Review: Thinking, Fast and Slow

Daniel Kahneman (2011)

Thinking, Fast and Slow is the behavioural economics book. In it, Kahneman brings out the three crucial distinctions underlying his view of how humans work:

  • System 1 thinking vs System 2 thinking
  • Rational economic agents vs Humans
  • Remembering vs Experiencing

System 1 and system 2 are anthropomorphisms of cognitive processes. System 1 processes are fast, automatic, effortless, and work with associations and averages. System 2 processes are slow, deliberate, effortful, and highly flexible, including responding reasonably well to teaching. System 1 is always active; system 2 is engaged only when required, and quite often only to approve the system 1 response.

While the characterisation of the systems might suggest that system 1 makes mistakes and it’s system 2’s job to detect and correct them, this is an overly-simple interpretation. Both systems make errors, and in characteristic ways. System 1 errors are errors of association, such as anchoring,1 halo effects,2 and availability heuristics.3 System 2 errors are errors of statistics,4 or of perseverance substituting questions requiring difficult mental work for easier ones.5

Both system 1 and system 2 can learn, but both require frequent and timely feedback to do so, and will only learn the parts of tasks for which they have timely and frequent feedback. Psychotherapists, for example, rapidly learn to manage clients’ emotions and to predict clients’ immediate responses with good accuracy. Despite this expertise, they remain unable to offer worthwhile predictions for clients’ prognoses over the course of a year, because the feedback for these predictions, if they are even routinely made, is sparse and much delayed.

Through contrasting rational economic agents with humans, Kahneman explains prospect theory, the work for which he is best known. Prospect theory is the foundation of behavioural economics, and offers two key insights: people are sensitive to changes in their fortunes more than to the absolute level,6 and that losses and gains are perceived differently. The implications of these insights are explored in detail, with examples ranging from toy economic games to employer-employee negotiations and peace talks.

In the latter part of the book, the two selves are introduced. These are the experiencing self and the remembering self. The former is the one which gets the joy from an ice cream or endures the last few minutes of a hard run; the latter is the one which reflects on the value of a life which affords opportunities for ice cream, or the satisfaction of staying in shape as you get older. Although the experiencing self is arguably the only one which should concern us, we consistently make choices to satisfy the remembering self.

Kahneman doesn’t delve deeply into the philosophy of the difference, but it’s worth noting that the longstanding battle between lower and higher pleasures in the utilitarian framework can be somewhat decomposed along the lines of the experiencing versus remembering selves, with the happy pig scoring high on minute-to-minute happiness through its life, and the sad philosopher scoring highly on overall life satisfaction. It also allows us to understand the pain of a bullshit sinecure.

Perhaps the biggest concern in the book relates to the reliability of some of the research findings presented particularly towards the beginning. Kahneman is effusive in his praise for much of the research he presents, and it means that findings for which the evidence is shaky, such as priming7 and ego depletion8 effects, are presented with as much gusto as those with a more solid empirical foundation. My advice to non-expert readers is to take the results of any experiment in which the manipulation was asking participants to imagine, read, or write something, or where they have to adopt some facial expression, with a hefty pinch of salt.

Overall, Thinking, Fast and Slow is a worthwhile read for anyone looking to get a decent understanding of modern thinking about how people make and explain their choices, and how well the explanations match underlying reasons. If you just want a tour of the various heuristics and biases to which humans are liable, Irrationality by Stuart Sutherland is older, but also a good choice.


  1. Anchoring occurs when a non-relevant numerical estimate interferes with an estimation task. If people are asked to estimate the probability of some event (e.g. a white christmas), their estimates will be lower on average if they’re presented with a random number generator which produces a low number (e.g. 6) rather than a higher number (e.g. 86). This occurs despite people being well aware that the random number generator is not providing helpful information for the task. 

  2. Halo effects refer to the tendency to think that people we like excel in all areas: your partner is not only attractive, they are also intelligent, funny, a better-than-average driver, etc. 

  3. The availability heuristic substitutes the ease with which examples can be called to mind for the frequency of occurrence. This is one reason why we think that rare-but-widely-reported events, such as terrorist attacks, are more worrying than common-but-seldom-reported events such as car accidents. 

  4. Of all the statistical quirks to which we fall prey, regression to the mean is perhaps the most interesting. Because our performance is a mixture of random factors (luck) and systematic factors (skill), the best estimate for a performance which follows a well above average one is slightly above average. This is neatly illustrated in the debunking of the ‘hot hand’ phenomenon in basketball. 

  5. A good example is the representitiveness heuristic, when we substitute how well a candidate entity typifies a category for the probability the candidate belongs to the category (while ignoring the background probability, or base rate). This is why we’re more inclined to think a politically-active, outspoken young lady is more likely to be a feminist bank teller than just a bank teller (despite the fact that the highest probability we can rationally assign to ‘feminist bank teller’ is the probability we assign to ‘bank teller’, because all feminist bank tellers are bank tellers). 

  6. Sensitivity to changes rather than absolutes is a generic feature of nervous systems, from adaptation at the single-cell level through to predictive-processing accounts of conscious experience. 

  7. Priming occurs when the retrieval of one idea is made easier by the presence of another idea. It is commonly demonstrated in stem completion tasks, where participants might complete the stem SO_P as SOAP rather than SOUP after hearing the prime ‘wash’. The effects which have suffered particularly poorly in replication attempts are social priming effects, where manipulations such as reading words associated with elderly people allegedly cause participants to walk more slowly. 

  8. Ego depletion, currently the subject of a fierce discussion, casts willpower as a finite resource akin to muscle power. The central idea is that exerting willpower on one task (e.g. doing a particularly difficult psychological task requiring sustained concentration) reduces the willpower available for other tasks (e.g. making healthy snack choices). It’s an appealing idea, but its evidence base needs to be adequately established.