| A review of Thinking, Fast and Slow by Daniel Kahneman JDN 2456059 EDT 15:43. Daniel Kahnenman is (along with his friend Amos Tversky) the founder of behavioral economics as we know it. He was trained in psychology but won the Nobel Prize in Economics (and one of my goals in life is essentially to follow in his footsteps). His first book, Thinking, Fast and Slow, summarizes three decades of his work, from the availability heuristic to prospect theory. If you are looking for an introduction to the state of the art in behavioral economics, look no further. And even if like me you already have some familiarity with the field, some of the details of Kahneman's many experiments will surprise you. The style is very accessible, even though many of the ideas are cutting-edge science. The two appendices are reprints of published scientific papers, and they too are remarkably readable, a far cry from the usual obscurantism of scientific jargon. The book is a tad long, but it needed to be in order to include everything that needed to be included; also it reads quickly and is always compelling, so the length isn't really something you'll notice as you read it. Kahneman staunchly resists the idea that his work shows people are irrational, and I'm not quite sure why. He writes (p.411): “I often cringe when my work with Amos is credited with demonstrating that human choices are irrational, when in fact our research only showed that Humans are not well described by the rational-agent model.” (The capitalized “Humans” is meant to distinguish between “Humans” Homo sapiens and “Econs” Homo oeconomicus. I find it a bit grating honestly.) But why not come out and admit it? Behavioral economics shows that human beings often behave irrationally. I mean, come on, 40% of Americans are Young-Earth Creationists; clearly our species is not the paragon of reason we'd like to think it is. And some of the experiments show that people are remarkably irrational. As in, we not only get hard problems wrong—we get easy problems wrong. We make stupid, boneheaded mistakes as a matter of course. Truly the remarkable thing is that our civilization has not destroyed itself already. For example, consider framing effects: People will often answer a question—even a moral question—completely differently depending on the wording. When an intervention is worded as “saving lives” people will be risk-averse (preferring small, certain gains to large, uncertain gains); but when the exact same question is worded as “letting people die” people will be risk-seeking (preferring a chance of preventing large losses to a certainty of preventing a small loss). They will “save 200” rather than “take a 1/3 chance of saving 600”; but they will “take a 1/3 chance that no one will die out of 600” rather than be certain “only 400 will die”. They will respond differently to the exact same situation, involving human lives at stake, based on the wording of the question. If that's not irrationality, I don't know what is. Nor is this the exception; it is plainly the rule. People make decisions—even serious, life-altering decisions—based on all the wrong criteria. They take chances they shouldn't (like playing the lottery) and refuse to take ones they should (like investing in stock). They base their hiring decisions on how people look instead of how qualified they are. They incur suffering to make more money, even though the data shows that money won't make them any happier. They are insensitive to scale: They'll give scarcely more money to save 100,000 people than they would to save 1,000 people. If you present two questions separately, they'll answer one way; but when you present both together, they'll answer a different way. Indeed, the scariest part of all this research is that it applies almost as well to professional psychologists and economists as it does to ordinary folk. Education has a slight affect on improving rationality—but not a large one. I can find many of these biases in my own decisions (particularly loss aversion and the sunk cost fallacy), and if you are honest you can probably find many of them in your own. We are a kludge of the Blind Idiot God; our brains are as badly designed as our bodies. What this all means for economics is that rational-agent models, even rational-expectations models, are just blatantly, obviously false. What's amazing is that these models ever get close to describing real-world economies, because they plainly don't describe real-world humans at all. (On the other hand, rational-agent models did fail miserably to predict the 2008 crash, so maybe they're just as ineffective as we'd expect of such a completely wrong model.) It is important to recognize that humans are not completely irrational; we could not have cured polio or landed on the Moon—indeed, could not have survived the Ice Age or invented fire—if we were completely irrational. We are not completely random in our behavior; we are capable of intelligent thought; but at the same time our cognition is lazy, inefficient, and often biased. We make the right decision in probably 80% or even 90% of cases—but woe betide that other 10 to 20 percent. We make our decisions most accurately in environments similar to those of our ancestors: So for instance we're very good at detecting cheating and maintaining social order in small groups. We are however positively terrible at managing globalized economies of billions of people. Kahneman explains many of the results of behavioral economics using a framework called “System 1” and “System 2”. I did not like this terminology at first, because giving numbers to things instead of explanatory names always strikes me as a bad idea. I particularly dislike the terms “Type I error” and “Type II error”; knowing which is which is pure memorization and has no logic whatsoever. (Compare “false positive” and “false negative”, which mean the same thing and are no harder to say.) But in fact “System 1” and “System 2” are somewhat better than this, because they really are actually in order of priority: Your brain first uses System 1, and only if System 1 cannot return an unambiguous answer is System 2 even called into action. On the vast majority of computational problems you face every day, you use only System 1 and never even realize that System 2 was an option. (In fact, here's a fascinating application: In my experience autistic people appear to have an undeveloped System 1 for interpersonal interaction. We lack many of the intuitions other people have about social interaction, and as a result have to consciously learn the stupid rules normal people follow using System 2. This hypothesis has a lot more explanatory power than the common idea that autistics “lack empathy”—why then do we almost to a man love animals?) As a result, I can forgive “System 1” and “System 2” even though I would have used the more descriptive terms “intuition” and “cogitation”. After you've used the concept several times, there is no difficulty remembering which is 1 and which is 2, because System 1 is always used first at highest priority. (This is profoundly unlike “Type I error”, which I've noticed still slips up even professional researchers on occasion.) Kahneman also rightly points out that it is not always the case that System 2 is more rational. This is typically the case, but there are important exceptions—particularly in cases where there are strict time constraints on decisions. A firefighter in a burning building, a soldier in a warzone, even a chess master on the clock, cannot afford to engage the full power of System 2 and “think slow”; they must think fast, using their honed intuition, simply in order to get a viable answer in an acceptable amount of time. Indeed, one terrifying fact about our lives is that we are almost always under time pressure, to greater or lesser extent. We cannot afford to spend seven hours deciding which clothes to wear, or ten years deciding which college to attend. We have more time for the latter than the former—but we are limited in both cases. (Indeed, in support of my theory that autistics have an underdeveloped System 1, consider that some autistics are avolitional—they spend far too long trying to decide how to act even in very simple low-stakes situations, and as a result often fail to act at all. The “lack of empathy” theory simply could not explain this behavior, but it makes perfect sense if System 1 is underdeveloped.) How then do we balance the use of System 1 and System 2? Kahneman offers a few hints as to how this might be done, but this is the most disappointing part of the book. After learning all the things that can go wrong with human decisions, we are offered hardly any insights into improving our decision-making. The best advice Kahneman offers us is to take an “outside view” and try to think of ourselves in third-person terms; this may sometimes be useful, but it's hardly a good general strategy. My proposal is different; it is to work to use System 2 in order to train and improve our System 1. Learn about the biases and heuristics, and then train yourself to catch them when you see them. Play around with numbers and hypothetical scenarios until it finally makes sense—intuitively, on a System 1 level—that you should act the right way. I for one have found that as soon as I reframe problems of “lives saved” as “lives lost” and vice versa, my desire to choose differently completely disappears. Shut up and multiply; use the expectation value. 1/3 of 600 is exactly the same as 200, so literally there is no way to decide. Flip a coin, or use the cheapest one. But 1/3 of 800 is quite a bit more than 200, so if you are risk-averse in this scenario you should stop and think about the 66 extra people (on average) you just let die. (Note that the marginal utility of a human life is pretty much constant.) In fact, you should probably not be risk averse even when it comes to money, unless the amounts involved are huge. Given the choice between a guaranteed $1 million and a 25% chance of $5 million, you should indeed take the $1 million, because that's a truly life-altering level of wealth (hence diminishing marginal utility really does apply, and you'll almost certainly never get a similar chance again). But given the choice between a guaranteed $100 and a 25% chance of $500, you should definitely take the latter, because if you're given similar options ten times you'll make an extra $250. (Given it a hundred times you'll make an extra $2500.) If you don't take the risk, you should ask yourself why you're willing to throw away that $25. (This is also what you should say to yourself if you think that your feelings about risk should guide your decisions. Why are your feelings worth losing that money?) It is far from easy to rewire your automatic responses, but it can be done. Indeed, I think a major goal of cognitive therapy should be to do exactly this: Apply the plasticity of the brain to retrain our automatic responses to be more rational. This task is difficult and time-consuming, but it works: If you want to make yourself healthier, happier, and generally better at life, cognitive therapy is the way to do it. It is proven effective; indeed, it's one of the few things that is. Of course, if you're not rational, it may be difficult to motivate yourself to go through the effort to become more rational... this is a general problem, and I haven't yet come up with a compelling solution. (The problem of persuading Creationists is really a special case.) Indeed, one of the first steps to rewiring your cognitive algorithms may be to read Thinking, Fast and Slow. There you will learn about a great many defects in the human mind that have probably passed your notice. I came out of reading it wondering most of all how we ever manage to function at all, though I think I have a partial answer to that question: many of these heuristics are at least reasonably good most of the time. Availability has something to do with probability, and representativeness tells you something about category membership; they're better than guessing randomly. They're just far from perfect, and it's easy to find cases where they fail. But one thing I don't quite understand, and maybe the answer is evolutionary: Why aren't we good at multiplying four-digit numbers? It's a computationally trivial task—it takes about 32 bits of memory and about 16 bit-shift operations. It's something that a high-end computer does hundreds of millions of times per second. So why are our brains, somewhere near 100 terabytes of memory, somewhere near 100 gigahertz of processing power... completely awful at this task? Why does System 2 even need to be called in for something millions of times simpler than the System 1 task of recognizing a face? What does this difference tell us about our evolution? Is it something that may be curable through training or medication—or if necessary, through genetic engineering or cybernetics? Can we make human beings more rational? Yes, I think we can. It is difficult, and we are only beginning to understand how. Yet, already our generation has a higher IQ than the previous (and the next may be higher still). Already we have come so very, very far from our origins as savannah hunter-gatherers; the computer on which you read this attests to that. It can be done—and it must be done, lest our technology outpace our reason. The first step is realizing you have a problem. Most economists don't; they still think that rationality is a good model of human behavior. Kahneman knows better, and will tell us how; but we must get them to listen. |