Gerd Gigerenzer on Gut Feelings
What this covers
Gerd Gigerenzer of the Max Planck Institute discusses his framework for when simple rules of thumb outperform elaborate analytical models. Hosted by Russ Roberts, the conversation centers on the distinction between risk and uncertainty—a divide that determines whether calculation or intuition works better in practice. Gigerenzer argues that many behaviors labeled as cognitive errors in behavioral economics are actually sensible responses to ambiguity and incomplete information. The discussion spans psychology, medicine, finance, and policy, tracing how complex models tend to excel at fitting historical data but fail in prediction when the future diverges from the past.
The conversation moves through several concrete domains. Gigerenzer uses examples like the gaze heuristic—the simple rule outfielders use to catch fly balls without calculating trajectories—and the recognition heuristic, where limited knowledge sometimes beats comprehensive information. He introduces the bias-variance dilemma to explain why overfitting past data sabotages future predictions, and the fast-and-frugal tree as a decision structure superior to models that attempt to balance competing factors. He examines failures of sophisticated systems: Google Flu Trends collapsing despite 160 variables, Value at Risk missing financial crises, and screening programs generating masses of false positives. The conversation also challenges behavioral economics itself, introducing the concept of bias bias—the tendency to find irrationality where none exists—and reinterpreting framing effects and the conjunction fallacy as intelligent adaptations to language and context rather than probability errors. A closing note distinguishes between defensive decision-making in medicine and the question of whether ignorance about uncertain futures can sometimes serve human flourishing better than information.
Gigerenzer argues that under genuine uncertainty (as opposed to calculable risk), simple heuristics and gut feelings built on experience outperform complex models, because complex models overfit the past and fail to predict the future, and that many alleged irrationalities in behavioral economics are actually sensible adaptations to ambiguity.
- The risk/uncertainty distinction determines whether calculation or heuristics work better
- Complex models excel at fitting the past but fail at prediction (bias-variance dilemma)
- Many 'biases' (framing, conjunction fallacy) reflect intelligent reading of language and context, not error
Models that fit data well create false confidence in explanation by conflating correlation with causation and noise with signal.
- An 'as-if' model (Friedman's methodology, Dawkins' ball-catcher solving differential equations) can predict well while being a completely false description of the actual process; the danger in economics is that researchers confuse model confirmation by data with the model being reality, leading to unwarranted inferences like assuming monopsony power when a minimum wage doesn't reduce employment.
“a model can be very effective at predicting even though it's a very poor description... Economists start to confuse the two. All the time.”
- A model that fits historical data well creates an illusion of explanation by implicitly assuming correlation is causation, when in reality much of what it captures is noise rather than signal—which is why such models fail to predict going forward even when they explain only half the variation in past outcomes.
“you're under the illusion you've explained the past because you're really fundamentally in that case, assuming that correlation is causation--when it is not.”
Experimental and lab findings do not reliably transfer to real markets and populations because critical contextual differences change what drives outcomes.
- Decisions about money made in a laboratory setting are misleading because markets provide information not available in the lab, a point made eloquently by Vernon Smith—so context is crucial and experimental findings about financial irrationality may not transfer to real market environments.
“the idea that people make decisions about money in some sort of laboratory setting is, of course, ridiculous. Markets provide information that aren't available in the lab.”
- Randomized controlled trials, though called the gold standard, are oversold because many findings do not hold up with a different or larger population in a different country—because too much that drives outcomes will differ in the future from the past setting where the experiment was run.
“many of the findings which are in RCTs, randomized controlled trials, which is the so called gold standard, don't seem to hold up so well with a different population--with a larger sample, in a different country.”
Dismissing evolved heuristics and traditions as mistakes ignores that they embody information unavailable to individual reasoning.
- Chesterton's Fence is the principle that a rule or tradition you don't understand may exist for a good reason you cannot see, so tearing it down is dangerous; many heuristics and rules of thumb evolved over time consistent with how our brain works, and arrogant experts wrongly dismiss them as mistakes when respecting tradition actually accesses information otherwise unavailable.
“the Chesterton Fence is this idea that you come across a fence in the middle of a field... when you tear it down, you find out it had a purpose.”
- A Harvard MBA CEO attributed his company's bankruptcy to having chosen the wrong case study, illustrating that case studies function as heuristics or taxonomies for decisions and that picking the wrong rule of thumb can be costly—showing heuristics can fail and require care in selection.
“I asked him what went wrong. He said, 'Oh, I chose the wrong case study.'... case studies are heuristics... he'd fallen from grace because he had chosen the wrong rule of thumb.”
Criticizing bad applications of science is not anti-science but necessary to prevent scientific methods from becoming dogma.
- Criticizing regression analysis, cost-benefit analysis, or big data is not anti-science but anti-bad-science; science is valuable but must be understood with its limitations, otherwise it becomes the equivalent of a religion.
“I'm often accused of being anti-science. And I respond by saying, 'No, I'm anti bad science.'... we have to understand it's limitations. Otherwise, it becomes the equivalent of a religion.”