And What It Means for the Algebra of Change
Every case in the preceding essay shared a constraint none of its protagonists could see. The Chagga built their furrows over five centuries; Sawadogo’s zaï spread over four decades. The algebra worked slowly because the minimum viable act required a human body, human time, and knowledge transmitted person to person. We are now watching that constraint dissolve.
The cost of intelligence, the marginal cost of reasoning, diagnosing, teaching, designing, is collapsing without precedent. This is not a better tool. It is a change in the nature of a resource.
Expertise has always been rivalrous: a doctor treating one patient cannot treat another simultaneously. A trained AI model is non-rivalrous, copyable at near-zero cost, usable by one person without depriving another. But intelligence as an economic good splits into two layers: a non-rivalrous core, the weights, wrapped in a rivalrous substrate of compute and energy. Everything here turns on that split, because enclosure does not need to capture the core. It only needs to own the substrate the core runs on.
The most durable transformations in the preceding essay were not programmes but conditions; systems that reproduced themselves because every participant held both the incentive and the knowledge to maintain them. The mfongo worked for five centuries because every family that depended on the water also held the knowledge to maintain it. AI, as currently structured, is the opposite. Frontier models are held by four or five companies, trained on data extracted on terms still being contested, and licensed rather than owned — revocable by terms of service, answerable to shareholders. A village doctor cannot be switched off. A subscription can be.
This creates the problem of artificial scarcity: institutions built to manage genuine scarcity manufacturing scarcity where the technology has eliminated it, IP regimes, compliance costs only incumbents can absorb, subscriptions calibrated to extract maximum surplus. Each is individually defensible. Their aggregate effect rebuilds scarcity in the one domain where it was just abolished.
Open-weight models are the counter-pressure, but the internet is the precedent, and it is not reassuring: its protocols never closed, yet enclosure happened in the layers above them : search, social platforms, marketplaces. Open weights may persist the same way, while enclosure forms beneath them in chips and energy, and above them in the distribution surfaces through which people actually reach the capability. An open model is necessary for a commons. It is nowhere near sufficient.
Of the four conditions for durable change, a specific privation, a minimum viable act, a network effect, sufficient time, AI changes only one. It compresses time: a network effect that took the zaï forty years of farmer-to-farmer teaching could happen in a season. It does not change the need for a specific privation; the technology has no mechanism for identifying the right problem, only for optimising whichever one it’s given, efficiently, even if it’s the wrong one. And it does not, on its own, create the condition. A solution delivered without transferring the understanding of how it works produces dependency, not capacity. A programme that uses AI is still a programme.
There is a version of this transition that reproduces every failure mode of development orthodoxy: not the East India Company accumulating a subcontinent through small decisions, but three or four companies accumulating a civilisation’s cognitive infrastructure through licensing agreements, the dependency accumulating quietly behind a freemium model that feels free for exactly as long as serves the acquisition strategy.
There is another version: open models adapted to a single district — its crops, its disease burden, its language — by the institution that already serves it, the adaptation documented plainly enough that the next district copies the method rather than licenses the product. Small enough for an agricultural university or a health directorate to complete. Legible within a season. Replicable with what the next institution already has. And it transfers understanding, not just the answer — the test that separates a condition from a programme. This is the zaï pit of the new terrain. Whether anyone digs it is the institutional choice the two versions divide on.
The algebra has not changed — only the speed at which it runs. The question of who is doing the arithmetic used to wait decades for an answer. It will not wait now.