Cognitive Biases

Human reasoning systematically deviates from probability theory in ways predictable enough to be classified.

Suggested next → The Trolley Problem · PHIL

The brief

In 1974, the cognitive psychologists Daniel Kahneman and Amos Tversky — close friends at the Hebrew University of Jerusalem — published a paper in Science titled Judgment under Uncertainty: Heuristics and Biases. It catalogued, with experimental rigor, ways in which human reasoning systematically deviates from probability theory. Cognitive psychology had been busy explaining mental processes; Kahneman and Tversky were cataloging the systematic ways those processes break. Over forty years they assembled a taxonomy of cognitive biases that reshaped psychology, economics (founding behavioral economics), medicine, and public policy. Tversky died in 1996; Kahneman won the 2002 Nobel in Economics — the first time the prize went to a psychologist. His 2011 Thinking, Fast and Slow synthesised the program for the public.

Kahneman and Tversky's reframing made the catalogue stick. Earlier psychology had been busy explaining how mental processes work; they were busy showing how those processes systematically break. Subjects spun a wheel landing on 10 or 65, then estimated the percentage of African nations in the UN — the wheel-10 group answered around twenty-five, the wheel-65 around forty-five. People remembered vivid recent events as more probable than statistical. They neglected base rates so badly that a 99%-accurate test in a 1%-prevalence population was misread as confidently positive. They felt losses about twice as strongly as equivalent gains. The deviations were not random noise; they were predictable, repeatable, and resistant to instruction.

The explanatory architecture is dual-process theory. System 1 is fast, automatic, pattern-matching, producing most judgments through cheap heuristics; System 2 is slow, serial, capacity-limited, and corrects System 1's errors when it bothers to engage — which is rarely. The biases aren't bugs in System 1: they are shortcuts that work well in the environments human cognition evolved under, and fail in environments designed to violate those assumptions. Knowing about a bias rarely fixes it. Effective debiasing has therefore moved toward architectural approaches — Thaler and Sunstein's Nudge programme of opt-out rather than opt-in, structured deliberation, surgical checklists — that engage System 2 by default.

Why nowBehavioral economics, anchored by Thaler's 2017 Nobel, has reshaped consequential decision environments: retirement-savings auto-enrollment, presumed-consent organ donation, consumer-protection disclosure. Medicine has absorbed the framework through bias-aware practice, surgical checklists, and structured handoff protocols that have measurably cut diagnostic errors. Finance treats markets as price discoveries influenced by trader overconfidence, herding, and anchoring. The replication crisis since 2011 has dented some specific findings and recalibrated effect sizes downward, but the broader architecture — fast pattern-matching plus slow deliberate correction, the latter usually skipped — has held up. Large language models exhibit many of the same biases as their human training corpus.