Every AI product has an implicit theory of people: what they know, where they fail, and what should be automated. That theory deserves scrutiny.