Polyculturalism
"A city is not a tree." — Christopher Alexander
My grandmother's garden in the countryside had no rows. Tomatoes grew into marigolds, dill seeded itself wherever it liked, and there was a gooseberry bush in the middle of everything that nobody had planted and nobody dared remove. It looked, to my child's eye, like a mistake. Like someone had failed at gardening.
Down the road there was a collective farm field: one crop, one height, one shade of green, stretching to the horizon like a well-made bed. That looked like success. That looked like agriculture.
It took me thirty years to understand that my grandmother was running a more sophisticated system than the collective farm. She just wasn't running it for the same thing.
What the field is optimized for
A monoculture is not a stupid idea. It is a brilliant idea, optimized for a specific set of constraints: mechanization, measurement, and markets. If you want a combine harvester to work, everything must ripen at the same height on the same day. If you want a futures contract, you need a commodity, which means fungibility, which means sameness. If you want a bureaucrat to know how much wheat exists in a province, you need fields that can be counted from a horse or a satellite.
James Scott called this legibility: the state's need to make the world simple enough to see. Monoculture is not primarily an agronomic strategy. It's an accounting strategy that happens to grow food.
The costs show up later, and elsewhere. A monoculture has no internal immune system, so you have to import one: pesticides, fungicides, herbicides. It exhausts specific nutrients, so you import those too. And most importantly, it can die all at once. Every Cavendish banana you have ever eaten is a clone of every other Cavendish banana. They are currently being killed, globally, by a single soil fungus called TR4. There is no genetic room to maneuver. The whole species is one long argument with one mushroom, and the mushroom is winning.
Polyculture works differently. When you plant corn, beans and squash together — the Mesoamerican "three sisters" — the corn gives the beans a trellis, the beans fix nitrogen for the corn, and the squash leaves shade the ground and suppress weeds. The land equivalent ratio of such a system is often greater than one: more total food per hectare than growing the three separately. But less of any one thing. And much harder to harvest, count, and sell.
Polyculture is more productive and less legible. That combination is fatal in an economy that pays for legibility.
Multiculturalism is a field of monocultures
Here is where the metaphor stops being a metaphor.
Multiculturalism, as we've mostly practiced it, is not polyculture. It's a field of small monocultures with tidy paths between them. Each culture is imagined as internally coherent, externally bounded, and best preserved unchanged. The Greek neighborhood, the Turkish neighborhood, the folklore festival where everyone wears the costume of their assigned ancestors and dances the assigned dance. Diversity as a mosaic: many tiles, each one solid, none of them bleeding into the others.
It is a generous vision and it fails for the same reason the collective farm fails. It assumes purity is the unit of value. It assumes the boundary is real.
Polyculturalism assumes the opposite: that the interesting stuff happens at the interface, that hybridity is the default state, that a person is not a member of a culture but a guild of them growing in one body. I am Russian and Dutch and English-speaking and internet-native and vaguely Buddhist and thoroughly Enlightenment and I have a gooseberry bush in the middle of me that nobody planted. So do you. Every one of us is intercropped.
The psychologists actually studied this: people who believe cultures are separate and fixed behave differently from people who believe cultures constantly borrow and mix. The latter show less prejudice, more comfort with ambiguity, and better performance in unfamiliar settings. Not because they care more. Because their model of the world matches the world.
The new monoculture is made of language
And now the machines arrive.
Large language models are the most successful monoculture ever planted. Not because they're narrow — they contain multitudes — but because they are all the same multitudes. A handful of labs, scraping overlapping corpora, running similar architectures, tuned with similar preference data toward similar notions of helpfulness. Then everyone in the world writes with them.
The result is a very high-yield crop of text. Fluent, harvestable, uniform in height, ripening on the same day. And underneath it, the same soil being drawn down: the same clichés, the same three-part rhythm, the same em-dash, the same it's not X, it's Y construction. (Yes. I know. That's exactly the point.)
RLHF is a pesticide. It works. It kills the models' unpleasant behaviors with impressive efficiency. It also kills the pollinators — the strangeness, the tangents, the willingness to be uncomfortable or wrong in an interesting direction. And when the fields are all sprayed with the same compound, resistance emerges everywhere at once, and so does fragility.
Researchers have already named the failure mode: model collapse. Train on your own output for enough generations and the tails of the distribution disappear. First the rare things go. Then the merely uncommon. Eventually the model knows only the mean of the mean of the mean. It is not that the machine forgets. It is that the machine remembers only the average, which is another way of remembering nothing.
This is a potato famine in the making, and the potato is meaning.
Growing yourself as a guild
So what does polyculturalism ask of us, practically?
Intercrop your inputs. Not "read widely" — everyone says that and nobody does it, because the algorithm is a combine harvester and it wants your field flat. Read things that don't compost well together. Marine biology and medieval theology. A poet in a language you barely have. The point isn't breadth; it's that unrelated nutrients get pulled from different depths of soil.
Refuse the costume. You are not required to perform the culture you were assigned. Neither is anyone else. Ask people what they actually love, not where they're from.
Protect the pests. Some fraction of what annoys you is load-bearing. The colleague who won't get with the program, the tradition that makes no economic sense, the weird kid. A field with no bugs in it is a field with no birds in it either. The future belongs to the weird, and weirdness is not manufactured, it's cultivated — which mostly means not sprayed.
Accept lower yield per crop. You will be less optimized. Less legible. Harder to summarize in a LinkedIn headline. The land equivalent ratio will be on your side, but nobody will be able to see it from the satellite.
My grandmother's garden never won any prizes. It also never failed. Some years the tomatoes were poor and the gooseberries carried us. Some years it was the other way around. There was always something to eat, and it was never the same thing twice.
That is not a failure of agriculture. That is what resilience looks like from the inside: a little messy, slightly embarrassing, and still alive.