Thinking Fast and Slow (book by Kahneman)

Thinking Fast and Slow

Daniel Kahneman

Thinking Fast and Slow

 Daniel Kahneman is a psychologist who shared a Nobel Prize in Economics for the work he did (with Amos Tversky) on the processes people use to make decisions. This work was the foundation of what is now called behavioral economics. This book explains their view of how we see the world and react to it. This is a big book with some big ideas. Here is a quick summary of some of these big ideas. (1) We react to the world with two thinking systems. The first is quick, easy, efficient and always functioning; it makes mistakes of a certain kind. The second system is slower, effortful, and what we usually associate with “thinking”, but it makes a different class of mistakes. Understanding the difference helps know when to “go with your gut” and when to “slow down and think about it”. (2) People are not as rational as they think they are, or as rational as some economic models assume they are. People are both rational (logical) and irrational (intuitive), but they do not know when they are irrational. Policies that depend on consistent logical behavior by people fail to account for their human behavior. This contrast might be called Econs vs. Humans. (3) We operate with two “selves”. The experiencing self is mostly concerned with the present, while the remembering self is concerned with the past. People routinely remember the past incorrectly due to failure to consider the duration of events (we don’t remember long painful episodes), and thus undertake things that their “experiencing” self would reject. This same problem makes it difficult for us to practice empathy with others in many circumstances.

Two systems for thinking

 This first section explains the idea that we have two processes for perceiving and reacting to our experiences. Called System 1 and System 2 as a way to contrast the two processes, it is critical to understand that there are not literally two systems in the brain processing inputs. It is just a way to represent the contrast between two types of cognitive behavior.

 

System 1 is fast, reactive, attentive, illogical, and easily misled. This is the fast in the title. System 2 is slow, logical, but inattentive. System 1 is always ON; its primary role is surveillance and defense. If you are walking in tall grass and suddenly jump out of the way of a snake; System 1 saw the snake and made you move before you were consciously aware of the snake. Advertising and political slogans appeal to System 1. System 2 can be dormant for long stretches, but is called on when something seems wrong. System 2 is problem-solving in the sense of diagnosis, hypothesis generating active thinking. When you say “On second thought”, you are invoking System 2.

 

One of the origins of this work was the question: are people intuitive statisticians? There are a number of aspects of life that are intuitive (grammar is learned intuitively), but statistics turns out not to be one of them. In other words, people have a poor grasp of probability and its consequences. So in many cases, people make decisions inconsistent with statistical logic. These are not always reflexive decisions either. In many cases, there is no time pressure to overcome logical thinking. Careful examination also shows that these decisions are not the result of emotion dominating logic; these are the result of systematic cognitive “errors”.

 

Intuition is an extremely important cognitive function, and System 1 presumably evolved to increase our survival prospects. One of the early findings was that intuition was a form of mental recognition. People with greater expertise have greater stores of experience to use in recognition. Herbert Simon noted that “The situation has provided a cue; this cue has given the expert access to information stored in memory, and that information provides the answer. Intuition is nothing more and nothing less than recognition”.  Intuition is very useful when it works, but a problem when it leads to actions that don’t work – or lead to harm.

 

Another feature of System 1 is that it is automatic in the sense that it is always on (it operates whether you are awake or asleep). It does not require any “effort” to engage. System 2 is on when you are awake, but minimally operating until engaged. The phrase “pay attention” expresses the call to switch from System 1 alone to using System 2 as well. System 2 requires considerable effort, so it is difficult to pay attention to more than one thing at a time or to anything for extended periods. System 2 is an energy drain, and consequently we use it when we have to – but no more than that. In the end, everybody is lazy (this is biologically sensible – save your energy for when you really need it). There are many demonstrations of the energy drain that paying attention causes. One of the best can be found at attention test . When we focus enough attention on something, we may become blind or deaf to our surroundings. Paying attention to one thing distracts us from others. We are unaware that we are blind under these conditions.

 

In general, research showed that many of our decisions are made by System 1 and System 2 usually goes along with that decision. In other words, many of our decisions are intuitive and we don’t really think about them. This exposes us to a set of risks that result from the effects of cognitive conflicts, illusions, and useful fictions.

 

Cognitive conflicts arise when we are asked to examine things which are out of place. For example, if you are asked to read a list of words related to color where the words themselves are in color, you will read the word RED much faster than the word RED. There is no conflict when the color and word are coherent, but a conflict when they are not. This is an example of familiarity – we can see and understand what we are familiar with much more easily than what is new to us. System 1 prefers to examine what we are familiar with, and ignore what we are not. Illusions are perceptions that are not overturned by information. One common illusion is shown below.

 

 

 

 

The upper line looks longer, but it is not. Even when you know they are the same length, the upper line looks longer. That is an illusion. System 1 sees and acts on the illusion. System 2 knows that there is an illusion and acts on the basis of reality. Useful fictions are used to enable expression. For example, the book and this summary treat Systems 1 and 2 as if they were things (or people), but they do not exist. It is a useful fiction to treat them this way because it makes the concepts easier to understand. We create useful fictions both in learning or understanding events and in expressing them. Sometimes, the analogy is close enough that decisions work, but sometimes the analogy is essentially false. Based on the false expression of circumstances, we make faulty decisions.

 

Central to understanding the interplay between Systems 1 and 2 is the energy requirements of paying attention. The author suggests that you do a simple test of this. Go on a walk with a friend and while walking ask them to multiply in their head two numbers (like 23 and 51) right then. Almost certainly, they will stop walking to carry out the task. The brain does not have enough capacity to walk and multiply bigger numbers at the same time.  When people are under heavy System 2 loading, they begin to make more “selfish” choices. Research suggests that there is a single pool of mental energy which is depleted by thinking, physical and emotional effort. As the pool of mental energy is drained, a person begins to cut corners to preserve remaining energy. The energy referred to here is real metabolic energy – not some sort of metaphoric energy. The brain’s primary fuel is glucose and the brain is one of the major consumers of glucose in the body and generators of heat. A bout of heavy thinking can measurably decrease the concentration of glucose in the blood. Because System 2 creates this drain, it is very efficient to defer as much to System 1’s automatic associative thinking in order to save energy.

 

System 1 uses an association-driven approach to decision making, where the definition of decision is very broad. Associations between ideas are not treated as ideas by System 1 but realities. If a person is shown two “disgusting” words, they grimace as if experiencing the reality of disgust. Two “happy” words induce a smile (these gestures may not be visible to the eye, but can be detected with instruments). Over a person’s life, they build up a huge network of associations. This network enables a phenomenon called “priming”. Priming occurs when an input triggers an association network such that the next bit of information leads to a decision. The book gives the example of SO_P. If you are shown the word EAT first, then you know that the word is soup. But if you are shown the word CLEAN first, you know the word is soap. The flood of information we experience is constantly priming us for decisions to come. For most occasions, the priming is useful and we make decisions automatically. Driving a car is mostly an automatic experience, where events prepare us for decisions that we will need to make. We knew that person was going to change lanes without signaling.

 

Such priming acts in many less obvious ways. If you make somebody ashamed, they are more likely to make the word SOAP out of the incomplete word S__P than soup. And the effect can be very specific. If you make them tell a lie by email, they will subsequently prefer to buy soap over mouthwash (to wash their hands). If they deliver the lie over the phone, they prefer the mouthwash (to wash their mouths). This is one of many such studies, each of which shows that we are easily influenced by our history and circumstances into taking predictable actions. Our self-image is that of independent thinkers who take each instance as it comes and make decisions based on the relevant information. This self-image is not what is observed in these experiments.

 

What’s more, people are unaware of these tendencies. The idea you should focus on, however, is that disbelief is not an option. The results are not made up, nor are they statistical flukes. You have no choice but to accept that the major conclusions of these studies are true. More important, you must accept that they are true about you.*

 

In a practical demonstration of the impact of priming on people, an office kitchen had a box for people to put money into along with a price list for coffee, tea etc. The money collected did not always cover the consumption observed. One day a small banner photo was placed right above the price list showing a pair of eyes. The photo was changed each week, sometimes showing flowers and sometimes showing eyes. Collections were much higher in weeks showing eyes than in weeks showing flowers. The suggestion that someone was watching was sufficient to trigger payment.

 

If association is a main drive of decision making, the second main influence is “cognitive ease”. In a way this is the inverse of cognitive dissonance. If something makes sense, we are at ease. A false statement creates a feeling of unease that cues us to look more closely. The determination of ease is done by System 1. Information that is easily and quickly processed is viewed as more valid than information that is difficult to process. A statement that is difficult to read due to font, contrast or color is judged less true than one that is easier to read. Thus short simple sentences are more believable (bulleted lists might be even better). Ease can derive from feelings of familiarity, truth or effortlessness. These feelings can be based on experience and lead to good decisions. They can also be based on inappropriate analogies with past instances and mislead. For example, a repeated false statement will become familiar. When probed later, you might recognize the statement as familiar and accept it as true.  Combining a true statement with an untrue statement will result in you attributing “more” truth to the whole because you will create an illusion of truth about the whole. Expressed differently, humans prefer repetition to novelty, so false repetition may be preferable (to System 1) over a novel truth.

 

Anytime that System 2 is engaged, cognitive ease drops. Thus, anytime we really need to think about something, we are less comfortable. But this has an interesting corollary. There are a variety of cognitive puzzles that can be presented to people. One classic is:

In a lake, there is a patch of lily pads. Each day, the patch doubles in size. If it takes 48 days for the patch to cover the whole lake, how long would it take to cover half the lake? 24 days or 47 days?

A group of Princeton students was presented a test containing puzzles like this, but half were presented the puzzle in a crisp clear print and half were presented the puzzle in a faint but legible print….90% of the students who saw the CRT in normal font made at least one mistake in the test, but the proportion dropped to 35% when the font was barely legible. You read this correctly: performance was better with the bad font. Cognitive strain, whatever its source, mobilizes System 2, which is more likely to reject the intuitive answer suggested by System 1 (the correct answer to the puzzle is 47 days).

 

One interesting application of cognitive ease relates to a “sense of knowing”. In effect, people often know when a problem can be solved even before they actually know the solution. People in a good mood are more accurate in their perception than people in a bad mood….when in a good mood, people become more intuitive and more creative, but also less vigilant and more prone to logical errors.

 

The importance of familiarity is also shown by the speed by which abnormal events are detected. System 1 is monitoring the environment and making predictions about what will happen next. Events that are consistent with the predictions cause ease, but events contrary to prediction cause surprise. Such surprise has been shown to arise in tenths of a second. This is part of a cycle of cause-and-effect. While the standard logical thinking about this is that “causes induce effects”; the mind treats cause-and-effect as a symmetric condition. Causes have effects and effects have causes. Presented with an effect and a list of coincident events, System 1 tries to assemble the information into causes-and-effect. Nicholas Taleb’s book, The Black Swan, describes reporting on Bloomberg News on the day that Saddam Hussein was captured. Early in the day, prices rose and the headline was U.S. TREASURIES RISE; HUSSIEN’S CAPTURE MAY NOT CURB TERRORISM. Prices fell later that day and the headline was U.S. TREASURIES FALL; HUSSEIN’S CAPTURE BOOSTS ALLURE OF RISKY ASSETS. One event “caused” opposite outcomes. More likely, there was no connection between Hussein’s capture and price changes, but there is a need to believe that there must be a cause. The most prominent coincident event must be the cause – especially if they precede the event of interest. People are very adept at creating cause-and-effect meaning from their experiences. This expectation of causation prevents us from seeing events in a probabilistic context. System 1 seeks causation, but it is System 2 that uses statistical reasoning. The expectation of causality combined with the constant prediction of System 1 creates a wonderful ability to jump to conclusions.

 

One expression of cognitive ease is exaggerated emotional coherence, better known as the halo effect. The effect arises when we use the answer to one question (Ann is a good speaker) to answer another question for which no information is available (Is Ann generous?). For System 1, this is a quick way to solve problems based on an assumption that all things associated with something (Ann) are correlated. Assignment of this overarching status on something eliminates inconsistency in my “feelings” about Ann and puts my mind at ease. If I am seeking to maintain valid opinions about a topic, I need to de-correlate my information. This may require that I deliberately create “independent” thoughts about the topic. This is a function of System 2. One common way to improve decision making is to ask for other opinions. The goal is to get an independent (uncorrelated) perspective on a situation. This a potential benefit of engaging a group in a decisions. But group discussions often build correlation in the group and degrade the independence of thought. A simple rule can help: before an issue is discussed, all members of the committee should be asked to write a brief summary of their position….The standard practice of open discussion gives too much weight to the opinions of those who speak early and assertively, causing others to line up behind them.

 

The basics of making judgments involve some basic assessments. Basic assessment is a broad topic but two specific aspects of interest are “sets and prototypes” and “intensity matching”. One of the basic assessments that System 1 makes is where to categorize a situation. This is a sophisticated version of pigeonholing. Almost anything can be pigeonholed into an existing category based on previous experience. Intensity matching is a process where the answer to one question is obtained by answering a different question involving an emotional element. For example, the answer to the question “how much would you contribute to save Honduran frogs” would be answered by considering the question “how much do I care about frogs”.  This process matches the intensity of my commitment to frogs with my willingness to contribute and supplies a quick answer.

 

Most judgments are made using heuristics (which are simplified procedures, technically speaking) but in this context are substitutions of one question for another. The major heuristics are:

• The 3D heuristic – This heuristic interprets a situation based on a perceptual illusion. For example, you look at a photograph and interpret it in 3D, even though it is in 2D. This is an illusion that can’t be overcome with information.
• The mood heuristic for happiness – The heuristic substitutes an evaluation of events for the answer to the question – am I happy? The book cites an experiment where student’s answers to the following questions were uncorrelated.
  • How happy are you these days?
  • How many dates have you had in the last month?

But asked in the opposite order, the answers were highly correlated. In the latter case, feelings about recent romantic experience substituted for general happiness. The first answer primed System 1 for the second question.

• The affect heuristic - In an affect heuristic, people let their likes and dislikes determine their beliefs about the world. For example, if you don’t like red meat – you probably don’t think it is safe.

 

All of this description of System 1 makes it seem like a danger, but it is a vital part of good function. System 1 is the result of a constant learning process, guided by System 2. System 1 handles many routine functions with minimal effort while monitoring the environment for threats. When you react to something “before” you are even aware of it; that is System 1’s work. Most of the decisions made by System 1 are justifiable, useful and effective. But on occasion, the results are objectively wrong.

 

Logical and Intuitive Interplay

Perhaps the most important bias that people have relates to meaning; events must have meaning. System 1 is quite good at finding meaning because it can quickly make associations between things. If I tell you that the US counties with the highest incidence of kidney cancer are rural, sparely populated, and located in the west, south and Midwest, you can probably identify 2 or 3 reasons that this would be true. If I then inform you that the counties with the lowest incidence of kidney cancer are also sparely populated, and located in the west, south and Midwest, you begin to realize that your explanation may not be valid. The key observation is bth “kinds” of counties are sparely populated. One or two cases of kidney cancer will really stand out in a small population. We routinely draw incorrect inferences when we are analyzing a small set of data.

 

This kind of mistake often has important policy consequences. For example, the Gates Foundation undertook to understand what the best schools had in common. After some study of schools in Pennsylvania, one of the most important factors turned out to be school size; small schools performed especially well. The Foundation invested a large amount of money in making schools smaller. If the Foundation had searched for the characteristics of the worst schools, they would have found out that they were smaller than average. If small schools were compared to large schools, what would have been clear is that small schools have a larger range of performance. More great schools will be small and so will more awful ones.

 

One example of the effect of a small sample is seen in the effect of anchors. One of the famous experiments use to demonstrate anchoring involved setting up a “wheel of fortune” wheel with numbers from 1 to 100, but rigged to stop at 10 or 65. The audience could see the wheel being spun and settle on a number. They were asked to write down the number then were asked two questions: one was “What is the percentage of African countries in the UN?” The average response of people seeing the 10 was 25% and the average for those seeing 65 was 45%. Though the subjects knew that the number was unrelated to the question, their answers were biased by the number. Similar results have been obtained in many environments making the anchoring effect one of the best tested psychological effects. Anchoring seems to be a form of priming combined with range finding. Even though we know some anchors are irrelevant, our associative thought processes use the anchor as a first estimate.

 

Anchoring is important in many real-life situations. Estimates of probable value, duration and outcomes of many types are subjected to anchoring effects. People selling things know the value of anchoring a high price and buyers know the value of anchoring a low price. Shoppers in one study bought twice as many cans of discounted soup (7) when there was a specific limit on purchasing (12) than when there was no limit. They did not approach the limit in either case, but the anchor of 12 raised buyers’ notion of how much to buy.

 

In a way, anchoring is a form of cognitive easing; the brain uses a recent piece of information to start estimation. Another form of easing is called “availability”. While the human memory is apparently quite large, we actually do not know all the details that we need to know to make decisions. So a common tactic is to equate two different kinds of information and make a decision based on the information we are familiar with. The information we are familiar with is “available”. For example, you might be asked what percentage of Hollywood marriages end in divorce. You would then try to recall instances of Hollywood divorce, and estimate based on HOW EASY it was to recall instances. You would not do an inventory or estimate the number of Hollywood marriages that do not end in divorce or recall instances of marriages remaining intact. It is easy to recall high profile divorces, so you answer with a high estimate. Similarly, a recent plane crash alters your estimate of flying safety because a recent incident comes to mind easily.

 

The role of ease is demonstrated by a counterintuitive experiment. People were asked to record 12 instances of their behaving assertively. Others were asked to record 6 instances. Both groups were then asked to rate their assertiveness. The group asked to record 12 rated themselves much less assertive than the group asked to record 6. The explanation related to the difficulty of coming up with 12 instances; it was hard enough to decrease people’s impression of their own assertiveness. If subjects in these sorts of experiments are given an explanation for the difficulty (for example, most people struggle to come up with 10 examples of anything), they increase their rating of their assertiveness to nearly the same level as the subjects asked for 6 examples. In fact, the explanation need not be particularly rational; any excuse will do. Availability bias is a System 1 attribute; switching the focus to content information engages System 2 and disengages System 1. A number of conditions promote availability bias with the most interesting being a “feeling of being powerful”. If you remind people of times they have been powerful, you can measure their increased confidence in their own intuition.

 

Availability has a big effect on our risk assessment. In effect, if we are asked whether to take a particular chance, we consult our memory for examples of the situation. If we can remember an instance easily, we use that ease to determine whether to take the chance; if it easy to remember a bad outcome, we judge it a high risk. In a very similar way, we use availability to make our assessments of probability. When examples come readily to mind, we assume that that is because they are more probable. So right after airplane accidents, people assume that air travel is less safe than when an accident is a distant memory.

 

We make some odd judgments based on lazy logic. Consider which of these statements is more probably true.

• A massive flood somewhere in North America next year, in which more than 1000 people will drown.
• An earthquake in California sometime next year, causing a flood in which more than 1000 people drown.

Most people judge the second statement to more probable, apparently because it contains more detail. This exposes another kind of logic flaw which is easier to understand with the example: Which is more probable?

• Mark has hair
• Mark has blond hair

Blonde headed people are a fraction of the people with hair. There really can’t be more people with blond hair than people with hair. As the situations get more complicated, we seem to tell ourselves stories using the information provided. The “better” the story, the more probable we think the outcome is. These are examples of failing to account for the base rate of events. Giant floods and big earthquakes are rare in North America, so the chance of a rare big earthquake leading to a rare big flood in a part of North America seems to require a lot of bad luck. But all that detail seduces our two Systems; this is the narrative fallacy.

 

Our inclination to assume that there are clear cause-and-effect relationships between events causes us to make a different kind of error. Looking back, we can see all the signs that predicted the outcome that actually happened – and we are blind to signs of other outcomes. This is known as hindsight bias and makes it hard for us to judge decisions after-the-fact. But in a world where some outcomes are almost completely unpredictable, this means that people will be harshly judged for their decisions as if they (and their decisions) were flawed. As a defensive response, standard operating procedure and bureaucracy arises. When decisions turn out badly, the person can point to the fact that they were following procedure. This will result in risk aversion (see loss aversion later).

 

Taken together, you should pay close attention when a person expresses uncertainty about a plan, but pay less attention when people express confidence. We place excess faith in this confidence which may be entirely based on the coherence of the story that the person has constructed for themselves. Compelling stories may just be well crafted stories. This observation was expanded based a metaphor of “foxes” and “hedgehogs”. Hedgehogs “know one big thing” And have a theory about the world; they account for particular events within a coherent framework, bristle with impatience toward those who don’t see things their way, and are confident in their forecasts. They are also especially reluctant to admit error. For hedgehogs, a failed prediction is almost always “only off on timing” or “very nearly right”. They are opinionated and clear….Foxes, by contrast, are complex thinkers. They don’t believe that one big thing drives the march of history (for example, they are unlikely to accept the view that Ronald Reagan single-handedly ended the cold war by standing tall against the Soviet Union). Instead the foxes recognize that reality emerges from the interactions of many different agents and forces, including blind luck, often producing large and unpredictable outcomes. Phillip Tetlock, who developed this metaphor for decision making, tested political and economic experts for their ability to predict events in their fields and asked about the basis of these decisions. Most were simple predictions of decrease/no change/increase, and almost 80,000 predictions were collected. The results were devastating. The experts performed worse than they would have if they had simply assigned equal probabilities to each of the potential outcomes. In other words, people who spend their time, and earn their living, studying a particular topic produce poorer predictions than dart-throwing monkeys… When foxes and hedgehogs were compared for accuracy, foxes scored slightly better; they were very poor instead of awful.

 

Paul Meehl carried the examination of expert opinion one step further by comparing the predictions of experts given access to a range of information with a formula developed using a subset of the data. The formula made better predictions than the experts across domains like college grades, parole violations, cancer survival, evaluation of credit risk, and training airplane pilots. Many domains can be analyzed to create a formula that performs as well as a person does, in part because the formula is not distracted by details or extraneous events. The author had a chance to apply this himself when he was ordered to revamp the Israeli Army’s interview procedure. Traditionally, soldiers were assigned after a 20 minute open interview, but follow up studies showed that the interview did not predict fitness to the different roles. Kahneman had read Meehl’s work and decided to use a 6 stage interview with fixed topics. Questions were factually oriented and interviewers would score the interviewee on each topic before moving to the next topic. Ultimately, these scores were fed into a simple formula that directed the assignment. This proposal caused a backlash, so Kahneman added a final evaluation; the interviewer could add a final overall score after all other scores were computed. Two things were learned from the change. First, soldiers were a better fit to their assignments in the new system based on the 6 scores and the formula. Second, the overall score provided at the end of the defined process produced almost as good a prediction as the formula. In other words, a disciplined interview process set the stage for either the formula or the interviewer’s intuition to make a valid prediction of the fit of recruit to role. Given the right facts and anchoring, people were quite capable of making better judgments.

 

There are domains where intuition is entirely valid and perhaps the only valid approach to decision making. Intuition seems to work best in complex environments which offer clues to the circumstances. Known as “recognition-primed” decision making, this process seems to involve a hybrid of Systems 1 and 2 in which System 1 proposes an action and System 2 then reviews the proposal for utility. Intuition is a form of recognition of past events that relate to the current event. Decision making in firefighting, chess, combat and reading are all examples of recognition-primed thinking. Presented with new information, System 1 searches our memories for information then applies them to the current circumstance. Intuition seems to work best where there is a possibility for regularity and practice – like chess. Chess is a limited field in terms of pieces and moves, which can be practiced extensively. In contrast, it is hard to predict the economy’s performance in the next 12 months. Economists make these predictions routinely and incorrectly. Economic performance is not “regular” in the same sense. Many things influence economic performance, there are complex feedback loops and non-economic factors (weather, war, political change) can influence the economy. Economic prediction is a domain of low validity, so we become more dependent on our narrative to make a prediction and less receptive to information that does not conform to the narrative.

 

One of the problems of a narrative is that they tend to be specific; they ignore the statistics of the class to which they belong. Kahneman was part of a group writing a textbook and the group was about 18 months into the process, when they stopped to estimate how much longer it would be. The most common estimate was about 2 years. One of the team had supervised other book writing teams and was asked what their results had been. Forty percent of other teams never finished, no team that finished their book took less than 6 years, and closer to 10 years was the norm. This person had estimated 2.5 years despite knowing that no group had ever finished in 2.5 years, or even close to it. Most people would not continue such a high-risk long project, but the team carried on anyway (of course, they thought they were not typical) and the book was completed 8 years later. Guided by their narrative about why this would work, which ignored the role of job changes, divorce, other job obligations and so on, each individual created an internal best-case-ever scenario to justify continuation. Exceptualism is a very dangerous mental trap and it depends on the story that we can create.

 

More broadly, Kahneman calls this the planning fallacy – the belief that the best-case scenario is the most likely scenario and that better (more detailed, conscientious) planning efforts can make it so. Mitigation of the planning fallacy depends on using information from similar cases rather than internal estimates. The prevalent tendency to underweight or ignore distributional information is perhaps the major source of error in forecasting. Planners should therefore make every effort to frame the forecasting problem so as to facilitate utilizing all the distributional information that is available [quoted from Bent Flyvbjerg] ….This might be considered the single most important piece of advice regarding how to increase accuracy in forecasting through improved methods. Using such distributional information from other ventures similar to that being forecasted is called taking an “outside view” and is the cure to the planning fallacy. Why didn’t the team give up the project when they realized that there was a good probability of failure or a long period of effort? One reason was that it was somebody else’s information and experience – it was not their own experience. Second, there was no crisis to deflect them from their plan. Nothing had happened to show that they were on the path to 8 more years of effort. The sunk-cost of the effort gave them momentum to carry on without really questioning whether that was a good idea.

 

This raises the interesting observation that capitalism depends on the combination of the planning and sunk-cost fallacies. Starting a business is not a rational choice. Most new businesses fail and many old ones do too. Most innovations fail, some quickly and some slowly after huge effort and investment. Yet our economy depends on people irrationally deciding that they are exceptions and forging ahead. An unbiased appreciation of uncertainty is a cornerstone of rationality – but that is not what people and organizations want. Extreme uncertainty is paralyzing under dangerous circumstances, and the admission that one is merely guessing is especially unacceptable when the stakes are high. Acting on pretended knowledge is often the preferred solution. Business depends on exaggerated optimism. It seems that the businesspeople making decisions are not especially risk embracing; they seem more risk ignorant. The result is that decisions are made in the context of “bold forecasts and timid decisions”. The effects of high optimism on decision making are, at best, a mixed blessing, but the contribution of optimism to good implementation is certainly positive. The main benefit of optimism is resilience in the face of setbacks.

 

One of the main contributions of Kahneman and Tversky was development of “prospect theory”. Prospect theory describes the kind of decisions people make between choices and illustrates the gap between economic theory of rational choice and the observed behavior of people. For example, you could be given a choice between a chance to flip a coin for $100 ($0 if you are wrong) or to take $45 for sure. Most people choose the $45, but an economically rational person would flip the coin. Most people seem to pay more attention to the worst possible outcome ($0) and thus choose the sure thing rathe than attend to the best outcome ($100). For many people, this is an expression of risk aversion. To get a sense of prospect theory, consider the following case. Today Jack and Jill each have a wealth of 5 million. Yesterday, Jack has 1 million and Jill had 9 million. Are they equally happy?  Utility theory suggests that they should be as they have the same current condition, but most everybody would think that they would have quite different feelings because they compare their current condition to their previous condition. Prospect theory takes this exact scenario and projects it into the future. How will you feel in the future about your decision? The other focus of prospect theory is change. People pay less attention to their state than potential changes: gains and losses. Kahneman and Tversky rapidly learned was to focus on loss. Consider the following two situations. (1) Would you rather have $900 or a 90% chance to win $1000? Most people choose the sure thing rather than take a 10% chance of having nothing. (2) Would you rather lose $900 or a 90% chance to lose $1000? Most people choose the 10% chance that they will lose nothing in comparison to a sure loss of $900. In both cases, people tend to take the option which seems to have the least probability of loss, though formally the expected value of each choice is the same.

 

These examples set the stage for more realistic examples. Would you rather have a 50% chance of losing $100 or a 50% chance of winning $150? The fear of losing $100 is compared to the attraction of gaining $150; this comparison is the essence of loss aversion. To understand this better, imagine that the gain is varied from lower values (like $125) to higher values (like $300) and the range of values at which people accept the bet is examined. To be risk-indifferent, you must accept a bet very close to $100, while the higher the required gain is, the greater the loss aversion. At the same time, when people are faced with a loss and are offered a chance to avoid the loss, they are quite willing to gamble.

 

The focus on loss led to an examination of how people think about loss. The example was provided by two employees who were getting promotions. The company would offer one $10,000 and the other 12 days of vacation a year. Neither employee had a preference, so they flipped a coin to see who would get what. After a year, they were offered a chance to switch. The question was: would they switch? In a variety of similar cases, the answer is clearly no. Once you possess something, even something that you originally viewed as equal to something else, you will not exchange it. Kahneman and Tversky called this the endowment effect. Possession increases something’s value so even exchanges are losses. Gains and losses seem to be in relation to some ‘reference point’, and the endowment effect seems to be one of the ways that we identify the reference point. If we think it is ours, we can fear losing it. This is an important point. While possession enhances the endowment effect, we begin to consider something ours when we expect to possess it. For example, if we knew that there was a 95% chance of winning $1 million tomorrow, we would be very anxious about the 5% chance that we would not win. This anxiety can invoke our loss aversion.

 

Our reaction to gains and losses was condensed into what Kahneman calls the fourfold pattern.

 

                                    

We react to the possibility of winning and losing in a sort of inverse way. We are risk averse when we think we have a high chance of winning or a low chance of losing. We are risk seeking when we have a high chance of losing or a small chance of winning. It is not hard to see how the prospect for personal gain, in the form of performance bonuses, could invoke either risky or conservative behavior under different circumstances.

 

Almost all of the discussion of probability up to this point has been about relatively probable events. It is odd to think about 5% being relatively probable until you begin to think about quite improbable events – one in a million. The author recounts visiting Israel during a period of active suicide bombing. There were 23 bombings over a 2.75 year period that killed 236 people. People stopped riding buses or were nervous during rides. Daily bus ridership was about 1.3 million people, so the objective odds of dying were something like one in 200 million. The author mentions that he did not even want his car near a bus, though he knew there was no objective risk. However, the vividness of the risk was such that it set off an “availability cascade” that made him risk averse. We are poor at dealing with rare events. We greatly overestimate their probability and we are too heavily influenced by them in our decisions. The more vivid an event, the more prominent it is in our recall and the more probable we think that it is. Expressed differently, we confuse impact with probability. We think of air travel as dangerous and car travel as safe, because car accidents are less vivid for most of us than airplane crashes. The inability to compare risks, due to cognitive processes like availability, leads to people declining to vaccinate their children out of fear of side effects. As we have less consciousness of the impact of major diseases (TB, mumps, etc.), these lose their vividness in our minds while the fear of autism remains vivid.

 

Many people are aware of the concept of framing. Framing is liked to System 1 and System 2 thinking because System 1 reacts to an issue in the context of its frame. The frame is the set of circumstances that System 1 uses to identify the right association and decision rules. Reframing is an activity of System 2 and this requires effort. Consequently, we resist reframing to the point where we can’t always reframe a situation. The words used in framing issues are very important. For example, the word “loss” is very powerful; we don’t want anything involved with losses. Tendencies to approach or avoid are evoked by the words, and we expect System 1 to be biased in favor of the sure thing when it is designated as KEEP and against the same option when it is designated to LOSE. The following pair of comparisons illustrates this concept. In version 1, you must choose between A and B.

 

Imagine that the United States is preparing for the outbreak of an unusual Asian disease, which is expected to kill 600 people. Two alternative programs to combat the disease have been proposed. Assume that the exact scientific estimates of the consequences of the programs are as follows:

• If program A is adopted, 200 people will be saved.
• If program B is adopted, there is a one-third probability that 600 people will be saved and a two-thirds probability that no people will be saved.

More people choose option A than B; they want to keep the certainty of 200 lives saved. Version 2 the options are:

• If program A’ is adopted, 400 people will die.
• If program B’ is adopted, there is a one-third probability that nobody will die and a two-thirds probability that 600 people will die.

More people choose option B’ than A’; they want to avoid the certain loss. The two versions offer the identical outcome, yet the common reaction to them is quite different; the fourfold pattern is seen at work. What is most likely is that we did not reframe version 1 spontaneously to examine the question from another perspective; we stayed with the frame presented. By doing so, we are bound by our frame rather than bound by our reality. This fact is well known by salesmen, marketers and propagandists who work hard to frame choices in terms that favor loss aversion or gamble-seeking, depending on how they want you to decide. In other words, it is easy to manipulate our decisions by how a situation is framed, and we must expend some mental energy to reframe the situation and bring more rationality to our decision. As we have seen again and again, an important choice is controlled by an utterly inconsequential feature of the situation. This is embarrassing – it is not how we would wish to make important decisions. Furthermore, it is not how we experience the workings of our mind, but the evidence for these cognitive illusions is undeniable. Count this as a point against the rational agent theory.

 

Two selves

Our experiences influence how System 1 decides and how it views the frames we are presented or create. This suggests that our memories play a big role in our decision making. As elsewhere, we suffer from some problems with memory. One type of problem relates to the duration of an experience compared to its intensity. For example, consider two people’s experience of a painful medical procedure. One might experience a prolonged period of moderate pain, while the other experiences a short bout of intense pain. While it might seem that the first person had a worse time of it because of its duration, patients report that the higher the peak intensity of pain, the worse the overall experience – regardless of duration. Numerous other studies show that we pay much more attention to intensity and neglect duration. We remember peak experiences (good or bad), not how long these experiences last. To talk about this contrast, the author describes two “selves”: a remembering self and an experiencing self. The experiencing self can report about the current experience but has almost no influence on memory. The remembering self apparently pays more attention to peak experiences and ignores the overall experience. Consequently, our memories of experiences can be wrong because they fail to account for the whole experience.

 

To give a sense of the phenomena, an experiment was conducted in which a person placed one hand in very cold water while their other hand recorded their pain level on a computer. At the beginning, people were told that they would have three rounds of testing. In one test, they placed their hand in the water for 60 seconds (none chose to remove their hand early) – this was the short version. In the other test, conducted a few minutes later, the other hand was placed in the cold water for 60 seconds and pain was recorded as before. But at 60 seconds a valve was opened that let warmer water into the water bath raising the temperature slightly over the next 30 seconds. In other words, the second experiment had all of the first test’s bad experience plus more not-quite-as-bad experience. People experienced the two tests in random order. They were then asked which of the two experiments they would do for their third round. About 80% chose the longer trial for their third test, even though it involved more total pain. They remembered the lower final pain intensity in the long test and used that as the overall experience memory.

 

The “two selves” concept can be applied to many situations to understand the difference between the experience of an event and the memory of the same event. Our sense of well-being, the degree to which life is worth living or the joy of being with children can be significantly different depending on when we are asked. What makes this more interesting is that this bias does not apply when we look at this information for another person. For example, as outsiders we might think that paraplegics have a low quality of life because we assume that the duration of their handicap is wearing. The daily lives of paraplegics and able-bodied people can be compared on both a constant basis (the experiencing self) and retrospectively (the remembering self). While paraplegics experience many moments of physical and emotional pain, they report the same retrospective quality of life as the able bodied. It may be a significant error to think that people in end-of-life situations view their own situations the same way that others view them. It might be an error to assess the quality of a work or family experience based on retrospective survey. Even minutes after an experience, the remembering self has already created a record that ignores the bulk of the experience in favor of the peak moments.

 

In summary, we live our lives as if we have two minds in our head. One mind is quick, effortless, mostly right and necessary for our daily functioning. However, it takes short cuts that can lead us to making mistakes. Our other mind is much more rational, but it is slower, requires more energy (and we would rather be lazy) and is easily distracted. The balance between these two minds rests on how much we are willing to slow down. When we are pressured (by ourselves or others) to make a quick decision, we go with our gut and are subject to the biases and shortcuts of System 1. Deliberately slowing down a decision gives our rational mind (System2) a chance to contribute. Understanding how our mind(s) encode our current and past experiences can help you know when going with your gut is absolutely the best option (and there are clearly many parts of our life that we should be thinking fast) and when we should let our “brain” participate (by thinking slow) to keep us from making a mistake.  As a practical matter, people trying to manipulate you, know all of this and tailor their presentations to take advantage of you. You need to know about this too, to know when to slow down and think more clearly.

 

Comments & interpretations

• System 1 might be an important part of associative creativity. The sort of “instant” connection may be how ideas come “out of the blue”.
• It is common advice to focus on a task. If a task is simple or complicated (which is really just a long series of simple tasks), focus reduces the error rate. But when we “focus” on a complex problem, we are more likely to miss important unexpected cues that emerge from the work. There is probably an optimal level of attention for different kinds of work, and it is not always intense.
• The idea that a person has a single energy pool for all activities is not a surprise. What is interesting is the impact of depletion on subsequent decisions. For example, if I spend a lot of time thinking hard about something, I lose interest in eating healthy foods, exercise, and being social. But the same must be true in reverse. If I must expend a lot of energy being social, then I am less likely to have the energy to eat well or exercise properly. We know that the US population has been gaining weight for two decades. Could this be a consequence of our increasingly complex social and work environment which creates greater cognitive loading and results in diminished control?
• Almost everybody hates Power Point presentations. I wonder how much can be traced to our intuitive understanding that Power Point corrupts communication and understanding. On the one hand, the bulleted lists create the illusion of factuality and logic because they are easy to read. On the other hand, we can load up a slide with lots of detail and know that nobody will actually look at it, because it is too much work. Presenters often use this, by showing us the slide and skipping over it. Here is the detail you need, but I won’t take you through it because it is hard. Power Point manipulates us in ways that almost no other medium does. We know this, even when we do not know how or why.
• Is one reason that people resist new ideas and innovation that they lack familiarity? I’d assume that the greater the novelty of an idea, the more difficult it is for people to accept. Perhaps this also links to all the stories of innovators who ultimately succeed by persistence and repetition. Eventually, a novel idea becomes familiar.
• Collecting and analyzing data is hard. In some cases, there is almost no data to collect. Analysis of these sparse data sets must run a high risk of being wrong, yet collecting more data is not possible. This must be a significant part of why many initiatives fail to match predictions. Like the school quality example, the range of outcomes is very wide, but all that we can usefully discuss in the average outcome. This is why innovation is so closely associated with risk taking. For something that is really innovative, there may be little suitable data to analyze.
• I wonder how anchors guide our impressions of success and failure. I saw a “spoof” presentation that talked about “mononumerousis” and the idea that success or failure hinges on meeting or beating a precise target - no matter how valid the target is.
• It is often notes that leaders need to “go with their gut”. The observation that people who are made to feel powerful embrace their intuition presents an interesting situation. As people rise to more powerful positions, they seek to have issues framed in less detailed terms. It is more common to have requests for “just a summary” or “to stay out of the weeds”. The ability of leaders to impose this constriction of information reinforces their confidence in their own ability because it is easier to understand the limited information. This would seem to create a spiral where people become more confident in their analytical abilities with decreasing amounts of information. Applied to the banking crisis, where top bankers said that the system had become too complex to understand, might an alternative analysis be that they had cut themselves off from the information that they needed to understand out of misplaced confidence in their ability to understand.
• There are many ways to think about risk aversion. The idea that availability is a major driver fits with the observation that most company cultures spend quite a bit of time thinking about failed ventures. A big project that does poorly will get everybody talking and pretty soon everybody will know that similar things must be risky and should be avoided.
• Base rates are so easy to ignore. If you engage in attempts to innovate, it is probably discouraging to know that most innovations actually “fail” in economic terms. Everybody pays a lot of attention to Apple over the last few years because it seems to go from success to success. What is remarkably unclear is how many companies did essentially the same things that Apple does and fail anyway. This might be why two important traits for innovators are optimism and a healthy sense of blissful ignorance of certain facts. You must also wonder about survival bias in the telling of Apple success stories.
• There are many complaints about complicated processes and sign-offs today. It is interesting to think that our inability to think about randomness, hindsight bias and a desire to place blame conspire to make decision making difficult and slow. When large companies look at the “instant” decision making of small companies, I wonder how much they are trapped by a mental model of certainty (everything important can be foreseen and controlled unless you are lazy or incompetent) while small companies are very aware of the randomness of life and so decision makers do the best they can and accept that results will vary.
• This book induces a kind of skepticism about rationality as we normally think about it. One particular thing that jumped out at me was the poor performance of experts in predicting events that are human (versus mechanical or natural). It is apparently one thing to hire an electrician to repair an appliance, but another to hire one to cope with a human system. This brings me to consultants, as a class. Are consultants more like pundits with a good story to tell or appliance repairmen? How could we tell them apart before hiring them? Are they really experts (from whom we expect a result) or a form of risk aversion (we shift the blame for the undesired results to them)?
• Advocates of defined interview processes should explain the basis for a fact-based approach to candidate selection to the rest of us better. I wonder how much effort is typically dedicated to interviewer preparation.
• Eisenhower once said something like ‘plans are useless, planning is everything’. Project managers are taught that an hour or planning may save 6 hours of work, and once the plan is agreed there is no reason it can’t work as stated. But the observation is that most plans do not work as expected, and if they look like they are on time/budget, it was not actually performed as the plan predicted. The desire for certainty seems to create a dual need to produce a plan that looks “good” and then believe that it is a valid prediction of the future. This seems to highlight the need for really extensive project after-action studies to get a detailed characterization of the effort, what went well and poorly, in order to compile information for “distributional planning” in the future.
• I have always believed that scientific research is another domain where…optimism is essential to success: I have yet to meet a successful scientist who lacks the ability to exaggerate the importance of what he or she is doing, and I believe that someone who lacks a delusional sense of significance will wilt in the face of repeated experiences of multiple small failures and rare success, the fate of most researchers. Some years ago, I learned that I have both a strong fear of failure and a strong drive to suppress it. I decided that this was a core element of my ability as a scientist. My experimental success rate was tiny for long periods of time, punctuated by bursts of success and capitalizing on that success. A good year was when 5-10% of new experiments worked. The fear of failure motivated thinking about failure closely and the suppression of the fear supported trying again. Sometimes you had to walk away for a while to recover and sometimes you had to give up. I would not have used the word “delusional”, but maybe that is actually the right way to think about it; I have a useful delusion.
• Sprinkled throughout the book are examples that demonstrate how we make independent decisions differently from comparative decisions. In other words, when we decide about something in isolation, we make decisions one way, but if we are considering two possible options – we reach different conclusions. It makes me wonder how much we should decide about things one at a time. Making a decision by comparison to a visible reference would probably be better than making a decision by many unconscious comparisons to varied references.

 

*Text in italics is quoted directly from the book