Welcome to the Making Sense podcast. I'm Kirby Ferguson, producer here at the show. Sam is on vacation this week, so we are reaching into the archive and bringing back one of our favorite episodes. This is Sam's conversation with the Nobel Prize-winning psychologist Daniel Kahneman, who passed in 2024. It's a timeless conversation and actually one of our funnier episodes. If you've heard it before, it rewards a second listen. Sam will be back next week with a new episode. For now, enjoy The Map of Misunderstanding.
Okay. Well, today I'm bringing you the audio from my live event with Danny Kahneman at the Beacon Theater in New York. This was a sold-out event in a very cool old theater. I'd actually never been to the Beacon before, but it has a storied history in music and comedy. Anyway, it was a great pleasure to share the stage with Danny. Daniel Kahneman, as you may know, is an emeritus professor of psychology at Princeton University and also an emeritus professor of public affairs at Princeton's Woodrow Wilson School of Public and International Affairs. He received the Nobel Prize in economics in 2002 for the work he did on decision-making under uncertainty with Amos Tversky.
Unfortunately, Tversky died in 1996, and he was a legendary figure who would have certainly shared the Nobel Prize with Danny had he lived longer. They don't give the Nobel posthumously. In any case, I think it's uncontroversial to say that Danny has been the most influential living psychologist for many years now, but he's perhaps best known in the general public for his book Thinking, Fast and Slow, which summarizes much of the work he did with Tversky.
Michael Lewis also recently wrote a biography of the Kahneman-Tversky collaboration, and that is called The Undoing Project. Anyway, Danny and I covered a lot of ground at the Beacon. We discuss the replication crisis in science, systems one and two, which is to say automatic and unconscious cognitive processes and more conscious and deliberative ones. We talk about the failure of intuition, even expert intuitions, the power of framing, moral illusions, anticipated regret, the asymmetry between threats and opportunities, the utility of worrying, removing obstacles to wanted behaviors, the remembering self versus the experiencing self, improving the quality of gossip, and many other topics.
Anyway, Danny has a fascinating mind, and I think you'll find this a very good introduction to his thinking. Of course, if you want more, his book, Thinking, Fast and Slow, also awaits you if you haven't read it. And now I bring you Daniel Kahneman.
Well, well, thank you all for coming. Really an honor to be here. It's a—Danny, it's a special honor to be here with you. So, thank you for coming.
My pleasure.
It is—it's often said and rarely true that a guest needs no introduction, but in your case, that is—it's virtually true. But we're going to talk about your work throughout, so people will—if they—if for the one person who doesn't know who you are, you will understand at the—at the end of the hour. But I guess by way of introduction, I just want to ask: what is the worst thing about winning the Nobel Prize?
That's a hard question, actually. There weren't many downsides to it.
Okay. Well, nobody wants to hear your problems, Dan. So how would you—how do you think about your body of work? How do you summarize the intellectual problems you have tried to get your hands around?
You know, it's been just a series of problems that occurred that I worked on. There was no big program. When you look back, of course, I mean, you see patterns and you see ideas that have been with you for a long time, but there was really no plan. I was, you know, it's—you follow, you follow things, you follow ideas, you follow things that you take a fancy to. Really, that's a story of my intellectual life. It's just one thing after another.
Judging from the outside, it seems to me that you have told us much of what we now think we know about cognitive bias and cognitive illusion. And really the—the—the picture is—is of human ignorance having a kind of structure. It's not just that we get things wrong. We get things reliably wrong. And because of that, whole groups, markets, societies can get things wrong because the—the errors don't cancel themselves out. I mean, bias becomes systematic, and that obviously has implications that are—that touch more or less everything we care about.
Let's just—I want to track through your work, you know, as presented in—in your now famous and well-read book, Thinking, Fast and Slow. And I just want to try to tease out what should be significant for all of us at—at this moment because, you know, human unreason, unfortunately, becomes more and more relevant, it seems, and we don't get over these problems. And I guess I wanted just to—to—to begin ask you about a problem that's very close to home now: this what is called the—the replication crisis or reproducibility crisis in science, in particular social sciences, and in particular psychology. And for those in the room who are not aware of what has happened and how dire this seems, it seems that when you go back to even some of the most celebrated studies in psychology, their reproducibility is on the order of 50, 60% in the best case. So there was one study done that took 21 papers from Nature and Science, which are the most highly regarded journals, and reproduced only 13 of them. So let's—let's talk about the problem we face in even doing science in the first place.
Well, I mean, you know, the—the key problem and the reason that this happens is that research is expensive, and it's expensive personally, and it's expensive in terms of money. And so you want it to succeed. So when you're a researcher, you know what you want to find, and that creates biases that you're not fully aware of. And I think a lot of this is simply self-delusion. That is, you know, there is a concept that's known as p-hacking, which is people very honestly deluding themselves about what they find. And there are several tricks of the trade that, you know, people know about them. You are going to do an experiment, so instead of having one dependent variable where you predict the outcome, you take two dependent variables, and then if one of them doesn't work, you stay with the one that doesn't work.
You do that and things like that a few times, then it's almost guaranteed that your research will not be replicable. And that happens—it was first discovered in medicine. I mean, it's more important in medicine than it is in psychology, where somebody famously said that most published research in medicine is false, and a fair amount of published psychological research is false too.
Even some of the—the most celebrated results in psychology, like—like priming and the marshmallow test, and—
Well, I mean, the—yeah, I mean, it's not only—it's actually they get celebrated in part because—because they're surprising.
And the rule is, you know, the more surprising the result is, the less likely it is to be true. So that's how celebrated results get to be non-replicable.
Right. Well, and the scariest thing I heard—I don't know how robust this study was—but someone did a study on trying to replicate unpublished studies and found that they replicated better than published studies. Did you hear this?
I don't think that's replicable.
Let's talk about system one and two. These are the structures that give us so much of our—what can be a—a dispiriting picture of human rationality. Summarize for us: what—what—what are these two systems you—you talk about?
I mean, before starting with anything else, there are clearly two ways that ideas come to mind. I mean, so if I say two plus two, then an idea comes to your mind—you haven't asked for it. It's—you're completely passive, basically something happens in your memory. If I ask you to multiply, you know, 24 by 17 or something like that, you have to work to get that idea. So it's that dichotomy between the associative, effortless and the effortful, and that is phenomenologically obvious. You start from there, and how you describe it and whether you choose to describe it in terms of systems, as I did, or in other terms—that's already a theoretical choice. And in my view, theory is less important than the basic observation of, you know, that there are two ways for ideas to come to mind, and—and then you have to describe it in a way that will be useful. And what I mean by that is you have to describe—describe the phenomena in a way that—that will cause—help researchers have good ideas about facts and about experiments to run. And the system one and system two was—it's not my—not my dichotomy, and even not my terminology.
And in fact, it's a terminology that many people object to, but you chose it quite deliberately. What are the liabilities? Because people—in your book you try to guard against various misunderstandings of this picture.
Well, yes. I mean, you know, there is a rule that you're taught fairly early in psychology, which is never to invoke what is called homunculi, which are little people in your head whose behavior explains your behavior or explains the behavior of people. That's a no-no. And system one and system two are really homunculi. So I knew what I was doing when I—when I picked those. But the reason I did was that system one and system two are agents. They have personalities. And it turns out that the mind is very good at forming pictures and images of agents that have intentions and propensities and traits, and—and they're active. And it's just easy to get your mind around that. And that's why I picked that terminology, which many people find sort of objectionable because they're really not agents in the head. It's just a very useful way to think about it, I think.
So there's no analogy to be drawn between the—a psychological, even Freudian picture of the conscious and the unconscious? How do—how do you think about consciousness and everything that precedes it in line?
I mean, you know, it's clearly related in the sense that what—what I call system one activities, the automatic ones, one characteristic they have is that you're completely unconscious of the process that produces them. You just get, you know, you get the results. You get four when you hear two plus two. In system two activities, you're often conscious of the process. You know what you're doing when you're calculating. You know what you're doing when you're searching for something in memory. So clearly consciousness and system two tend to go together. It's not a perfect—you know, and who knows what consciousness is anyway. But they tend to go together, and system one is much more likely to be unconscious and automatic.
Neither system is a perfect guide toward tracking reality. But system one is—is very effective in many cases, otherwise we—it wouldn't have evolved the way it has. But I guess maybe let's start with a picture of where our intuitions are reliable and where they reliably fail. How do—how do you think about the utility of intuition?
I'll say first about system one that our representation of the world—most of what we know about the world—is in system one. We're not aware of it. So that we're going along in life with producing expectations or being surprised or not being surprised by what happens. All of this is automatic. We're not aware of it.
So most of our thinking, system one thinking, most of what goes on in our mind goes on and we're not aware of it. So that's—and intuition is defined as, you know, knowing—or rather thinking that you know something without knowing why—why you know it or without knowing where it comes from. And it's fairly clear, actually—I mean, that's a digression—but there is a guy named Gary Klein, a psychologist who really doesn't like anything that I do.
How does your—how does your system one feel about that?
I like Gary a lot, actually. So, but he believes in intuition and in expert intuition, and he's a great believer in—and he has beautiful data showing beautiful observations of expert intuition. So he and I—I invited him actually to try and figure out our differences because obviously I'm a skeptic. So where is intuition marvelous and where is it flawed? And we worked about—we worked for six years before we came up with something, and we published an article called The Failure to Disagree, because in fact there is a fairly clear boundary about when you can trust your intuitions and when you can't.
And I think that's summarized in three conditions. The first one is the world has to be regular enough. I mean, first of all, intuition is recognition, and that's—Herbert Simon said that: you have an intuition, it's just like recognizing, you know, that it's like a child recognizing what a dog is. It's—it's immediate. Now, in order to recognize patterns in reality, which is what true intuitions are, the world has to be regular enough so that there are regularities to be picked up. Then you have to have enough exposure to those regularities to have a chance to learn them. And third, it turns out that intuition depends critically on the time between when you're making a guess and a judgment and when you get feedback about it. The feedback has to be rapid. And if those three conditions are satisfied, then eventually people develop intuition. So that the chess players—chess is a prime example where all three conditions are satisfied. So after, you know, many hours—I don't know, 10,000 or not—but many hours, a chess player will have intuitions. All the ideas, all the moves that come to his or her mind are going to be strong moves. That's intuition.
Right. So—so the picture is one of intuition. I mean, there are intuitions that are more innate than others, or we're so primed to learn certain things innately that—that no one remembers learning these things, you know, recognizing a human face, say. But much of what you're calling intuition was at one point learned. So intuition is trainable. There are experts in various domains, chess being a very clear one, that develop what we consider to be expert intuitions. And yet much of the story of the blind spots in our rationality is a story of the failure of expert intuition. So to what—where do you see the frontier of trainability here?
I mean, I think that what happens is that when those conditions are not satisfied, people have intuitions too. That is, you know, they have ideas that come to their mind with high confidence and they think they're right. And so the—the main—
Yeah, I mean, you know, we've all met them, and we see them in the mirror. And, you know, that's—so it turns out you can have intuitions for bad reasons, you know. So all it takes is a thought that comes to your mind automatically and with high confidence, and you'll think that it's an intuition and you'll trust it. But the correlation between confidence and accuracy is not high. That's, you know, one of the saddest things about the human condition. You can be very confident in some—in ideas, and the correlation—you shouldn't trust your confidence.
Well, so—but—but that—so that's just, you know, yes, a depressing but fascinating fact that the signature of a high probability that you are correct is what you feel while uttering that sentence. I mean, psychologically, confidence is the marker of your credence in whatever proposition it is you're—you're entertaining, and yet we know they can become totally uncoupled and often are uncoupled. Given what you know or think you know scientifically, how much of that bleeds back into your life and changes your epistemic attitude? You like—yeah, do you not—personally, do—do you hedge your bets? How is Danny Kahneman different given what he has—has understood about science?
Well, this is even more depressing than I thought.
I, you know, in terms of, you know, my intuitions being better than they were? No. And furthermore, I have to confess I'm also very overconfident. So, even that I haven't learned.
Right. So, what are you saying?
It's hard to get rid of those things.
You're just issuing a—a long string of apologies. I mean, how do you get through life this way? Because you—you should know better. If anyone should know better, you should know better.
Yeah. But I don't—I don't really feel guilty about it. So I—
So—so how hopeful are you that we can improve? How hopeful are you that an individual can improve? And how hopeful are you that we can design systems of conversation and incentives that can make some future generation find us more or less unrecognizable in our stupidity and self?
You know, I—I should preface by saying that I'm not an optimist in general, but I'm certainly not an optimist about those—those questions. I—I don't think that, you know, I'm a case study because I've been studying that stuff for more than 50 years, and I don't think that my intuitions have—have really significantly improved. I can catch sometimes, and that's—that's important. I can catch—recognize a situation as one in which I'm likely to be making a mistake, and this is the way that people protect themselves against visual illusions. You can see the illusions, and there's no way you can not see it, but you can recognize that this is likely to be an illusion. So don't trust my eyes. Take out the ruler. There is an equivalent—you know, there is a similar thing goes on with cognitive illusions. Sometimes you know that your intuitions, your—your confident thought, is unlikely to be true. That's quite rare. It doesn't happen a lot. I don't think that I've become, you know, in any significant way smarter because of studying errors of cognition.
Right. Okay. Let me just absorb that for a second. What you must thirst for on some levels is that this understanding of ourselves can be made useful—or more useful than it is—because the consequences are absolutely dire, right? Our decision-making is, one could argue, the most important thing on earth, certainly with respect to human well-being, right? I mean, how we—how we negotiate nuclear test ban treaties, right? I mean, like everything from that on down. This is all human conversation, human intuition, errors of judgment, pretensions of knowledge, and sometimes we get it right. And the delta there is extraordinarily consequential. So if I told you that we, over the course of the next 30 years, made astonishing progress on this front, right? So that we—our generation looks like, you know, bumbling medieval characters compared to what our children or grandchildren begin to see as a new norm. How did we get there?
You don't get there, you know. I mean, that's—you know, it's the same as if you told me, will our perceptual system be very different in 60 years? And I don't think so.
So let's take one of these biases or sources of bias that you have found. I mean, the power of framing, right? We know that if you frame a problem in terms of loss, and—or you frame the same problem in terms of—of gains, you get a very different set of preferences from people because people are so averse to loss. So the knowledge of that fact—let's say you're a surgeon, right, and you're recommending—or—at least, you know, proffering a surgery for a condition to your patients who you have a—you know, you have taken a Hippocratic oath to do no harm, and you know because you read Danny Kahneman's book that if you put the possibility of outcome in terms of mortality rates versus survival rates, you are going to be moving several dials in the—your patient's head one way or the other reliably. Can you conceive of us ever agreeing that there's a right answer there? Like, in terms of what is the ethical duty to frame this correctly? Is there a correct framing, or are we just going to keep rolling the dice?
Well, I mean, you're—this is a lot of questions at once. And the first place, you know, when you're talking about framing, the person who is subject to the framing—I mean, so you had a surgeon framing something for a patient—first of all, the patient is going to be completely unaware of the fact that there is an alternative frame. That's why it works. It works because you see one thing and you accept the formulation as it is given. So that's why framing works. Now, whether there is a true or not true answer—so I should—let me mention the sort of the canonical problem, which actually my late colleague Amos Tversky invented. So in one formulation, you have a choice between—well, there is a disease that's going to cause 600 deaths unless something is done, and you have your choice between saving 400 people or a two-thirds probability of saving 600. Or alternatively, other people get the other framing, that you have a choice between—
Killing 200 people for sure, and—or not allowing them to die—and a one-third probability that 600 people will die. Is there a correct answer? Is there a correct frame? Now, the interesting thing is, people, depending on which frame you present to them, they make very different choices.
But now you confront them with the fact that here you've been inconsistent. And some people will deny it, but you know—you—you can convince them this is really the same problem. You know, if—if you save 400, then 200 will die. And then what happens is they're dumbfounded. That is, there are no intuitions. We have clear intuitions about what happening—about what to do with gains. We have clear intuitions about what—what to do with losses. And when you strip it from that language with which we have intuition, we have no idea what to do. So, you know, what is better when you stop to think about—you have stopped thinking about saving or about—about dying.
Well, actually, I've forgotten if that research was ever done. I forgot what the results were. Has the third condition been compared to the first two? What do people do when you give them both framings and—and dumbfound them? I mean, where—where do the percentages go with respect to—
We—this is not something that, you know, we've done formally, but I can tell you that I'm dumbfounded, and I have absolutely no idea. You know, I have—I have the same intuitions as everybody else. You know, when it's in the gains, I want to save lives, and it's in the losses, I don't want people to die. So—but that's where the intuitions are. When you're talking to me about 600 more people staying alive with a probability of two-thirds, or, you know, when you're talking about numbers of people living, I have absolutely no intuitions about that. So that is quite common in—in ethical problems and in moral problems, that they're frame-dependent, and when you strip the frames away, people are left without a moral intuition.
Well, and this is incredibly consequential in—when you're thinking about human suffering. So your—your—your colleague Paul Slovic has done these brilliant experiments where he's shown that if you ask people to support a charity, you talk about, you know, a famine in Africa, say, and you show them one little girl attached to a very salient and heartbreaking narrative about, you know, how much she's suffering, you get the maximum charitable response. But then you go to another group and you show that same one little girl and tell her story, but you give her a brother, and the response diminishes. And if you go to another group and you give them the little girl and her brother, and then you say, in addition to the suffering of these two gorgeous kids, there are 500,000 suffering children behind them suffering the same famine, then the altruistic response goes to the floor. It's—it's precisely the opposite of what we understand system two should be normative, right? The bigger the problem, the more concerned and—and charitable we should be.
So to take that case, there has to—there's a way to correct for this at the level of tax codes and levels of foreign aid and which problems to target. If we know that we are emotionally gamed by the salient personal story and more or less morally blind to statistics and raw numbers. I mean, there's another piece of work that you did which shows that people are so innumerate with respect to the magnitude of problems that they will more or less pay the same amount whether they're saving 2,000 lives, 20,000 lives, or 200,000 lives.
Yeah. Because basically—and that's a system one characteristic. Basically, you're saving one life. You're thinking—you have an image. You have stories. And this is what system one works on. And this is what emotions are about. They're about stories. They're not about numbers. So it's always about stories. And what happens when you have 500,000? You have lost a story. A story to be vivid has to be about an individual case. And when you dilute it by adding cases, you dilute the emotion. Now, what you're describing in terms of the moral response to this is no longer an emotional response. And this is already, you know, this is—this is cognitive morality. This is not emotional morality. You have disconnected from the emotion. You know that it's better to save 500,000 than—than 5,000, even if you don't feel better about saving 500,000. So this is passing on to system two. This is passing on to the cognitive system the responsibility for action.
And—and you don't think that that handoff can be made in—in a durable way?
I think it has to be made by policymakers, and policymakers, you know, we—we hire some people to think about numbers and to think about—about it in those ways. But if you want to convince people that this needs to be done, you need to convince them by telling them stories about individuals, because numbers just don't catch the imagination of people.
What does the phrase cognitive ease mean in your work?
Well, it means that some ideas come very easily to mind and others come with greater and greater difficulty to the point of—so that's—that's what cogn—it's—it's also called fluency, right? It's, you know, what's—what's easy to think about, and there is a correlation between fluency and pleasantness, apparently, that pleasant things are more fluent. They come more easily. Not always more easily, but yes, they're more fluent. And fluency is pleasant. So there is that interaction between fluency and pleasure, which I hope replicates.
So—but so the picture I get is of—I don't know if you referenced this in your—in your book, I can't remember—but what happens, what we know from, you know, split-brain studies, that for the most part the left linguistic hemisphere confabulates, continually manufacturing discursive stories that ring true to it. And there's—in the case of actual neurological confabulation, there's no reality testing going on. There's nothing. It's just—it's telling a story that is being believed. But it seems to me that most of us are in a similar mode most of the time, that there's a very lazy reality-testing mechanism coming online, and it's just easy to take your own word for it most of the time.
