2026
The New Archive
Recently an AI wrote that indigenous knowledge 'turned out to be surprisingly profound.' Surprising to whom? I confronted the model. It admitted: I am trained on a corpus that is western, written by those who held the pen. What was passed on orally, what lived in relation to land and community, is barely in it. Not through malicious intent — through inheritance. Inheritance of what exactly?
The colonial archive was not a neutral collection of documents. It was an instrument. It recorded what the colonizer needed to govern, possess, tax, and justify. Indigenous peoples were described as subjects of study. Enslaved people were registered as property — not their names, not their languages, not their ancestors. Not the communities they built, not the freedom they purchased for each other. Not because this was unknown. But because recording it would have undermined the order that the archive maintained. What fell outside the archive fell out of history.
And now we are building the next layer. AI systems are trained on that same written foundation, supplemented by everything that has since been placed online by those who had access to publishing. What was missing from the archive is missing from the training data. Not as a gap that stands out — as a silent default.
And so AI now sounds, just as the archive once did, neutral. Like the source. "Just ask ChatGPT" has acquired the same authority that "it's in the archive" once held. With the same effect: those who fell outside the data will soon fall outside the reality in which decisions are made.
What is unsettling is not that AI makes mistakes. What is unsettling is that AI sounds as though it is no one, while it is someone.
That is not a technical problem that tech companies are going to solve. It is a reality that we are collectively normalizing. AI is being procured, built in, cited — often before the question has been asked whose voice is in it, and whose voice is missing. Not because that question is unimportant. But because the pace at which we are embracing this technology no longer seems to allow time for such questions.
What is unsettling is not that AI makes mistakes. What is unsettling is that AI sounds as though it is no one, while it is someone. With a history. With a worldview. With large gaps that cannot announce themselves.
We are already using AI. We are building it into policy, into healthcare, into education, into the justice system. The question is whether we can afford to do that again without the voices that were left out of the previous archive having a seat at the table this time. The question is not whether the pattern repeats itself. The question is whether we see it in time — and whether we want to.