User Tools

Site Tools


books:gut_feelings

Gut Feelings

The Intelligence of the Unconscious
Gerd Gigerenzer

With this book, I invite you on a journey into a largely unknown land of rationality, populated by people just like us, who are partially ignorant, whose time is limited and whose future is uncertain. This land is not one many scholars write about. They prefer to describe a land where the sun of enlightenment shines down in beams of logic and probability, whereas the land we are visiting is shrouded in a mist of dim uncertainty. In my story, what seem to be “limitations” of the mind can actually be its strengths.

This is a very interesting book about the intelligence of our unconscious intuitive judgments. One of the main messages supports our interpretation of the puzzling evidence from the judgment and decision literature, which is that many of these findings of cognitive bias are misleading due to the contrived nature of the experiments, narrow interpretation of the “rational” result, and also the lack of appreciation that, unlike in economics or academic theory in general, in the real world that human intuition was optimized to work in, predictability is poor, information incomplete, and there is not in any case enough time or brainpower to consider all things.

Unpredictability

We strongly agree with the author's identification of fundamental intractable unpredictability in the world as an extremely important factor in the human condition. As Nassim Taleb observes in The Black Swan, this basic observation is obscured by a strong human tendency to exaggerate our ability to predict, a bias which seems completely unaffected by the repeated failure of prediction.

Simple Rules are Good

A unique emphasis of this book is the idea that simple decision rules work as well or even better than theoretically optimal statistical decision procedures. This is a general point about what decision rules are good in the real world and not specifically an observation about intuition, and in fact he goes on to argue at some length that we should consciously adopt simple rules for medical decision-making because they are easy to understand, apply, and explain, they work very well, and (unlike intuition) it is possible to standardize best practive. The virtue of simple rules does however tie in with his argument defending the rationality of intuition because he believes that simple rules underlie intuitive decision-making.

For example, in the real world, investing equal amounts in several different stocks works better than the Nobel-prize-winning theoretically optimal portfolio theory. Medical researchers also found this true for the decision of which possible heart-attack patients to send to intensive care and which to put in a regular bed for observation. Fast and frugal decision trees

The author's explanation of why simple rules that only look at a couple factors can outperform statistical rules that consider all factors is the consider-all-things rules end up incorporating spurious information that doesn't have predictive value. Though he doesn't use the term, this basically amounts to a claim that simple decision rules are less vulnerable to Overfitting that consider-all-things rules. This does raise the question of how true his implication is that simple rules outperform consider-all-things rules. The concept of overfitting implies that judgment is involved in creating a good statistical decision rule, and that if you compare to overfitted rules you end up looking good.

Another way to look at the success of simple rules is suggested by The Black Swan: that statistical rules often assume Gaussian distributions (no outliers) and thus are overly confident that the future will resemble the past.

Intuition is Simple Rules?

The author argues (unconvincingly we feel) that intuition is generally based on these sorts of “fast and frugal” rules. Certainly in some cases (such as how baseball players catch fly balls) we can infer simple rules that underlie their behavior, and these simple rules don't at all resemble the complex physics that predict the ball's trajectory. What he seems to completely miss in the example of motor skills there is an entirely different explanation for why a simple rule is effective, and that is feedback. As he notes it is really not possible to predict exactly where a ball will land at the time it is hit because of wind changes, turbulence, etc. So even if the human brain could do so, it would be futile to solve complex differential equations on the fly. Feedback is necessary, and once you have feedback, all sorts of approximations are tolerable.

Choose the Best

He argues that the simple rule “choose the best” underlies most intuitive decision making. This is a good strategy, because, for example, handball players intuitively choose the best play first, and if you stop and give them more time to think, they actually come up with worse results because the other solutions that they generate are not as good as the first one, and once the bad option is on the table, the players sometimes choose it because their conscious theories of play are actually rather ill-informed.

But “choose the best” entirely begs the question of the nature of intuitive decision making. What this really says is that all the meat of simple decision making like handball play is not rule-based. All the action is in determining what is best, and often this will involve the exact sort of weighing of many considerations that the author argues against. Another body of research strongly supports the idea that our intuitive reactions are strongly based on many subtle considerations that we are not consciously aware of, and that is priming.

Less is More?

One of the authors unifying themes is “less is more”. He uses this to bundle a bunch of fascinating observations, but we feel that he generally is describing the consequences of being a largely unconscious being. He is describing the output from the unconscious into the conscious, and not the actual workings of the unconscious. Although he clearly has a sophisticated understanding of mind, he is still falling for the user illusion. Yes it helps us that we forget unimportant things and recall important ones. This data reduction is crucial to enable effective conscious processing because the bandwidth of consciousness is so limited (see Level Map and The User Illusion.) Similarly, we think that his preference for a rule-based unconscious may because he's confusing the interpreter's stories with the actual workings of the unconscious.

Social Environment

Any attempt to investigate human decision-making and intuition that ignores the dominance of social considerations in human lives is doomed to confused conclusions—man was a social animal long before he evolved into a conscious animal. The author gives a simple example of this sort of error from rat behavior experiments. A hungry rat is repeatedly put at the base of a T maze that has food rewards at the ends of the top. Though the food rewards are not always the same, there is generally 10 times as much food on the left arm as the right. Rats do notice the difference, and favor the left arm, choosing it 9 times our of 10. The experimenters were puzzled because “rationally” they should always choose the left arm—silly rats! But in the real rat race, a lone rat is a rarity, and in a pack of rats it is optimal for 10% of the rats to turn right.

The author makes some excellent observations of human behavior that largely duplicate the theory of Cultural Evolution in Not by Genes Alone, but he neglects to put these social behaviors in an evolutionary context. In particular, he notices the workings of Conformity Bias and Prestige Bias and how these lead to successful behavior in a social context.

The Conjunction Fallacy

Linda is thirty-one years old, single, outspoken, and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice and also participated in antinuclear demonstrations.

Which of these two alternatives is more probable?

  • Linda is a bank teller
  • Linda is a bank teller and active in the feminist movement

If you say she's gotta be a feminist, you've just fallen for the Conjunction Fallacy. Or have you?

The author argues that though this is indeed a violation of the way that you are taught to parse logical propositions out of natural language, that this result also make complete sense in the context of normal human communication where people follow the “maxim of relevance.” In conversation it's a fundamental principle that everything the other person says is relevant to what they are trying to say. We routinely ignore minor logical boo-boos like saying the opposite of what you mean.

He shows that if you explicitly state the question in a frequentist interpretation (if there's a hundred people like Linda, how many are…), then people give the “right” answer. Clearly some proposed cognitive biases are semantic artifacts of the poor mapping of logic onto natural language.

The Recognition Heuristic

This book has a wonderful example of how humans can socially construct accurate knowledge (but fails to adequately emphasize that this is a cultural process.) Suppose that you ask German students which of a bunch of pairs of German cities is larger, then also ask American students the same questions. The Germans will do a lot better, right? Not! In fact, the Americans do as well or better by using the recognition heuristic—Americans haven't heard of most small German cities, but have heard of many large ones, so they correctly guess that if they've heard of one city and not another, that the familiar one is larger. The Germans can't do this because they suffer from the disadvantage of having heard of all the cities, yet still don't know the actual populations. There was a similar result for English soccer teams with English or Turkish subjects.

These are true examples of when “less is more”; for use of the recognition heuristic, it is optimal to have heard of only 1/2 of the items. The author discusses two other cases where asking the man on the street which items he recognized predicted more accurately than the experts did: stock picking and winning Wimbledon.

The expert stock pickers did particularly poorly. Experts asked to pick which of a pair of stocks would do better only predicted the correct stock 40% of the time. Though non-experts only performed at the chance level (50% accuracy), this was still better than the experts. The stock market is almost completely unpredictable; clearly the experts mostly aren't earning their paychecks by making accurate predictions. The author says that the reason for worse-than-chance performance is that for marketing reasons the experts are diversifying their opinions (away from reality.) We infer that investors shop around for an expert with a story they like (and refuse to learn from past prediction failures), so any plausible story can be the basis of a successful investment adviser business.

A less trivial example of less-is-more comes from the Wimbledon tennis matches. The experts performed well above chance, but tennis amateurs who had heard of 1/2 of the players still outperformed rankings by three different groups of experts. Random people who had heard of well less than 1/2 of the players didn't do quite as well as the experts, but still much better than chance.

Why does this work? It's socially constructed knowledge. Ripples of information are constantly washing over us, and we continuously assess what is important enough to remember. The simple decision of remembering or forgetting that someone is a tennis player encodes a considerable amount of knowledge about that player's skill. An expert's detailed knowledge, sophisticated theories (and the need to differentiate themselves in the sports pundit market) impair their ability to make the obvious decision.

A fascinating paradox of the recognition heuristic is that if you are on a team of three people playing a trivia game, and you are asked whether Detroit or Milwaukee is bigger, and none of you actually know for sure, but two of you have heard of both cities and one has only heard of Detroit, then you should guess Detroit. With the recognition heuristic, ignorance actually adds information. Fascinatingly, in experiments groups of people actually did defer to the more ignorant person, though it was rare that anyone articulated the argument that since they hadn't heard of it Milwaukee must be smaller. It seems that the group responded to the quick confident response of the ignorant person. By intuitively judging how confident other people are we can assess the reliability of other people's statements. Emoting your degree of confidence is an equally important social skill.

Futility of Introspection

It is an interesting point about the human condition that baseball players have no idea how they catch a fly ball—why to jog over rather than run flat out and then wait for the ball. The players are using simple rules that they have learned somehow, but they have no awareness of this. Though there may be some cases where it is beneficial to not understand what we are doing (repression and Denial), we believe that the inaccessibility of intuitive processing is primarily due to Representational Opacity. In this case, though we can understand the motor skill as being based on a simple rule, the actual architecture of the brain is not a rule-based system, so we can't discover the rule by introspection, only by Reverse engineering (see Intentional Design.) And intellectually understanding the skill is often little help in actually mastering it, since the opacity works both ways. You can't pull the rule out of the neural net, and you can't push it in either (except by training.)

You could leave a comment if you were logged in.
books/gut_feelings.txt · Last modified: 2012/07/06 23:33 by ram