Amos Tversky

Introduction

Amos Tversky (March 16, 1937 – June 2, 1996) was an Israeli cognitive and mathematical psychologist who changed how the social sciences describe human judgment. Working for more than a decade in near-total intellectual partnership with Daniel Kahneman, he showed that departures from rational choice are not random noise but patterned, predictable, and traceable to a small number of mental shortcuts. Their 1979 paper on prospect theory is the founding document of behavioral economics, and their 1974 Science article on heuristics and biases is one of the most cited papers in the history of psychology.

Tversky died of metastatic melanoma at 59, six years before Kahneman received the 2002 Nobel Memorial Prize in Economic Sciences for their joint work. The prize is not awarded posthumously. "I feel it is a joint prize," Kahneman told The New York Times. "We were twinned for more than a decade."

Kahneman died on 27 March 2024, aged ninety. The death was confirmed to The New York Times by his partner Barbara Tversky — Amos Tversky's widow, at the time a professor of psychology at Stanford and later at Teachers College, Columbia — and the obituaries were obliged to explain, once again, that the prize cannot be shared with a dead man.

Early Life and Education

Tversky was born in Haifa, in what was then British Mandatory Palestine. His father, Yosef Tversky, was a veterinarian born in Poland; his mother, Jenia (née Ginzburg), was a Lithuanian-born social worker who served about fifteen years in the Knesset for the Mapai labor party. A sister, Ruth, was thirteen years his senior.

At high school he was taught by the literary critic Baruch Kurzweil and counted the future poet Dahlia Ravikovich among his classmates. He was also a member and leader of Nahal, the youth movement that combined farming with military service. The combination of literary training and paramilitary discipline — close reading of texts, and a tolerance for physical risk — marked him for life.

He served in the Israel Defense Forces as a paratrooper, rose to the rank of captain, and was decorated for bravery. He parachuted into combat during the Suez Crisis in 1956, commanded an infantry unit during the Six-Day War in 1967, and served in a psychology field unit during the Yom Kippur War in 1973. Colleagues later noted that a man who had commanded men under fire was not easily rattled by academic argument.

Tversky took his bachelor's degree at the Hebrew University of Jerusalem in 1961, studying philosophy and psychology, and completed his doctorate at the University of Michigan in 1965. Michigan in that period was the centre of mathematical psychology, and Tversky's training was in the theory of measurement — the question of what has to be true of the world before numbers can legitimately be assigned to it. In 1963 he married the American psychologist Barbara Gans, later a professor of human development at Teachers College, Columbia University; they had three children.

Academic and Professional Career

Tversky returned to the Hebrew University to teach after Michigan, and it was there, in 1969, that the collaboration began. Kahneman invited him to speak to a seminar; Tversky presented experiments from the laboratory of Ward Edwards, Kahneman objected from the floor, the two argued, and the argument became a friendship. They wrote together by sitting in a small room with a notepad, negotiating every sentence, and by one account deciding first authorship with a coin toss. Tversky won the toss.

He joined the faculty at Stanford University in 1978 and stayed for the rest of his life. The collaboration continued across an ocean and then across a continent, by long telephone calls and by taking turns at the burden of drafting. Their peak decade was roughly 1971 to 1981. In the mid-1980s the partnership cooled — Tversky's MacArthur Fellowship and election to the National Academy of Sciences, honors Kahneman did not share, were part of the strain — and they reconciled in the months before Tversky's death.

His honors included election to the American Academy of Arts and Sciences in 1980, the American Psychological Association's Distinguished Scientific Contribution Award in 1982 (shared with Kahneman), a Guggenheim Fellowship and a MacArthur Fellowship in 1984, election to the National Academy of Sciences in 1985, and the University of Louisville Grawemeyer Award for Psychology in 2003, again shared with Kahneman. He worked at home until the last days of his life.

Major Contributions to Psychology

Foundations of measurement

Before the biases, there was measurement. Tversky was a co-author, with David Krantz, Duncan Luce and Patrick Suppes, of the multi-volume Foundations of Measurement (1971), the definitive treatment of the conditions under which psychological attributes can be represented numerically. It remains the reason he is called a mathematical psychologist, and it explains the severity of his later work: he knew precisely what a rational model required before he set out to show where people departed from it.

Similarity and the contrast model

Tversky's 1977 paper "Features of similarity" dismantled the idea that similarity is a geometric distance. Similarity judgments, he showed, are asymmetric — people say North Korea is more like China than China is like North Korea — and can violate transitivity. Such patterns cannot arise in any spatial model. He replaced distance with a feature-matching contrast model in which common features increase similarity and distinctive features decrease it, weighted by salience. The paper is a standard demonstration that a concept everyone uses confidently behaves in no way that theory requires.

Heuristics and biases

The 1974 Science article, "Judgment under Uncertainty: Heuristics and Biases," argued that people estimate unknown quantities by substituting an easier question: how representative is this case of a category, how easily do examples come to mind, what number did I just hear. Representativeness produces neglect of base rates and of sample size — the "belief in the law of small numbers" documented with Kahneman in 1971. Availability makes rare but vivid events, such as plane crashes, seem common. Anchoring leaves estimates contaminated by whatever number was on the table first.

The most famous single demonstration is the conjunction fallacy. Given a description of Linda — single, outspoken, philosophy major, concerned with discrimination, active in anti-nuclear protests — most people judge it more probable that she is a bank teller and a feminist than that she is a bank teller, which is logically impossible. Tversky and Kahneman's reading was that representativeness had replaced extensional reasoning.

Prospect theory

Prospect theory, published in Econometrica in 1979, replaced expected utility with a model in which outcomes are evaluated as changes from a reference point, losses loom larger than equivalent gains, and decision weights are not probabilities. The value function is concave for gains and convex for losses, steepest near the reference point, and steeper on the loss side — loss aversion. The theory accounted in one structure for risk aversion in gains, risk seeking in losses, the certainty effect, and the isolation effect, and it explained why framing the same problem as lives saved or lives lost reverses preferences. In 1992 Tversky and Kahneman extended it to cumulative prospect theory, which handles any number of outcomes.

Endowment effect, preference reversal, and support theory

Tversky, Kahneman and Paul Slovic documented preference reversal, in which people choose one gamble and price another higher, violating the assumption that preference is a stable ordering. The endowment effect — that ownership alone raises what people demand to give something up — was demonstrated in experiments with Thaler and interpreted through loss aversion. With Craig Fox, Tversky proposed comparative ignorance: people dislike ambiguity mainly when an unambiguous comparison makes the ambiguity noticeable. With Derek Koehler he developed support theory, holding that judged probability depends on how an event is described, so that unpacking a hypothesis into its parts raises its apparent probability — a direct violation of extensionality.

Major Publications

  • Krantz, D. H., Luce, R. D., Suppes, P., & Tversky, A. (1971). Foundations of Measurement (Vol. 1). New York: Academic Press.
  • Tversky, A., & Kahneman, D. (1971). Belief in the law of small numbers. Psychological Bulletin, 76(2), 105–110.
  • Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5(2), 207–232.
  • Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131.
  • Tversky, A. (1977). Features of similarity. Psychological Review, 84(4), 327–352.
  • Tversky, A., & Kahneman, D. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263–291.
  • Kahneman, D., Slovic, P., & Tversky, A. (Eds.) (1982). Judgment Under Uncertainty: Heuristics and Biases. Cambridge University Press.
  • Tversky, A., & Kahneman, D. (1983). Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment. Psychological Review, 90(4), 293–315.
  • Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5(4), 297–323.
  • Tversky, A., & Fox, C. R. (1995). Weighing risk and uncertainty. Psychological Review, 102(2), 269–283.

Impact and Legacy

Tversky's influence runs outward from psychology into economics, law, medicine, and public policy. The nudge movement, disclosure regulation, default options in retirement saving, and the design of medical consent forms all rest on findings he produced. A 2002 survey in Review of General Psychology ranked him the 93rd most cited psychologist of the twentieth century, tied with Edwin Boring, John Dewey and Wilhelm Wundt — a placement that understates his effect outside psychology, since his largest citations are in economics journals. Michael Lewis's The Undoing Project made the partnership itself a popular subject.

The most serious challenge came from Gerd Gigerenzer and the ecological rationality program, and it struck at the interpretation rather than the data. Gigerenzer argued that the standards of correctness were wrong: judgments that look erroneous under probability format largely recover when the same quantities are stated as natural frequencies, counts out of a concrete population. Gigerenzer and Ulrich Hoffrage showed this for Bayesian updating in 1995. Ralph Hertwig and Gigerenzer extended it to the Linda problem in 1999, contending that respondents were not violating probability theory but reading "probable" in its ordinary sense of plausible, and that the error rate falls sharply when the question is posed as "of 100 people like Linda, how many are bank tellers." Their broader charge is that heuristics-and-biases researchers display a "bias bias," classifying reasonable behavior as error by comparing it to an inappropriate standard, and that the heuristics themselves are specified too vaguely to be tested.

The exchange reached its sharpest form in a single volume of Psychological Review in 1996, where Kahneman and Tversky defended the reality of cognitive illusions and Gigerenzer replied that the norms were narrow. Tversky and Kahneman had answers that still carry weight. Making an illusion disappear by restating the problem does not show it was never there, any more than restating a visual illusion removes it. Some documented errors involve frequency data and cannot be repaired by translation into frequencies — the hot hand in basketball is an illusory positive autocorrelation with no frequency-format fix. And base-rate neglect persisted in an experiment in which participants received 250 trials with feedback, a design that gives frequency information everything it could want. Attempts to control the conversational reading of the Linda problem reduce the effect without removing it: when the first option is rewritten as "a bank teller whether or not she is active in the feminist movement," a majority still commit the conjunction error, and several later analyses concluded that the effect is a genuine error rather than a misunderstanding.

A second line of criticism concerns loss aversion itself. David Gal argued in 2006 that many classic loss-aversion results are better explained by inertia, a psychological law of status quo preference, and Gal and Derek Rucker argued in 2018 that loss aversion has been oversold as a general principle. The asymmetry between gains and losses is real in the laboratory, but how much of it survives outside the narrow choice problems that generate it remains contested.

What is not contested is the methodological legacy. Tversky insisted that departures from rationality be demonstrated in experiments simple enough to be run on a seminar audience and robust enough to survive repetition, and that the normative model be stated precisely enough to be violated. Most of the current argument about whether people are rational is conducted on ground he surveyed.

The same tests are now run on machines

The most unexpected afterlife of the 1974 Science paper is that Linda, the cab problem and the Asian-disease framing have become benchmarks for artificial intelligence. A 2025 review in Nature Reviews Psychology 4:777–792 reads large language models through dual-process theory and finds that they alternate between System-1-like answers carrying the classic biases and System-2-like answers under particular prompting regimes, while warning that the resemblance is not a mechanism: in a model the "bias" reflects patterns in the training corpus rather than a cognitive process, and models exhibit distortions humans do not, hallucination among them. A large-scale evaluation of 45 models and about 2.8 million responses across eight bias families found bias-consistent answers in 17.8% to 57.3% of instances, with models above 32 billion parameters less susceptible in roughly 39.5% of cases, and a causal study using cross-tuning traced the biases mainly to pretraining rather than to instruction tuning.

The sharpest result comes from the other side of the old quarrel. Arend Hintze and colleagues, with Ralph Hertwig — a long-time collaborator of Gerd Gigerenzer — asked how models weight probability in simple economic gambles, and most did not reproduce the human inverse-S-shaped pattern at all: some were risk-seeking, some showed strong recency effects, and the most accurate behaved like decision-from-experience rather than decision-from-description (Hintze, Bisht, Schossau & Hertwig, 2025, Computers in Human Behavior: Artificial Humans 6:100227). The conclusion is the one Tversky would have demanded. A system can depart from the normative answer without departing in the human way, so the existence of an error says nothing about what caused it. Diagnosing machines with the 1974 instruments has thus reinstated Tversky's own rule, that a departure from rationality has to be shown against a stated standard and traced to a mechanism before it means anything.

Learn More

References

  • Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124-1131.

Last updated: 2026-10-20

Menu
Loading