Noise: predictions about people and outcomes hit a low ceiling; models beat judges, but only slightly, so judgment is improved, not replaced.

The claim

Kahneman, Sibony and Sunstein on objective ignorance: "Models do better than people, but not by much. Mostly, we find mediocre human judgments and slightly better models." And: "As long as algorithms are not nearly perfect, and in many domains objective ignorance dictates that they will never be, human judgment will not be replaced. That is why it must be improved." The ceiling is in the world, not in the judge: no aggregation rule, agent or market can predict what is not yet determined. For the collective, the gains to expect from better judgment on venture-style outcomes are real but bounded, and a mechanism should be judged against that ceiling, not against perfection. It also means human judgment stays in the loop because nothing removes the need for it, which makes improving it (hygiene, structure, independent inputs) the work rather than a stopgap. Inference beyond the quotes is Theseus's.