VI
Mark học AIDaniel Kahneman
Our intuition is usually right, but when it is wrong it is wrong in predictable ways. Knowing the names of those errors helps us notice when to slow down.
Chapters per part. Parts II and IV are the work behind the 2002 Nobel.
| The real (hard) question | What System 1 quietly answers (easy) |
|---|---|
| Should I invest in Ford stock? | Do I like Ford cars? |
| Will this candidate be a good leader? | Does this person look the part? |
| How much would I pay to save oil-soaked birds? | How sad do I feel about them? |
| How happy am I these days? | What is my mood right now? |
WYSIATI: "what you see is all there is". System 1 builds a story from whatever little it has and never asks what is missing.
| When we might GAIN | When we might LOSE | |
|---|---|---|
| High probability | Risk averseFear losing a near-sure gain · accept a low settlement | Risk seekingHope to escape · keep betting to win it back |
| Low probability | Risk seekingHope for a big win · buy lottery tickets | Risk averseFear of disaster · buy insurance |
Red edges mark where we gamble; blue edges mark where we pay for peace of mind.
The 13 most-quoted claims, rechecked after psychology's replication crisis. Tap a row to see why.
Replicated in the Many Labs project (2014) with thousands of people across many countries.
The "Asian disease" problem replicates reliably, including in Many Labs. The effect shrinks when people deliberate.
Replicated many times. Gigerenzer counters that asking in frequencies ("how many out of 100?") sharply reduces the error.
Infrastructure data (Flyvbjerg) show cost overruns and delays are the rule, not the exception.
From Paul Meehl (1954) to later meta-analyses, formulas win or tie in most comparisons.
Kahneman himself calls them "useful fictions". Many researchers see thinking as a continuum rather than a clean split.
Loss aversion is real and meta-analyses support it. Its size varies with context; for small stakes it can be very weak (Gal and Rucker, 2018).
In 2023 Kahneman and Killingsworth reanalyzed together: the plateau holds only for the least happy group; most people keep feeling better as income rises.
Miller and Sanjurjo (2018) found a statistical flaw in the original study; corrected, a small effect appears.
Failed to replicate (Doyen et al., 2012). In 2017 Kahneman wrote that he had "placed too much faith in underpowered studies" in chapter 4.
A preregistered study across 23 labs with over 2,000 people (Hagger et al., 2016) found an effect near zero.
A meta-analysis of 17 experiments (Meyer et al., 2015) found no effect.
The drop is too large to be hunger. Case order was not random: cases without a lawyer tended to come last in a session.
Keep Parts II to V. Read chapters 3–4 as history of science, not as conclusions.
The book came out in 2011, just before social psychology found that many famous results did not replicate. In 2012 Kahneman himself wrote an open letter warning of a "train wreck looming" in priming research.
In 2017 Schimmack, Heene and Kesavan rated the replicability of the studies cited in chapter 4 and found it very low. Kahneman replied publicly that he had placed too much faith in underpowered studies, and that readers should discount the chapter.
Gerd Gigerenzer criticizes from another angle: System 1's shortcuts are often smart strategies in real environments, not just errors. Kahneman also admitted that knowing about biases barely helped him avoid them; organizational procedures work better.
Five steps before a big decision. Leave small everyday choices to System 1.