In Defense of Intuition (with Gerd Gigerenzer)
What this covers
Psychologist Gerd Gigerenzer and host Russ Roberts trace a genealogy of how intuition became devalued in modern thought. Gigerenzer argues that intuition—rapid judgment built on years of domain experience—is not opposed to conscious reasoning but works alongside it, and that the System 1 / System 2 framework has merely repackaged a centuries-old gendered dichotomy in which women's thinking was cast as intuitive and inferior, men's as rational and superior. The larger claim is sharper: much of the behavioral-economics literature on cognitive bias rests on statistical errors by researchers themselves, not genuine errors by ordinary people. Famous findings like the hot hand fallacy dissolve under correct analysis, and phenomena like framing effects reflect intelligent communication rather than irrational bias. Gigerenzer calls this pervasive misdiagnosis the Bias Bias—a tendency to see irrationality everywhere, even where none exists.
The conversation moves through concrete domains where this error matters. In medicine, mammography screening for asymptomatic women shows no net life extension in randomized trials; communicating tiny benefits as large percentage reductions exploits confusion over Relative vs Absolute Risk. In policy, nudging—manipulating defaults to steer people toward "rational" choices—rests on the assumption that people cannot learn; Boosting vs Nudging offers an alternative that builds competence and preserves human agency instead. Organ donation provides a key case: opt-out defaults raise the roster of potential donors but not actual donations, which depend on organizational capacity and family choice. Gigerenzer defends Ecological Rationality, the principle that heuristics and intuitions are well-matched to their environments—base-rate neglect is optimal when the world changes rapidly, poor when it is stable. He argues Risk vs Uncertainty differs in ways AI cannot bridge: current systems excel at chess and fixed correlations but founder precisely where human judgment excels, in genuine uncertainty and domains that require embodied experience.
Gigerenzer argues that intuition is a form of expertise-based intelligence that complements rather than opposes conscious reasoning, and that much of the 'irrationality' literature reflects researchers' own statistical errors and a paternalistic agenda rather than genuine human bias.
- The System 1/System 2 dichotomy is a flawed reframing of a historically gendered intuition/reason split
- Many famous 'biases' (hot hand, framing) dissolve under correct statistical analysis or ecological context
- Boosting (making people competent) is preferable to nudging (manipulating defaults) for both effectiveness and democracy
The System 1/System 2 dichotomy falsely opposes intuition and deliberation when they actually work together cooperatively.
- There is no opposition between intuition and conscious thinking—the big mistake of the System 1/System 2 framing; for example, a doctor's unexplained feeling that something is wrong (intuition) inspires deliberate diagnostics, so intuition and deliberation work together rather than against each other.
“There is no opposition between intuition and conscious thinking. That's the big mistake of the System 1 or System 2.”
- Into the 20th century, psychology textbooks promoted the idea that men's intellect (rational, slow, reliable) differed from women's (intuitive, fast, unreliable), constructing a dichotomy that did a disservice both to women and to the understanding of rationality and intuition.
“into the 20th century, textbooks in psychology were promoting the idea that men's intellect is very different from female's intellect. And of course: We men are rational, with reason; while women are intuitive.”
- The historical gendered dichotomy (intuitive/female vs rational/male) was reinterpreted in the 21st century as System 1 versus System 2, with Kahneman resolving the Gigerenzer–Kahneman debate by attributing errors to System 1 and correct answers to System 2.
“this dichotomy turned, now in the 21st century, a distinction between a System 1 and a System 2. System 1 was the idea about women, System 2 about men.”
- Intuition is a feeling based on years of experience that arrives quickly in consciousness telling you what to do or not do, which you cannot explain the source of; it is not arbitrary, not a sixth sense, and not exclusive to women—everyone with experience in a domain has it.
“Intuition is a feeling based on years of experience that comes fast into your consciousness--so, you feel what you should do or what you shouldn't do--and where you have no way to explain it, where it's coming from.”
Cognitive biases are overstated; many supposed biases are not universal errors but reflect ecological rationality matching strategy to environment.
- The 'bias bias' is the temptation to see cognitive biases everywhere even when there are none; many famous biases (overconfidence, hot hand fallacy, conjunction error, base rate neglect, framing) are not biases at all but may be biased only in certain situations, and the framework has been misused to blame people and justify paternalism.
“the bias bias is the temptation to see biases everywhere, even if there are none.”
- The supposed 'hot hand fallacy' is not an error by coaches and fans but by the researchers themselves: as Miller and Sanjurjo showed in Econometrica, the original studies failed to understand that the statistical properties of a sample are not the same as the population, so the hot hand actually exists (though small).
“Now we know from a great article in Econometrica by Miller and Sanjurjo that it's not that the coaches have a problem with statistical thinking. It's the researchers, who didn't understand that the properties of a sample are not always the same as the population.”
- Heuristics like overconfidence and base rate neglect are not generally wrong: base rate neglect is the best strategy when the world is changing quickly, though poor in stable conditions—what matters is ecological rationality (matching strategy to environment), not pure logic.
“base rate neglect is the best thing you can do if your world is changing quickly, but not good if it's fractured.”
AI hype and screening overconfidence exploit communication bias and misaligned incentives; prevention and fundamental research matter more than technological cure-alls.
- The belief that AI will solve all problems is a religious faith with no basis; previous technologies like the Human Genome Project were similarly hyped as cure-alls (cancer, Alzheimer's) but reality proved far more complicated due to gene interactions.
“there's no method or no technology that can solve all problems. That's a religious faith. It's like God.”
- Mammography screening for asymptomatic women prolongs life by zero days on average according to randomized trials: of 1,000 screened women aged 50+, four die of breast cancer after ~10 years versus five among the unscreened (0.1% absolute reduction, communicated misleadingly as a 20% relative reduction), and an additional woman in the screened group dies of another cancer, leaving no net benefit.
“the answer is mammography screening prolongs life by zero days. Zero.”
- Overstated AI claims over the last 50 years are driven in part by funding incentives—telling politicians and funders believable stories to get money, then dismissing skeptics as anti-technological—whereas the real breakthroughs (lasers, foundational tech) come from basic fundamental research by people like Planck and Einstein.
“if you look behind why these claims have been made about AI for the last 50 years--overstatements--is in part because of funding.”
- Opt-out organ donation policies increase potential donors but not actual donations; studies of 35 OECD countries and a Max Planck study of countries that switched policies show no rise in actual donations, because real donation depends on a fast, well-funded organizational system and on relatives—who usually make the final decision—not on the default.
“you have only more potential organ donors. Potential. What you want are real organ donors; and there is much more to do than just having an opt-in or opt-out policy.”
- To win the war against cancer the target is prevention, not screening: about half of all cancers are due to behavior (smoking, eating too much and the wrong things, not moving), so prevention should start early—though it is opposed by the tobacco, fast food, and sugar industries.
“if you are going to win the war against cancer, it is not screening. It is prevention. About half of all cancers are due to behavior”
- The framing 'bias'—where '90% survive' versus '10% die' should logically be treated identically—actually reflects intelligent communication: people understand that doctors convey recommendations through framing, so '90% survive' signals 'do it' and '10% die' signals a warning, which research by Craig McKenzie supports.
“if the doctor says you have a 90% chance to survive, that means: Do it. If the doctor says 10% chance you die: That's a warning, not a recommendation, because the doctor knows more about that.”
- AI excels in well-defined, stable worlds (chess, Go, large language models trained on fixed text correlations) but struggles precisely with the highly unconscious, intuitive abilities humans have; predictive AI involving humans or viruses has little success because it must handle genuine uncertainty.
“those abilities that we have, which are highly unconscious and intuitive, are often the hardest for AI to do.”
People understand risk well when presented in natural frequencies; misunderstanding occurs only with unfamiliar forms like conditional probabilities.
- Communicating a 0.1% absolute risk reduction as a '20% reduction' exploits the fact that people don't distinguish relative from absolute risk, functioning as a form of nudging that makes a tiny effect look large.
“you assume that women do not know the difference between a relative and absolute risk. So, you communicate that in relative risk, because it looks different, and, ha, they believe it's a big effect.”
- Probability statements are ambiguous without a reference class: '30% chance of rain' could mean 30% of the time, 30% of the region, 30% of forecasters, or (correctly) rain on 30% of days like tomorrow; the general boosting principle is always to ask what the reference class is, which children by age 10 can grasp.
“the general principle: always ask for the reference class. So, is this 30% about time or region or days or meteorologists?”
- People often misunderstand risk only when it is presented in unfamiliar forms like conditional probabilities; when the same information is presented as natural frequencies, they understand it—and even seven- to eight-year-old children can learn Bayesian thinking this way, as shown with children in Germany and China.
“If you present them in frequencies, they understand... seven- to eight-year-olds can understand Bayesian thinking, if you present the information in an intuitive way.”
Nudging is paternalistic control disguised as behavioral science; boosting respects democracy by making people risk-literate and capable of judgment.
- Nudging (from Thaler and Sunstein, based on Kahneman and Tversky) rests on three claims—people don't understand risk, people cannot learn, and therefore government must steer them toward what they'd choose if rational—making it a 21st-century justification for paternalism.
“the argument is: People do not know how to deal with risks. Second, they cannot learn... third, therefore, because that has immense consequences for our lives, government needs to step in and nudge people”
- Boosting, formalized by Ralph Hertwig, means making people strong and risk-literate rather than nudging them like sheep; unlike nudging it is pro-democratic because democracy works only if people understand and take their part, whereas replacing human judgment with algorithms or authorities is an anti-democratic move toward autocracy.
“boosting means that you make people strong. You don't nudge them like sheep. You make them strong; and they can start already in schools, make them risk literacy.”
Intuition from experience and conscious reasoning serve complementary roles; moral behavior relies primarily on implicit cultural rules defended later by reasoning.
- A key smart rule for a research group is to hire a contrarian—someone who respectfully disagrees with the boss or group spirit on a factual basis without fear—because this is essential for correcting one's own and the group's biases; the same function can be served by a candid spouse or friend.
“one of the most important rules is: Hire a contrarian. That is, a person who respectfully disagrees with you as the boss or with the group spirit”
- A best friend is someone who criticizes your ideas when something is wrong and protects you from running into a knife, unlike the sycophantic behavior of ChatGPT-5.
“the best friend is not someone who claps you on the shoulder and nods and finds everything good, like ChatGPT 5: So sycophantic. I can't stand it.”
- Most moral behavior operates through implicit, intuitive rules absorbed from one's culture and parents, with conscious reasoning typically used afterward to defend what one already feels; per Darwin, morality and religion function evolutionarily to bond a group together and make it stronger for survival, because humans are a social species needing a moral fabric to glue them together.
“moral behavior is mostly true, as you say, through intuitive rules that are not explicit, that people follow, and often then defend after the fact with conscious reasoning, but that has a function.”