AI gravity and socio-economic operating systems

Gravity is the right metaphor. AI does not push economies; it bends them. Capital, talent, attention all curve toward the densest concentrations of compute. The curvature is not malicious. It is geometric. A region with a hyperscale cluster exerts a pull on the surrounding labour market that a region without one cannot match, no matter how generous its tax incentives.

A socio-economic operating system that does not account for that curvature will mistake drift for choice. Policymakers will look at the migration of their best engineers and conclude that wages are too low, when the actual force is gravitational. They will design retention schemes that fail because the schemes assume a flat space, and the space is not flat.

The first task of a serious AI policy is therefore astronomical. You have to map the curvature before you try to navigate it. Where are the dense bodies? In which directions are they pulling? What orbits are stable, and which ones will decay? The Caribbean, for example, sits in the gravitational field of two large bodies — the North American compute belt and the emerging Brazilian one. The orbits that are stable for us are not the orbits that are stable for, say, Singapore.

The Caribbean has lived inside gravitational fields it did not author before. The plantation economy was a gravitational field. The IMF structural adjustment programmes were a gravitational field. The pedagogy of that survival — the West Indian habit of finding agency inside conditions one did not set — is now globally relevant. The rest of the world is about to learn what it is to be a small body in someone else's solar system.

The future is still about figuring out what questions to ask. This is more important than finding the correct answer to the wrong one. A government that asks how to attract a hyperscale data centre is asking a 2018 question. A government that asks what kind of cognitive sovereignty its citizens will have when the inference layer is owned offshore is asking a 2026 question. The difference between the two questions is the difference between renting and owning the operating system you live inside.

There is a temptation, particularly among smaller states, to declare neutrality. To say: we will use whichever model is best, from whichever provider is cheapest. That is a coherent posture for cutlery. It is not a coherent posture for cognition. The model you use shapes the questions you can ask, and the questions you can ask shape the future you can imagine. Neutrality, in this domain, is a slow form of capture.

What would a non-neutral posture look like? It would look like a regional commons — shared compute, shared evaluation, shared refusal. It would look like a curriculum that taught children to interrogate models in the way an earlier curriculum taught them to interrogate texts. It would look like a procurement policy that paid a premium for systems whose weights could be audited. None of these are cheap. All of them are cheaper than the alternative.