Anthropology is the discipline that refuses to guess what people are like.

Every system built for people encodes a theory of people. Sometimes that theory is written down. More often it is inherited, borrowed from a market segment, a persona deck or an engineer's intuition about a reasonable user.

Behaviour rarely obeys those theories. A patient accepts a diagnosis but not the explanation attached to it. A clerk keeps a paper record alongside the digital one, not out of resistance but out of accountability. These are not deviations from the model. They are the evidence that the model was incomplete.

Understanding people is not a soft prelude to the real work. It is the part of the work that determines whether anything else holds.

Artificial intelligence is now infrastructure. It sits between a patient and a diagnosis, a student and a grade, an applicant and a decision. Because it is infrastructure, its failures are rarely dramatic — they are quiet, distributed and absorbed by the people least able to contest them.

Designing from the human outward changes what counts as a requirement. Explainability stops being a feature and becomes a relationship. Accuracy stops being sufficient and becomes one input into whether a person is willing to act.

Anthropology's method is patient and unglamorous: stay long enough that explanation replaces assumption. It studies what people do, in context, with their own categories intact — including categories that make no sense to a dashboard.

Medical anthropology brings a vocabulary for illness, care and authority. Digital anthropology brings one for platforms, data and mediated relationships. Together they make it possible to describe what an AI system actually does to a life, rather than what it was intended to do.