Thinking, Fast and Slow: Part 3, Chapter 22 Summary & Analysis

Kahneman next writes about his collaboration with Gary Klein, a colleague who did not agree with his work on experts and was deeply skeptical about the value of using rigid algorithms to replace human judgment. Klein had studied the expertise of firefighters and analyzed how experienced professionals develop intuitive skills. Kahneman invited him to join in an effort to try to discover when one can trust an experienced professional who claims to have prediction abilities.

Kahneman’s view of intuition was formed by observing the illusion of validity with his own work and Meehl’s work. Klein’s views were shaped by his work studying firefighting teams. The commanders could draw on patterns that they had learned and could intuit quickly the best option to fight a fire, without having to generate other options.

Klein calls this the recognition-primed decision (RPD) model, which also applies to experience in other domains like chess. In it, a tentative plan comes to mind by the associative memory in System 1. The next phase is a deliberate process in which the plan is checked by System 2. Kahneman believes that this kind of intuition is really recognition of information stored in memory.

Certain types of intuitions are stored in memory very quickly: a dubious dish that leaves us hesitant to return to a restaurant, or a bad experience on a certain street that causes us to remember it when we pass. This emotional learning is quick, but expertise takes a long time to develop. An expert chess player can understand a complex position at a glance, but it takes years (studies show at least 10,000 hours) to develop that kind of ability.

Becoming an expert at chess can be compared to learning to read. A first grader works hard to recognize letters and parse syllables, but adult readers can perceive entire clauses automatically. Chess is harder than reading because there are more letters in the “alphabet” of chess and longer “words,” but eventually chess masters can read situations instantly.

Kahneman and Klein realized that their disagreements stemmed from the fact that they had different experts in mind. Klein had worked with firefighters and nurses, while Kahneman worked with financial traders and political scientists. Klein argued that true experts know the limits of their knowledge; Kahneman argued that there are many “pseudo-experts” who don’t know what they don’t know.

Kahneman concluded that there are two conditions to acquire real expertise: an environment that is sufficiently regular so as to be predictable, and an opportunity to learn these regularities through practice. Bridge and poker players, nurses, physicians, athletes, and firefighters all satisfy these conditions. But these professionals use highly valid cues to make predictions, while political scientists make long-term forecasts that often have no valid cues.

Meehl’s work demonstrated that algorithms do better than humans, but even the algorithms did not have extremely high accuracy because the situations it evaluated did not allow for that accuracy. Statistical algorithms can find weakly valid cues and use them consistently, but if a strong predictive cue exists, humans can usually find it. Thus, it is wrong to blame anyone for failing to forecast accurately in an unpredictable world. But it does seem fair to blame professionals for believing they can succeed at something impossible.

There are a few conditions one can meet to become an expert, and those are immediate and unambiguous feedback as well as sufficient opportunity to practice. Chess is good for these conditions. But surgeons can be much more proficient in some operations than others. Psychotherapists get good feedback from patients as they speak with them, but do not have a chance to get feedback on the success of long-term treatments until many years later, and the results are often ambiguous.

The conclusion that Kahneman and Klein came to is that, for the most part, it is possible to distinguish intuitions that are likely to be valid from those that are likely to be bogus. If the environment is regular and the judge has had a chance to learn its regularities, System 1 will recognize situations and generate quick and accurate predictions and decisions.