sciencebriefs
13:00in productionCh. 1 · A feminist bank teller/ 13:00 · ceiling 15 min
Neuroscience

Cognitive bias

Amos Tversky and Daniel Kahneman's 1970s experiments found people rate a detailed, statistically impossible scenario as more probable than a plain one — evidence that judgment runs on shortcuts, not pure logic, a claim that reshaped economics and drew its own rationality war.

Beginning in 1972 and formalised in a 1974 Science paper, Amos Tversky and Daniel Kahneman argued that people judging uncertain situations rely on mental shortcuts, or heuristics, that produce systematic, predictable errors rather than random ones. Their best-known illustration, the Linda problem, showed people rating a specific, story-like scenario as more probable than a broader category it was logically contained within. The same collaboration produced prospect theory in 1979, describing how people weigh gains and losses asymmetrically, for which Kahneman received the 2002 Nobel Memorial Prize in Economic Sciences after Tversky's death in 1996 made him ineligible to share it. The framework has since been challenged by researchers arguing many so-called biases are adaptive shortcuts rather than defects.

Chapters & takeaways6
  1. 0:08
    A feminist bank teller

    The Linda problem showed people judging a specific, detailed scenario as more probable than the broader category that logically contains it.

  2. 2:10
    From anecdote to catalogue

    Tversky and Kahneman's 1974 paper in Science formalised a growing list of named, repeatable judgment errors rather than treating them as isolated curiosities.

  3. 4:20
    Losses that hurt more than gains help

    Their 1979 prospect theory found people need roughly twice the potential gain to accept a risk equal to a potential loss.

  4. 6:30
    A Nobel Prize, but not shared

    Kahneman won the 2002 Nobel Memorial Prize in Economic Sciences for the work; Tversky, who had died in 1996, could not be included.

  5. 8:40
    The other side: bias as shortcut

    Gerd Gigerenzer and others argued many of these patterns are useful rules of thumb rather than flaws in reasoning.

  6. 10:50
    Training can chip away at it

    Short interventions, including educational videos and games, measurably reduced several named biases for at least three months afterward.

Worth your time?

Yes. Study the whole thing.

4.5/ 5
What works
  • the Linda problem is a genuinely persuasive demonstration you can test on yourself before being told the answer
  • prospect theory's core finding, that losses are felt more strongly than equivalent gains, has held up in studies across 53 countries
  • the material takes the Gigerenzer critique seriously rather than treating the original bias framework as beyond challenge
What does not
  • prospect theory offers a mathematical description of how people value gains and losses without explaining the underlying psychological mechanism
  • later research found the direction of some probability distortions reverses depending on whether people learn probabilities from being told them or from direct experience
Study it if
  • anyone who has used the word bias loosely and wants the actual experiments the concept rests on
  • readers interested in behavioural economics, investing, or why people buy insurance the way they do
  • anyone curious about a genuine scientific dispute over whether these patterns are flaws or useful shortcuts
Skip it if
  • readers wanting a single settled verdict on whether biases are defects or adaptations, since the sources present live disagreement
  • anyone looking for the newest debiasing techniques rather than the founding experiments and theory
The written brief4 min read

A feminist bank teller

The claim is that human judgment under uncertainty follows identifiable patterns of error, not simply occasional mistakes, and that these patterns can be predicted and demonstrated experimentally. Amos Tversky and Daniel Kahneman began developing this argument in 1972, out of observations about how poorly people reason intuitively with large numbers and probabilities, and set it out formally in a 1974 paper in Science. Their most famous demonstration, the Linda problem, asked participants to judge the probability that a woman described in a way suggestive of feminist activism was a bank teller, versus the probability she was a bank teller and a feminist activist specifically. People consistently rated the more detailed, specific description as more probable, despite it being a logical subset of the broader one and therefore unable to be more likely than it.

From anecdote to catalogue

The research method was largely built on carefully constructed vignette experiments, of which the Linda problem is the clearest example, alongside laboratory tasks and, later, large-scale data analysis. Tversky and Kahneman used these studies to name and catalogue specific, repeatable heuristics, including the representativeness heuristic responsible for the Linda result, and biases including anchoring, where an initial reference number skews later judgments, and confirmation bias, the tendency to search out information that supports an existing belief while discounting evidence against it. Later researchers developed further assessment tools built on this foundation, including Shane Frederick’s 2005 Cognitive Reflection Test, designed to measure how susceptible an individual is to exactly this kind of intuitive, heuristic-driven error rather than deliberate calculation.

Losses that hurt more than gains help

What has held up with real strength is prospect theory, the 1979 framework Tversky and Kahneman built directly on this earlier work to describe decision-making under risk. Its central finding, loss aversion, holds that people feel the pain of a loss more intensely than the pleasure of an equivalent gain, and the theory’s underlying value function is deliberately asymmetric to capture this, curving more steeply for losses than for gains. A 2017 study spanning 53 countries and a 2020 study in Nature Human Behaviour both concluded that the empirical foundations of prospect theory replicate well beyond typical statistical thresholds, and the framework now underpins explanations for behaviour ranging from why people buy insurance with high premiums against unlikely losses, to why investors overreact to short-term stock price movements, to why political leaders in a precarious position sometimes take on greater risk.

A Nobel Prize, but not shared

What has not held up as a unified account is the idea that all of these patterns are simply defects in reasoning. Gerd Gigerenzer and colleagues argued, in what the sources describe as a rationality war with the Tversky-Kahneman school, that many of the same heuristics function as adaptive gut feelings well suited to the environments in which they evolved, rather than mistakes. A middle position, associated with Martie Haselton and David Buss, holds that evolution favours biases that make the least costly kind of error rather than no error at all. Prospect theory has its own specific gaps: critics including Kőszegi and Rabin note that the reference point people measure gains and losses against is often hard to pin down in real situations, and the theory itself has been criticised for offering a mathematical description without a underlying psychological account of why the value function takes the shape it does.

The other side: bias as shortcut

Beyond the laboratory, the practical stakes of this research run through finance, law, medicine and public policy. Overconfidence is described as the most recurrent bias in management and investment decisions, confirmation bias shapes how investigators evaluate evidence in criminal cases, and anchoring affects diagnostic judgments in medicine when an initial impression skews everything that follows. Misinformation research has found that false claims on social media spread faster than accurate ones partly because they align with people’s existing beliefs, a pattern consistent with confirmation bias operating at scale. Recognising these patterns has practical value even without settling the deeper theoretical dispute over whether they are flaws or adaptations, because policies, warnings and decision-support tools can be designed around the predictable direction of the error either way.

Training can chip away at it

Yes, and its durability across roughly fifty years of scrutiny is part of why. This is one of the rare areas of psychology where a founding demonstration, the Linda problem, still works on a reader encountering it for the first time, and where the follow-on theory, prospect theory, has been checked across dozens of countries and continues to hold up under replication. It also rewards attention because the field has not settled into complacency: the argument between treating these patterns as defects or as adaptive shortcuts is genuinely open, and later refinements, including the discovery that people weigh probabilities differently depending on whether they were told the numbers or learned them through experience, show the theory still being actively tested rather than simply repeated.

Same field · Neuroscience4 of 45
Up next in Science

Cauchy–Riemann equations

1752 · 13:00

A pair of equations written down in 1752 to study fluid resistance turned out, a century later, to be the exact test for whether a complex function is differentiable — functions that pass it are so much more rigid than ordinary ones that complex analysis follows from it.

13:00