CONCEPT
Noise in Human Judgment
Kahneman, Sibony, and Sunstein's 2021 framework for the random variability in professional decisions that should be identical — the under-recognized twin of bias, and the specific failure that AI systems most reliably eliminate.
Noise is unwanted random variability in judgments that should be equivalent. Two judges reviewing the same case with the same facts impose sentences differing by years. Two underwriters evaluating identical risks price them fifty percent apart. The same pathologist on different days reaches different diagnoses on identical biopsies. Where bias is systematic error that shifts the mean, noise is scattered error that widens the distribution. Kahneman argued that organizations obsess over bias and ignore noise, despite noise being at least as damaging. AI systems are, by architecture, noiseless: given the same input under the same conditions, they produce the same output. This is a genuine and substantial improvement in domains where consistency matters. But noise elimination has a cost the framework itself illuminates: some variability is not unwanted — it is the raw material of creative insight, and AI compresses human output toward its competent average with consequences for the tails of the distribution where distinctive work lives.
