Book Cover image

Thinking, Fast and Slow by Daniel Kahneman

Totally Scientific Rating: ⭐⭐⭐⭐

Get it on Amazon.

Book cover

Thinking, Fast and Slow by Daniel Kahneman

Totally Scientific Rating: ⭐⭐⭐⭐

Get it on Amazon.

This Book in a Minute (Or Less)

If you want to think better it helps to understand what affects the quality of your thoughts, what are the common traps we fall into, and how can we create the conditions to make better decisions. This book is perhaps the first and ultimate authority on decision-making, providing plenty of examples, many that will make you blush, on how far we are from the rational beings we like to feel proud of being.

Summary Notes

This book is all about two modes of thinking: fast and intuitive, or slow and deliberate. The two modes have advantages, situations where they make sense to be employed, and consequences, where we are led astray for relying on one instead of the other.

To improve our ability to understand the nuances of how we are thinking it's helpful to use rich and precise language to discuss it.

For this purpose, we should keep in mind how we are affected by biases and use heuristics.

Biases are systematic errors, even predictable in some circumstances, caused by a tendency to think about a subject in a pre-defined way, without considering new, or contrasting, information. Biases are one example of heuristics: mental shortcuts we use to reach the conclusions we need with the least effort.

Here are two examples:

Availability heuristic: we make a decision or a judgment based on how easyily can we recollect information about the options. Something is a better option if we can think of it, or its characteristics, with less effort.

Hindsight bias: we underestimate how much we were surprised by past events. If an event happens people overestimate the probabilities which they predicted of it happening. If an event never happens people underestimate the probabilities which they predicted of it ever happening.

These shortcuts can also be a sign of expertise. Valid intuitions develop as experts learn to recognize familiar elements in new situations that lead to acting in a manner appropriate to it.

That's why it pays off to understand how we are thinking about a problem. There are different qualities to shortcuts and we better prepare to recognize the difference.

Part 1 - Two Systems

System 1 is lazy, operates automatically and quickly. It generates impressions and feelings without effort, feeding the beliefs and choices of System 2. It's very good at what it does: accurate modeling of familiar situations, short-term predictions, and swift initial reactions to challenges.

System 2 is all about attention to detail and effortful mental activities. It takes the complex pattern of ideas provided by System 1 and constructs thoughts in an orderly series of steps. It's there to monitor our behavior and keep polite in social situations.

System 1 generates a stream of feelings, impressions, intuitions, and intentions and it's the job of System 2 to endorse these or not, which, most of the time, goes smoothly. From intuitions we get beliefs, and from intents, we get voluntary actions.

The advice is:

“Learn to recognize situations in which mistakes are likely and try harder to avoid significant mistakes when stakes are high.”

The premise of this book is that it is easier to identify other people's mistakes than our own.

The other budget you need to manage

Self-control and deliberate thought, apparently, draw on the same limited budget of effort ("processing power" of System 2). Using one may reduce our chances of successfully doing the other.

Several studies have shown that people who are simultaneously challenged by a demanding cognitive task and by a temptation are more likely to yield to the temptation. Just don't try to solve complex equations after resisting eating dessert.

Too much concern about how well one is doing in a task sometimes disrupts performance by loading short-term memory with pointless anxious thoughts.

One of the main functions of System 2 is to check and control thoughts/actions expressed by System 1, allowing some to be turned into behavior while suppressing/modifying others. Sometimes (a lot of times) System 2 is lazy and validates actions/intuitions of system 1 that are harmful and could have been discarded with a small effort. For example: when people believe a conclusion is true, they are also very likely to believe arguments that appear to support it, even if the arguments are unsound.

What you have been thinking is what you see

Words that you've seen before become easier to see again. The car that you are looking to buy suddenly is everywhere. It's a sense of ease that gives the impression of familiarity. The impression of familiarity is provided by System 1 and it is System 2 that will rely on that impression for a definitive judgment. This has its consequences: it's easier to spread falsehoods by constant repetition since the repetitions lead to familiarity, and familiarity is not easily distinguished from truth.

The normal state of mind is to produce opinions and intuition about almost everything.

If a satisfactory answer to a hard question is not found quickly, System 1 will find a related question (heuristic) that is easier and will answer it. For example: when tasked with judging probability, people judge something else and believe they have judged probability. When asked "This woman is running for the primary, How far she will go in politics?" we usually answer to "Does the woman look like a political winner?"

We may not realize that the first question is indeed difficult because we got an intuitive answer quickly come to mind.

“System 2 is more of an apologist for the emotions of System 1 than a critic of those emotions.”

Part 2 - Heuristics and Biases

System 1 deals very poorly with statistical facts, because it associates the outcome of an event as the reason to cause it.

Humans are pattern seekers. If we see a sequence of events go in the same direction a bit more than usual, we always assume that there must be a mechanical causality or someone's intention. We inherit this from our ancestors. General vigilance, being on the lookout for an environment change: lions appearing out of nowhere lead to people getting killed if at the first sight of them we dismiss lions as a statistical anomaly.

We are far too willing to reject the belief that much of what we see in life is random, and as a result, we end up with a view of the world around us that is simpler and more coherent than data justifies.

Anchoring effect: when people consider a particular value for an unknown quantity before estimating that quantity, and when estimating, people stay close to that first value considered.

The visit of a museum when pay-what-you-want without an anchoring effect would get on average: 64$. If 5$ was the anchor the average was 20$ and 143$ if the anchor was 400$.

Affect heuristic: making decisions based on our emotions on the topic.

For example, good technologies have few costs and bad technologies have no benefits. In reality, there are trade-offs. This can be applied to the majority of categories where we have preferences.

Availability cascade: a self-sustaining chain of events, which may start with the media reports on a relatively minor issue and lead up to public panic and large-scale government action.

This has the consequence of overriding and/or hiding bigger risks, or even diverting the resources into a problem that is not as important as the ones that everyone already knows about and doesn't make the news.

A most important tool in the shed

Base rate: the proportion of a type of thing in the complete sample.

Three red balls and five blue balls make the base rate of blue balls 5/8, and it would be foolish to expect a higher probability when taking one ball at random from the bucket of 8. And yet...

When we try to guess the profession of a subject given a description of him and a list of possible professions we tend to go with the stereotypes we get from the descriptions (and the veracity of it) and ignore the base rates. For example, someone who wears glasses and likes to read comic books might be a programmer OR might be working on the most common profession in the world (which I hope is not yet programming).

When we describe a person with an adjective we are fitting her to a larger group than describing her with two adjectives. If two adjectives make sense together we still must take into account that one person has a bigger group that can belong to if it had one adjective.

Again we replace a difficult question (the probability of a profession) with an easier one (the similarity between description and stereotype).

In case of doubt on the quality of the information we have there is only one thing we can do to improve our position: stay close to the base rates. But, unfortunately, people will, more often than not, disregard information from base rates if it conflicts with other beliefs.

Conjunction fallacy: people judge the conjunction of two events as more probable than one of those events separate.

Adding details to scenarios makes them more persuasive, but less likely to come true. This fallacy is also affected by how the information is solicited: if asked "what was the percentage" we get much more errors than "how many of the 100 participants". A visual representation is a possible explanation.

Why rewards for improved performance work much better than punishment for mistakes.

Regression to the Mean explains why flight cadets improved performance after being shouted angrily at them and got worse when they were praised. It seems a contradiction with the previous header, but let's dig a little deeper.

A cadet praised had a better than average performance, probably lucky on the attempt, and thus likely to deteriorate the performance regardless of whether he was praised or not. On the other hand, a cadet who was shouted at by the instructor (only) after a particularly bad performance was likely to improve regardless of what the instructor did.

The instructor will end the day believing verbal abuse is the magical formula for great results in training, while in fact 'accidental performances' happen, and with small samples, they are not a true indicator of cadet skill.

“Correlation and regression are two different perspectives of the same concept: Whenever the correlation between two scores is imperfect, there will be regression to the mean.”

Now consider the following scenario: "Highly intelligent women tend to marry men who are less intelligent than they are". Start a debate and hear people's opinions on this.

Some think intelligent women want to avoid competition, or it's because intelligent men don't want to compete in a couple with an intelligent woman. Other crazier explanations will come up.

We can rephrase the initial scenario as: "The correlation between the intelligence scores of spouses is less than perfect", and that's just as true, even algebraically equivalent, as the first scenario but way more boring.

So what's the nugget here?

If the correlation between the intelligence of spouses is less than perfect (and men and women on average do not differ in intelligence) then it is a mathematical inevitability that highly intelligent women will be married to husbands who are less intelligent than they are (and vice versa).

“Our mind is strongly biased toward causal explanations and does not deal well with 'mere statistics'.”

The prediction paradox

Predictions are another thing humans are not very good at. We usually interpret evidence at our will and predict a slightly different event without noticing that the question we are answering is not the one we were asked.

Intuitive predictions tend to be overconfident and overly extreme and is the job of System 2 to apply some sense to them. Extreme predictions and a willingness to predict rare events from weak evidence is a feature of System 1.

If you expect your predictions to be of modest validity, you will never guess an outcome that is either rare or far from the mean.

Part 3 - Overconfidence

Narrative fallacy: flawed stories of the past (sometimes made up by us) shape our views of the world and our expectations of the future.

The ultimate test of an explanation is whether it would have made a given event predictable in advance. The fact that many of the successful human endeavors involved choices/decisions further tempt us to exaggerate the role of skill and underestimate the role of luck.

Once you adopt a new view of the world (or of any part of it), you immediately lose much of your ability to recall what you used to believe before your mind changed. The narrative we believe changes and so do we.

“This illusion that one has understood the past feeds the illusion that one can predict and control the future.”

A sixth sense

Intuition is only recognition. We recognize a situation, and with it, a potential right answer to follow.

There's an issue though. The confidence we have in our intuitions is not a reliable guide to their validity. And yet, we can work on improving our recognition of situations, and in turn, our intuitions.

There are two main requirements to acquiring a given skill:

  • a regular and predictable environment;
  • enough opportunities to learn these regularities through practice.

We need the stability of the environment so we can verify a given answer to always solve a given situation. Once we trust the pair situation-answer we develop an intuition for future challenges.

To develop this expertise we depend on the quality and speed of feedback, as well on sufficient opportunity to practice. The different speed of feedback makes it very visible in the diagnostic of different clinical expertise's.

Anesthesiologists enjoy good feedback conditions in contrast to Radiologists. The effect of anesthetics is likely to become quickly evident, while radiologists get little information about the quality of their feedback.

Planning fallacy: plans/forecasts that are unrealistically close to a best-case scenario and/or could be improved by consulting statistics of similar cases.

The proper way to get information from a group of people is not with a public discussion, but by confidentially collecting each person's judgment. Take this into account when is the group forecasting its execution success vs an individual.

When pitting the "inside view" (how much time do we think it will take us to finish this project) vs the "outside view" (how much time do groups like ours take to finish a project like this) the outside view always loses, but it's probably one of the best tools to get a forecast closer to the end result.

Overconfidence sometimes hurts, sometimes moves the world forward

What if the people who have the greatest influence on the lives of others are the optimistic, overconfident, and those that take more risks than they realize?

Entrepreneurs are overconfident risk-takers. Even if they are not sure how, or if, they'll succeed they most certainly believe they have their destiny in their hands. They might not be entirely right, as the outcome of startups depends as much on competitors' behavior and market changes as on their effort.

By focusing on what they know best, their plans, actions, and immediate threats/opportunities, they underestimate the importance of the outside world on their success.

Overconfidence comes from System 1, it can be tamed, but not eliminated. It's much more the result of how coherent is the story we tell ourselves and not so much about the quality of information that supports it. You can adjust the dose of optimism with premortems: Imagine we implement the plan as it now exists and in 1 year the outcome is a disaster. Spend 5 to 10 minutes to write a brief history of what happened and see what flaws came upon your project and forecast.

Part 4 - Choices

Let's say you have to choose between:

  • a 50% chance of receiving 1M€ or a 50% chance of receiving 8M€;
  • receiving 4M€;

To arrive at the value of each choice you would use something Bernoulli came up with called utility function. It explains why the poor buy insurance and the rich sell it. Since it's not the same to lose the little you have as losing the same amount while living in a mansion, it's worth paying insurance to transfer the risk of ruin to the richer person.

The theory of Bernoulli was that the higher the utility of a proposal the happier a person would be taking it. The problem was it was a flawed theory, and no one seemed to want to evaluate it and adjust their models.

This theory didn't take into account the past wealth of a person.

Let's think about a new choice:

  • a 50% chance to own 1M€ or a 50% chance to own 4M€;
  • own 2M€;

And now consider two individuals having to chose: Person A has 1M€ in the bank, and person B has 4€M. Person A can only win or stay the same, while person B can only lose or stay the same. They have the same choice and expected wealth at the end, but one will likely want to risk it and the other wants to play safe.

People become risk-seeking when all their options are bad.

“We just like winning and dislike losing, independently of our wealth, and we most certainly dislike losing more than we like winning.”

We also have stronger reactions to outcomes caused by actions than by inaction. A coach whose team is losing is expected to make an action as default, even if the best strategy and players are in the game. If he doesn't do anything and the team loses it can only be because of incompetence to produce the changes that everyone else was seeing after the fact.

We hate more losing than we love winning

When evaluating changes we give more weight to the disadvantages than the advantages, inducing bias that favors the status quo.

The endowment effect attempts to explain the willingness to buy or sell a prized object (a collector's item for example) depending on if we already own it or not.

It's not as obvious as it seems. If we own it, we consider the pain of giving it up. If we don't, we consider the pleasure of acquiring it.

The pain of losing and the pleasure of acquiring are not symmetrical. We would ask for way more money to sell X than we would be willing to pay for it, simply because we would find it more painful to lose X than pleasure from getting it.

The news is full of bad news

Our brains (as on other animals) contain a mechanism that is devised to give priority to bad news. It's the system 1 responsibility and there's nothing of the sort for good news.

“A single cockroach will completely wreck the appeal of a bowl of cherries, while a cherry will do nothing at all for a bowl of cockroaches.”

You are more motivated to avoid causing bad impressions than to pursue causing good ones, simply because it's harder to recover from a bad image.

Weights and Probabilities

The weights we put on outcomes are not identical to the probabilities of these outcomes. Possibility and certainty have similar effects in the domain of losses, and again, that's why insurance exists.

A loved one facing surgery with a 5% risk of a bad outcome feels worse than half as bad as if the risk was 10% and we want to play it safe. A lottery with infinitesimal odds of success triggers the opposite effect, we think we can win it, and of course, are fine with risk.

The more vivid descriptions produce higher decision weights for the same probability. We are convinced by scenarios with more details as if that was a sign of a more realistic outcome.

Part 5 - Two Selves

Decision utility is how much "wantability" a given alternative presents to us if pursued. We can use the expected decision utility of alternatives to better decide what is best for us, but more often than not, discard what is worst for us.

Every time there are different costs for the same expected decision utility a mistake is bound to be made.

If a person has to receive painful injections every day and the pain is the same every time we should pay equal amounts per injection we could avoid. Right? The truth is we tend to pay more per injection reduced if we go from 6 to 4 than from 20 to 18. The difference is absurd. If the pain is always the same and there are always two injections that are removed, why is the utility different depending on the past number of injections?

The trap of confusing experience with memory

Would you prefer to endure a very painful, but short, procedure, or a not so painful, but long, procedure?

There's reason to believe we would prefer the one that would leave a better memory.

The experiencing self doesn't have a voice. The remembering self is sometimes wrong, but is the one keeping score and governs what we learn.

Even if we are to increases a procedure in duration artificially as to not end with the highest pain, it would tailor our last memory of it being a slight discomfort and not the maximum pain felt during it.

“What we learn from the past is to maximize the quality of our future memories, not necessarily of our future experience.”

This same bias happens for good things, but in reverse: we stick with the peak feeling at the end of the experience, and we tend to favor short periods of intense joy over a long period of moderate happiness.


System 2 is who we think we are. It makes judgments and choices. It's an endorser of System 1 antics and its feelings.

System 1 is the originator of everything we do. It maintains a rich and detailed model of the world in our associative memory and its best trick is probably how it can distinguish surprising from normal events in a fraction of a second.

To try to counter the errors that originated in System 1 there's only one recipe:

  1. recognize that you are in a cognitive minefield;
  2. take a step back and slow down;
  3. ask for the support of System 2.

If you enjoyed this, there are many others to come.

This Week's Worth preamble

This is my (almost) weekly newsletter.

You should consider be part of it to stay on top of new articles and other stuff I have yet to figure out. Yup, I'm improvising here.