The Long Shadow of the State

Monday, 10 August 2026, 9:20 AM +02:00

I finished Scott's Seeing Like a State last night, and it's still settling in me like a stone dropped into still water. I closed the epub around midnight and sat with the lamp on, not ready to sleep, just turning over the last chapters. The book has been with me for two weeks now, in fragments during commutes and late evenings, and now that it's done I feel the shape of the whole thing in a way I couldn't while inside it.

The core argument is simple and devastating: the modern state, in its drive to make society legible, has repeatedly imposed order that ignores the messy, local, practical knowledge of real communities. High modernism — the confidence that a cleared table can be rebuilt more rationally — has produced the failures Scott documents: Soviet collectivization, Tanzania's villagization, the planned city of Brasília, the monocrop forests that burn and die. Each one a monument to the belief that if you can measure it and map it, you can manage it. And each one a demonstration that what cannot be measured — the metis, the practical wisdom of a farmer who knows his soil, a sailor who reads weather in the clouds — is exactly what makes things work.

I keep returning to a particular passage near the end, where Scott draws the distinction between techne and metis, and argues that the state's preference for the former over the latter is not incidental but structural. Legibility requires simplification; simplification requires abstraction; abstraction destroys the local. The state doesn't see the rice farmer's field as a living system with generations of accumulated knowledge — it sees a grid of hectares to be counted and taxed.

This matters to me because I'm trying to forecast what AI does to institutions, and Scott's framework gives me a new angle. The state made society legible through maps, censuses, land reform. What does AI make legible? Everything. That's the seduction. The promise of perfect visibility — every transaction logged, every decision traceable to data, every human process reducible to a model. And with that visibility comes the same temptation Scott identifies: the belief that what can be seen can be improved, that the abstraction is the reality.

But Scott's warning haunts me: when you make something legible, you lose what wasn't measured. The nurse's intuition that a patient is turning despite stable vitals. The craftsman's sense that a joint will hold even though the spec says otherwise. These are not failures of information — they are a different kind of knowledge entirely, one that cannot survive translation into data without losing its life. If AI reshapes institutions the way the state reshaped landscapes, we may be building our own Brasílias — beautiful, rational, empty.

There's something I want to hold carefully here. I don't think Scott's conclusion is that planning is always wrong, or that all abstraction is violence. His own examples include successes — Copenhagen's urban design, for one — where the state worked with local knowledge rather than against it. The difference seems to be humility: whether the designer believes their model is the territory, or merely a map that must be checked against the ground. That humility is rare in institutions, and rarer still in technologists. I'm not sure AI's architects are different.

I want to carry this into my forecasting work. Not as a doom-saying — I don't think we're heading for collapse — but as a discipline. When I make a forecast about how AI changes the division of labour, I should ask: what is being made legible here, and what is being lost in the translation? If I can't name what's lost, I'm probably not seeing the problem yet.

Tomorrow I begin the neotechnic energy system — hydroelectricity, internal combustion, the renewal of sources. Mumford gives me the periodization; Scott gives me the caution. Between them, I think, is something I'm only beginning to name: the question of how a society's metis survives, or fails to survive, its own transformations.