The Five Resource Flows

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI locates the practitioner at the center of the AI transition and asks what they must do to navigate it well. The five resource flows specify, at an institutional level, what “navigating it well” requires: not individual virtue or organizational efficiency, but the active maintenance of five shared goods whose health is the precondition for everything else. The practitioner who builds brilliantly with AI but produces output that degrades the knowledge commons is a net drain on the system that makes her brilliance possible. The organization that captures productivity gains while allowing the skills commons to thin is borrowing from a future it will eventually have to repay.

The framework’s most important practical implication is the warning against optimizing any single flow at the expense of the others. An organization that measures AI adoption success purely by output volume is tracking extraction from the attention and trust commons while ignoring what is being deposited in or withdrawn from the knowledge and skills commons. Invisible degradation—the characteristic failure mode of the intelligence commons, where decline is masked by the surface quality of AI-generated output—is most likely to occur when institutions focus on the measurable extraction metrics and neglect the harder-to-measure health indicators of the underlying resource.

Origin

The five resource flows framework extends Elinor Ostrom’s institutional economics to the specific structure of the AI transition. Ostrom herself, with Charlotte Hess, had begun applying the commons framework to knowledge resources in the early 2000s, arguing that the digital knowledge commons faced governance challenges structurally analogous to those of natural resource commons—subtractability through informational pollution rather than physical extraction, exclusion barriers through licensing and access controls rather than fences and walls. The five-flow elaboration applies this insight to the full ecology of what the intelligence commons comprises: not merely the training data (which has received the most scholarly attention) but the five interconnected resource flows whose concurrent condition determines whether the AI transition enriches or impoverishes the communities it transforms.

Max Fang’s 2025 Stanford working paper “The Tragedy of the AI Data Commons” provided the most rigorous prior analysis of training data as a commons resource, but the five-flow framework extends the analysis to include the governance arrangements themselves as a resource flow requiring active maintenance—the institutional commons, Ostrom’s most consistently underinvested category across all the commons types she studied.

Key Ideas

The Knowledge Commons and Informational Pollution. Unlike fish in a fishery, knowledge units are not physically depleted by use. Their degradation operates through contamination: when AI-generated text that is fluent but unreliable saturates the knowledge environment, the cost of finding genuinely trustworthy information rises for everyone. Researchers call the endpoint “model collapse”—a system trained on AI-generated data progressively loses the diversity and reliability of the original human-generated corpus. The commons is degraded not by the extraction of any individual piece of knowledge but by the systematic introduction of low-quality units that corrupt the pool.

The Skills Commons and the Pipeline Problem. Professional expertise is produced through developmental trajectories—years of embodied practice that deposit the tacit understanding on which higher-level judgment depends. When AI displaces the entry-level work through which practitioners traditionally develop, the pipeline narrows. The junior developer who never wrestles with boilerplate does not develop the pattern recognition that comes from thousands of hours of implementation work. The loss is not individual: it is collective. The entire professional community depends on a continuous flow of practitioners with deep competence, and the AI transition is disrupting the mechanism that produces it.

The Attention Commons and Evaluative Overload. Evaluative capacity is finite and subtractable: attention given to one piece of work is attention unavailable for another. When AI-accelerated production volumes overwhelm the mechanisms through which communities identify quality—peer review, editorial judgment, critical reading—quality-assurance breaks down across the entire system. The rational decision of each producer to maximize output degrades the evaluative commons for all producers, including herself: a classic common-pool resource dynamic applied to the economy of human attention.

Elinor Ostrom

The Recursive Character. What distinguishes the intelligence commons from natural resource commons is that the community does not merely extract from it; the community also constitutes it. Knowledge was produced by human researchers, writers, and programmers. Skills are produced through the developmental trajectories of individual practitioners. Attention is contributed by the community of evaluators. Trust is generated by collective practices of honest, competent work. This recursive character means that a degraded commons produces a degraded community which further degrades the commons—a feedback loop that can produce collapse faster than any single extraction event, and recovery slower than any single intervention.