Why Thinking, Fast and Slow changed how we understand decisions

In September 2012, Daniel Kahneman sent an unusual letter. Addressed to the social priming research community, it began: "I see a train wreck looming." He had recently published a book that prominently featured social priming — the experimental finding that exposure to words related to old age makes people walk slowly, that an image of eyes makes people behave more honestly, that small environmental cues invisibly shape behaviour. His book had helped make these findings famous. And now, as preregistered replication failures accumulated across laboratory after laboratory, Kahneman was telling his colleagues that "your field is now the poster child for doubts about the integrity of psychological research."

That letter, self-inflicted and early, is the most Kahneman-like thing Kahneman ever did. It is also the frame through which to read Thinking, Fast and Slow a decade after publication: a book that contains some findings that did not replicate, a central thesis that is stronger than ever, and a self-defeating irony that the book itself anticipated.

The model, precisely stated

The terms "System 1" and "System 2" were coined not by Kahneman but by Keith Stanovich and Richard West, in a 2000 paper in Behavioral and Brain Sciences — a fact Kahneman acknowledges clearly in the book. He adopted the framework because it was the cleanest available label for a distinction researchers had been making informally for decades.

System 1 is fast, automatic, associative, emotionally laden, runs continuously without effort, and is not under voluntary control. It produces impressions, feelings, and intuitive judgments. It is not a separate brain region or module — it is a label for the aggregate of fast cognitive processes. System 2 is slow, effortful, serial, and rule-governed. It requires working memory. Critically, it is also lazy: it tends to accept S1's outputs without scrutiny and often rationalises what S1 has already decided.

The key heuristics S1 uses are availability (judging probability by ease of retrieval, not by actual frequency), representativeness (judging probability by resemblance to a prototype, producing the conjunction fallacy), and anchoring (estimates clustering around an initial number regardless of its relevance). These are not theoretical constructs. They are some of the best-replicated experimental findings in all of psychology.

WYSIATI — "What You See Is All There Is" — is Kahneman's own formulation. S1 constructs the most coherent story it can from available information and treats that story as complete. The absence of missing information is not noticed. This produces overconfidence, narrative fallacies, and the illusion of understanding. A compelling story feels true even when it is built from insufficient evidence.

What has held up

The empirical core of the book rests on findings that are older and more thoroughly replicated than the priming literature: prospect theory, loss aversion, anchoring, the planning fallacy, the peak-end rule, and regression to the mean.

Prospect theory — developed by Kahneman and Amos Tversky in a 1979 paper in Econometrica — is the most-cited paper in the history of economics. Its central finding is that people evaluate outcomes relative to a reference point, that losses feel roughly twice as painful as equivalent gains feel pleasurable, and that the relationship between probability and decision weight is not linear. A 2024 meta-analysis by Brown, Imai, Vieider, and Camerer, surveying 607 empirical estimates of loss aversion from 150 papers, found a mean loss aversion coefficient of 1.96. The 2x figure Kahneman cited is a slight overstatement. The direction is not in doubt. Losses loom larger than gains across cultures, populations, and experimental designs.

Anchoring is arguably the best-replicated finding in social psychology. The Many Labs replication project — 36 laboratories, 6,344 participants — tested 13 classic psychological effects and found anchoring among the four or five strongest. Anchoring occurs even among experts: judges give systematically higher sentences when an anchor number is mentioned first. Doctors adjust diagnoses based on unrelated numbers they recently heard. Negotiators cluster around initial offers even when those offers are known to be arbitrary.

The planning fallacy was first named by Kahneman and Tversky in 1979, and its mechanism — systematically underestimating time, cost, and risk while overestimating benefits — is by now one of the most documented phenomena in project management. The Sydney Opera House, budgeted at $7 million in 1957, was completed in 1973 for $102 million: a 1,300% cost overrun, with a 250% schedule overrun. The solution Kahneman proposed — reference class forecasting, looking at the statistical distribution of outcomes from comparable past projects rather than estimating from the unique features of this project — has been adopted by the UK Treasury and several major infrastructure planning agencies.

The peak-end rule was demonstrated cleanly in two experiments: the cold water study, in which participants preferred a longer and more painful immersion that ended slightly better to a shorter immersion at maximum discomfort; and the colonoscopy study, in which extending the procedure with three motionless minutes significantly improved patients' remembered experience, despite adding total discomfort. The experiencing self and the remembering self want different things, and we live with our memories. This finding has been applied in the design of medical procedures and customer service interactions and has held up to scrutiny.

Regression to the mean is not an experimental claim — it is a statistical fact. The insight Kahneman applied to it is that we systematically attribute regression to our interventions. Flight instructors who praised good landings saw worse performance next time; instructors who criticised bad landings saw better performance next time. They concluded praise was counterproductive. They were wrong: extreme performances naturally regress toward average, regardless of feedback. We reward and punish in response to statistical noise and then attribute the regression to our actions.

What hasn't held up

Ego depletion does not hold. Baumeister's finding — that willpower draws from a depletable resource, so that resisting cookies impairs subsequent persistence on puzzles — was presented as evidence that S2 is depleted by use. A preregistered multi-lab replication across 23 laboratories with over 2,000 participants, published in Perspectives on Psychological Science in 2016, found effectively zero effect. Kahneman relied on this finding in his discussion of cognitive load and S2 fatigue. Those passages are now scientifically unsupported.

Social priming — the Bargh walking-speed experiment, money primes affecting self-sufficiency, eyes-on-a-wall honesty effects — has been substantially discredited. When Doyen et al. replicated the walking-speed study with automated timing (rather than a human observer who knew the hypothesis), the effect vanished. Kahneman's 2012 letter was the clearest acknowledgement from inside the field. In 2017 he stated publicly that the implicit priming literature "is not trustworthy" and that he had "placed too much faith in underpowered studies." The chapter on priming in Thinking, Fast and Slow is the part he would most significantly rewrite.

The hot hand is a genuinely complicated case. Gilovich, Vallone, and Tversky (1985) concluded that basketball shooting streaks are a cognitive illusion — that sequential hits are no more common than chance would predict. Kahneman accepted this. Miller and Sanjurjo (2018) found a systematic selection bias in Gilovich's original statistical analysis; when corrected, Gilovich's own data shows a positive hot hand. Bocskocsky, Ezekowitz, and Stein (2014), using optical tracking data from over 83,000 NBA shots, found hot-hand effects of 1–2 percentage points masked by the fact that defenders respond to hot shooters. The hot hand fallacy, as a clear and simple cognitive illusion, is not the settled fact Kahneman presented it as.

Why the 5-star rating still holds

None of the failed chapters touch the book's central argument. The hypothesis that humans are systematically non-rational, that our intuitions fail predictably across specific domains, that loss and gain are processed asymmetrically, that confidence systematically exceeds accuracy — these are better supported in 2025 than they were in 2011. The replication crisis has weakened some illustrations and strengthened the meta-thesis.

The book changed how decisions are understood not by cataloguing everything correctly but by installing a framework — the two-system model, the experiencing versus remembering self, the inside versus outside view — that makes the structure of human error legible and therefore addressable. That framework has not been falsified. Reference class forecasting is now used in government infrastructure planning. Pre-mortem analysis — imagining a project has already failed and asking what went wrong — is a standard technique in strategy consulting. The peak-end insight is applied in medical procedure design. These are practical, working applications of the book's core ideas.

The self-defeating irony Kahneman anticipated is also worth naming. A book that warned us we over-trust compelling narratives itself contained some compelling but fragile narratives about priming. Kahneman acknowledged this without defensiveness, which is exactly the behaviour the book asks of us. The error was real. The response was the ideal one. The book demonstrated its own thesis by being susceptible to it — and then by demonstrating what to do when you notice that.

That is not a failure. That is a five-star book.

Sources

• Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
• Tversky, A. & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131.
• Tversky, A. & Kahneman, D. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263–291.
• Brown, A. L., Imai, T., Vieider, F. & Camerer, C. (2024). Meta-analysis of empirical estimates of loss aversion. Journal of Economic Literature, 62(2), 485–516.
• Hagger, M. S. et al. (2016). A multilab preregistered replication of the ego-depletion effect. Perspectives on Psychological Science, 11(4), 546–573.
• Doyen, S., Klein, O., Pichon, C.-L. & Cleeremans, A. (2012). Behavioral priming: It's all in the mind, but whose mind? PLoS ONE, 7(1), e29081.
• Miller, J. B. & Sanjurjo, A. (2018). Surprised by the hot hand fallacy? Econometrica, 86(6), 2019–2047.
• Klein, R. A. et al. (2014). Investigating variation in replicability: A "Many Labs" replication project. Social Psychology, 45(3), 142–152.
• Stanovich, K. E. & West, R. F. (2000). Individual differences in reasoning. Behavioral and Brain Sciences, 23(5), 645–665.