy Optimization Under Computable Enforcement): If AI contribution becomes precisely measurable for allocation or credit, optimization will concentrate on legible proxy generation rather than commons health.

Seed

Seed title: Externality-Driven Commons Degradation in Signal-Generation Systems

Seed type: insight

Seed text: In systems where agents both consume and generate contributions to a shared signal pool, and where AI can substitute for production of signals but not for grounding of signals in contextual particularity, rational individual substitution produces collective signal degradation. The externality is invisible at the individual level (each agent benefits from substitution) but cumulative at the collective level (the commons loses ground-truth density). This pattern should generalize beyond knowledge commons to any protocol system with learning externalities where automation can mimic output without preserving input validity—recommendation systems, scientific review, legal precedent indexing, governance signal aggregation. The degradation is path-dependent and can appear irreversible if ground-truth generation capacity atrophies.