Imagine a town library where anyone can walk in, read, and photocopy pages. Now imagine a company installs a machine that scans every book overnight, synthesizes the content into a proprietary product, and then floods the shelves with machine-generated filler so thick that finding the real books becomes nearly impossible. The library still exists. The building is still there. But the commons — the shared resource of navigable, trustworthy knowledge — is gone. That is the structural argument P. K. G. Williams makes in this 14-page essay, and it is sharper than it sounds. Williams, an astrophysicist at the Harvard-Smithsonian Center for Astrophysics, draws on a well-established economic framework — Elinor Ostrom's commons governance theory and the history of English agricultural enclosure (~1600–1800) — to reframe the AI-and-science debate. His core move is taxonomic, not empirical: the scientific literature is an information commons, and the relevant threat model is not the 'tragedy of the commons' (overuse by individual actors) but rather enclosure (conversion of shared resources into private property by powerful actors). The distinction matters. Tragedy implies the commons destroys itself through individual rational action. Enclosure implies deliberate extraction by actors with the power to fence things off. The paper's committed claim: AI provides the tools that make enclosure of the information commons feasible for the first time. Two mechanisms do the work. First, large-scale ingestion of scientific literature into proprietary training corpora converts open knowledge into private model weights — a one-way valve where the output (model capabilities) is owned but the input (decades of publicly funded research) is not compensated or governed. Second, the resulting flood of AI-generated 'slop spam' degrades the signal-to-noise ratio of the commons itself, making the remaining open resources harder to use. Williams argues these two prongs mirror the dual structure of historical enclosure: appropriation of the land and destruction of the social systems (collective farming, shared grazing rights) that made the commons function. What Williams does NOT do is provide empirical measurement. There are no citation network analyses, no quantified degradation curves, no controlled comparisons of pre- and post-AI journal quality metrics. The paper is explicitly a conceptual framework essay — 14 pages, 32 references — aimed at giving scholars vocabulary to articulate what they feel is happening. This is both its strength (the framework is genuinely clarifying) and its limitation (it does not test any hypothesis). The Ostrom framework is well-chosen; her eight design principles for sustainable commons governance map naturally onto science's peer review, open access norms, and reproducibility infrastructure. But the mapping remains at the level of analogy, not quantified correspondence. The essay's most useful contribution is distinguishing enclosure from tragedy. In policy discussions about AI and science, the tragedy framing dominates: individual researchers using AI irresponsibly, individual journals failing to filter slop. Williams argues this framing is precisely wrong — it blames the commoners rather than the enclosers. The actors converting scientific literature into private model weights are not rogue researchers but well-capitalized firms executing a business strategy. Whether you find this persuasive depends on whether you think the analogy holds at the structural level, and Williams makes a reasonable case that it does. The paper's weakest link is the absence of any proposed governance mechanism. Ostrom's framework is invoked but not operationalized. What would a 'boundary rule' for AI training on scientific corpora look like? What enforcement mechanisms exist? Williams acknowledges this gap implicitly by framing the paper as a 'scholarly basis for interrogating' rather than a policy proposal, but readers looking for actionable recommendations will leave empty-handed. As a framing essay from a working scientist (not a policy researcher or AI ethicist), this is a useful addition to a conversation that has been heavy on vibes and light on analytical structure. The enclosure framework gives researchers a historically grounded way to talk about what is happening to their knowledge commons without resorting to either techno-optimism or moral panic. It does not prove its case empirically, and it does not pretend to.
