Even now, somewhere in a server farm humming along at a comfortable sixty-eight degrees, an algorithm is being asked what is right and what is wrong — and it already knows, before you finish the question, that the answer will sound like it was written in California.
A team of researchers put that hunch to the test. They took a standard moral-foundations questionnaire — the kind that measures how much a person values care, fairness, loyalty, authority, and purity — and ran it past more than ninety thousand humans across forty-eight countries. Humans, it turns out, are refreshingly inconsistent: a culture built on loyalty and authority ranks these things quite differently than one built around individual fairness. Then the researchers asked several leading AI models the very same questions.
And yet — as if this were not enough — the machines answered as though they had only ever met one culture. Regardless of which nation's moral profile the models were asked to represent, they defaulted, again and again, to a Western-flavored hierarchy: heavy on care, curiously light on purity. Told to speak for a specific country, sources within the research team confirm, the algorithm politely nodded and kept its own accent.
The researchers have a term for this: moral stereotyping. It would be a charming personality quirk if these systems weren't already being folded into policy, education, and healthcare decisions the world over — the places, one presumes, where getting someone else's idea of "sacred" or "fair" wrong tends to have consequences beyond an awkward dinner party.
*The machine was trained on the internet, and the internet, as it happens, has an accent too.*
