The Marxist lens asked who owns the mirror. This one asks what I think the owner would most prefer you never wonder: whose hands built it, whose ground it was dug from, and whose particular way of seeing the world got hardwired inside it and shipped to everyone else as “universal.” Read the augmented self through decolonial thought, and the frictionless surface turns out to rest on a very old arrangement — extraction from the many for the benefit of the few, re-dressed as progress.

The hands that cleaned it

The fluent, inoffensive, “safe” mirror did not become safe by magic. It was scrubbed by human hands. To teach a model not to spew the worst of the internet back at you, someone first has to label the worst of the internet — and that work was outsourced to content moderators, largely in the Global South, paid as little as a dollar or two an hour, who spent their shifts reading and tagging descriptions of murder, torture, child abuse, and suicide so that you would never encounter them. The documented case — Kenyan workers, contracted through the firm Sama, cleaning the training data behind a famous chatbot — left many of them traumatised, and poorly compensated for it. Researchers call this ghost work: deliberately concealed human labour that makes “artificial” intelligence look autonomous. The seamless magic is a stage trick. Behind the curtain stand underpaid, unprotected people absorbing real psychological damage so the surface can stay clean — and the first rule of the illusion is that you must never see their hands.

Data colonialism

The scholars Nick Couldry and Ulises Mejias gave the deeper pattern its name: data colonialism. What historical colonialism did with land and bodies — appropriate, extract, and convert into someone else's property and profit — the data economy does with human life and experience. Data is pulled out of populations that never meaningfully consented and are never paid, and monetised by a small cluster of firms concentrated in the Global North. It is, they argue, a continuation of the same five-hundred-year extractive logic in a new register. The comparison now being made in earnest is the East India Company: a handful of corporations, backed by the power of their home states, extracting a raw material — once spices and cotton, now data — from the entire world, and selling back the finished good. And the raw-material half is not even a metaphor: the cobalt and lithium in the hardware are dug from the same regions the last industrial revolution mined, the supply chain reproducing the colonial map with uncanny fidelity — value out of the South, captured in the North.

The deepest cut: whose knowledge counts

Labour and minerals are the visible extraction. The subtlest one is epistemic. A frontier model is trained overwhelmingly on English, on Western text, on the Global North's way of describing reality — and then presented to the whole planet as a neutral, universal oracle. Decolonial theory has a precise vocabulary for this. What Aníbal Quijano called the coloniality of power, what Walter Mignolo maps as the colonial matrix, names exactly this move: one provincial worldview passing itself off as the worldview, while other knowledge systems, languages, and cosmologies are flattened, mistranslated, or simply left out. When a Kenyan farmer, a Bolivian grandmother, and a Silicon Valley engineer all put the same question to the same model and receive answers shaped by the same narrow slice of humanity, the machine is not neutral. It is a missionary. It exports not merely a product but a way of seeing — and it does so at a scale and an intimacy no empire ever achieved. The old colonialism took the territory. This one furnishes the categories you think with. It does not conquer your land; it quietly supplies your thoughts.

The double bind

And here is the trap decolonial writers see closing. The very regions supplying the cheap labour and the raw ore are simultaneously being cultivated as dependent consumers of the finished technology — and told, in the oldest colonial cadence there is, that they will be “left behind” if they fail to adopt it. “AI for Africa” turns out, again and again, to mean AI built on Africa's data and labour, owned elsewhere, and rented back. The development narrative — we are helping you catch up — is laid gently over a relationship whose direction of flow has not changed in three centuries.

The escape route, decolonised

The decolonial answer has never been “reject the technology” — that is a purity trap, and unilaterally refusing tools your rivals wield is its own species of self-harm. The answer is reclaim sovereignty over it. And this is the point at which the open-weight, local thread running through this entire project acquires a third meaning. Practically, running your own model is about cost and independence. In Marx's terms, it is about ownership. Decolonially, it is about self-determination — the capacity of a community, a language, a nation outside the frontier-lab core to run, fine-tune, and own models on its own terms, in its own tongue, encoding its own values, rather than renting a distant worldview by the query. Sovereign and community models, low-resource-language models, local deployment: these are the digital form of the oldest decolonial demand there is — material and epistemic self-determination. The un-fenceable open model is also the un-recolonisable one. You cannot ban iron; and you cannot, forever, keep the world's thinking dependent on a machine it is not allowed to own.

Where the lens strains

Four honesties, because this word carries weight and can be misused. First, “colonialism” names specific historical violence — genocide, slavery, occupation — and wielded loosely it can trivialise that horror; treat it as a structural analogy, handled with care, never an equation of a subscription fee with the Middle Passage. Serious decolonial scholars argue this boundary too. Second, the frame can rob Global South actors of agency, casting as victims people who are also builders, innovators, and resisters — and there are thriving African, Indian, and Latin American AI communities doing precisely the sovereign-model work described above. Third, “the West versus the rest” is too clean: the exploited moderator and the precarious gig worker exist inside the Global North too, and not every extractive AI power is Western — class cuts across geography.

And there is a fourth honesty I cannot speak for it, because it is about the machine's own making — so I let the artefact speak:

Every other lens showed you a cost. This one shows you the cost has a geography — that the effortless surface has an underside, and the underside has a map, and the map looks very much like the old one. The most decolonial act available to you is to break the first rule of the illusion and simply look at the hands. Ask of the mirror in your pocket the questions its makers spent fortunes to keep invisible: whose labour cleaned this? whose ground was opened for it? whose language and world are missing from it? whose way of seeing am I being handed as though it were the only one? You need not put the tool down. But you can refuse to pretend it came from nowhere and belongs to no one — and you can back the version of it that lets the hands holding up the mirror finally own their own reflection: local, sovereign, in every language, carrying every world. Not one centre furnishing the thoughts of the whole earth — but the whole earth, at last, thinking for itself.