Making Sense with Sam Harris - #268 — The Limits of Self-Knowledge

Episode Date: November 24, 2021

Sam Harris speaks with Stephen Fleming about the neuroscience of self-awareness. They discuss the nature and limits of metacognition, the relationship between self-knowledge and intelligence, error monitoring, theory of mind, mirror neurons, deception and self-deception, false confidence, probabilistic reasoning, where metacognition fails, cognitive decline, calibrating one’s confidence, and other topics. If the Making Sense podcast logo in your player is BLACK, you can SUBSCRIBE to gain access to all full-length episodes at samharris.org/subscribe.   Learning how to train your mind is the single greatest investment you can make in life. That’s why Sam Harris created the Waking Up app. From rational mindfulness practice to lessons on some of life’s most important topics, join Sam as he demystifies the practice of meditation and explores the theory behind it.

Thank you. Today I'm speaking with Stephen Fleming. Stephen is a professor of cognitive neuroscience at University College London and is the author of the recent book, Know Thyself, The Science of Self-Awareness. And self-awareness is the topic of today's conversation. We talk about the relevant neuroscience, the relationship between self-knowledge and intelligence, the evolution of metacognition, error monitoring, theory of mind, mirror neurons, deception and self-deception, false confidence, probabilistic reasoning, where metacognition fails, cognitive decline, those places where self-knowledge might be counterproductive, and other topics.

Anyway, I found it quite interesting, and I hope you do as well. And now I bring you Stephen Fleming. I bring you, Stephen Fleming. I am here with Stephen Fleming. Steve, thanks for joining me. Thanks very much, Sam. It's an honor to be here. So you've written a very interesting book on perhaps the most interesting topic. The topic is self-knowledge, self-awareness. The book is Know Thyself, the Science of Self-knowledge self-awareness the book is know thyself the science of self-awareness but i'm really eager to talk about the whole sweep of this but before we jump in perhaps you can summarize your background academically and intellectually sure yeah so i'm currently a professor of cognitive neuroscience at University College London.

And I guess, I mean, I've always been interested in the sciences. I left high school without really knowing what I wanted to do, wanted to be a musician. So I didn't apply to university like all my friends were doing. And instead, I took a year off back then and worked in an office job. And it was while I was commuting that I started reading popular science books on cognitive science. That's partly why I was also so interested in writing one myself when I got the opportunity and just found it absolutely fascinating. I had like no idea that there was a science of the mind out there. We didn't get exposed to that at school, at high school. So I then became fixated on doing experimental psychology. I went to Oxford and I was lucky there to have as a tutor, a guy called Paul

Lazzopardi who works on blindsight, this bizarre neurological condition of consciousness. And it was Paul who convinced me that there was a real rigorous science out there of consciousness and that it was possible to do you know good neuroscience on this and I then went on to University College London to do a PhD in neuroscience with and that was co-supervised with the psychologist Chris Frith and the neuroscientist Ray Dolan and both great people I haven't met either but I've obviously read their papers yeah no it was it was a fascinating time and in ray's lab he was focused on studying decision making using reinforcement learning models and in my phd i mostly focused on using brain imaging to study decision making but on the side i was continuing to kind of have this off on love affair with

consciousness, which has kind of continued with me now. And I guess towards the end of my PhD, I realized we could start applying some of the tools of decision-making research to also study how we make second order decisions. So how we think about and reflect on how we're performing on various tasks. And that's what psychologists refer to as metacognition or thinking about thinking. But at the time, there was a long tradition in psychology of studying this topic, but very few people were working on the neuroscience of metacognition. And I had the opportunity in a sense to get in on the ground floor of that and we ran a couple of early brain imaging studies looking at the relationship between prefrontal function and

metacognition. I then went off to New York to do a postdoc at NYU to learn how to build computational models of metacognition and then in 2015 moved back to London and UCL, where I now lead my own research group studying metacognition and consciousness. Nice, nice. Well, we should dispense with one possible source of confusion at the outset, because I'm not sure how familiar you are with my work, especially on the topic of mindfulness, meditation, the nature of the self. especially on the topic of mindfulness, meditation, the nature of the self.

And so I'm someone who's given to say at fairly regular intervals that the self is an illusion, or at best a construct. At bottom, it's not what it seems. But that's a very specific use of the word self. And when we talk about self-awareness, I think we're talking about something that is far more capacious than the sense of subject-object perception, which is really the linchpin of the self that I would argue that mindfulness ultimately reveals to be illusory. So we're talking about the whole person much more often than we're talking about the sense that there's a subject in the head independent of experience. So when we're talking about self-awareness, this is not in violation of anything I have said about the status of the self as subject in other contexts.

And I'm happy to talk about the self with you as well, but I just wanted to try to clarify that for people because there's going to be something I can hear in the heads of many listeners. If the self is an illusion, what could we possibly mean by self-awareness? Well, self-awareness extends to everything else we can reflect on and be aware of in a kind of second order way that relates to our experience, you know, our performance errors, the thing that we just experienced a moment ago, you know, lapses in memory, or, I mean, let's just dive into the topic. How do you describe metacognition at this point? Yeah, no, I think that's a really useful background to have in place, metacognition at this point. Yeah, no, I think that's a really useful background to have in place. Because I am talking about something distinct here to the philosophical notion of

self, which is a complex object. And here I'm talking about something more practical, something more functional, which is this capacity to be aware of our traits, our skills, our personalities, our behaviors, and in some sense, see ourselves like others see us. And we can study this in various ways. We can look at, in very simple tasks, how people realize they've made errors or how they're able to estimate their confidence in their skills and abilities and so on. And it's something that we often, I think, just take for granted. But the reason I find it so fascinating is because when you think about it for a moment, it is a kind of bizarre and wonderful feature of the human mind that

we can, in some sense, think about how our own minds are working. And this has very practical consequences. So the reason we write a shopping list when we go shopping is because in some sense, we realize that our memory is not going to be good enough to hold all those items in mind. And similarly, when we start to realize our sight is failing, for instance, it's not because we think the outside world has become blurry. It's because we realize that there's something in our perceptual systems that needs fixing with new glasses and so on. So it's this kind of practical reflective thought that's not always obvious from the outside, but it's something that we can study with the tools of psychological science.

Yeah. And one thing is increasingly clear is that other people and now even algorithms, can know what we're like better than we can, certainly on specific topics. I remember a friend once told me a story from a board meeting where he was engaged in a very stressful conversation with the group, And someone in the meeting commented on how emotional he was getting. And it just seemed like they might want to take a break. And he denied being overly emotional. And someone around the table suggested that he bring his attention to the sensations at his upper lip. And the moment he did that, he burst into tears. Wow.

I mean, apparently his lip had been quivering as though he was about to burst into tears, and it was noticeable to those in the room. And, you know, you can imagine just how much can be known about any one of us now based on our Google search history, say, or anything we do with our attention online. And when you look at the database of knowledge that is the profile of each of us that is accruing somewhere in the cloud, and what might be gleaned from that when you compare it to everything else, every other profile of every

other person, just the statistical knowledge there and the capacity to predict the next thing we'll find captivating. It is exactly what you said, to take the view of oneself that another person could have opens the door to sometimes mort sometimes mortifying you know at minimum interesting facts that are not necessarily visible or salient when one's simply living one's life and having one's experience yeah absolutely it's interesting the example you mentioned of the person in the boardroom because i i feel like I have through studying metacognition I've become more attuned in my own life to how I might have this fading out of self-awareness at certain moments and it's something that my wife has said to me on occasion when things are

stressful with grant applications or whatever that I just become you know a horrible person to live with on a for a few days at a time and I I used to deny this completely. I was like, I don't feel like anything's changing my behavior. And I now come to realize that how could I have possibly known at that time? I mean, there's a whole interesting story there about stress and how it is detrimental for metacognition itself. So you have this kind of paradoxical situation where the times when you might need to be aware of how your behavior is causing problems for others, those are the times when metacognition and self-awareness might actually be most impaired. But I have definitely, I think, become a bit more willing to accept in my own life that those fade outs of self-awareness can

happen and they do happen probably more often than I'd like to admit and I then have this stronger tendency to trust what say my wife is saying about my behavior and to try and correct it accordingly yeah yeah so let's um build up this picture of metacognition. I mean, the simplest or most common definition I think one encounters is the phrase, knowing that you know, right? There's the knowing of things, there's the cognition piece. that you have the knowledge, and this extends to knowledge in all of its forms. Semantic knowledge, if you ask me, could you name more than four states in the United States, I could say yes to that. I could be sure about my knowledge there without actually going through the exercise of listing any states. So I have this more abstract understanding that my knowledge bank contains at least four state names, and so it is with so much of what we know. And of course, we can be wrong about that. We can actually think we could produce specific

concepts or memories and, when asked, actually fail. But generally speaking, there's a representation of what is in our storehouse of knowledge that doesn't require us to actually go into the storehouse in order to cash it out in that moment. And so it is with even procedural learning or motor memory. So do you know how to ride a bike? Could you raise your hands over your head? We'd be surprised to have our confidence about that disconfirmed if we tried. But how would you build up the layers of what we're calling metacognition here? What is it beyond this representation to oneself that one

knows certain things? Yeah, I think that's a very nice way of thinking about it, this notion that there are representations that go beyond knowledge. And one analogy that I sometimes use, it's not a perfect one because it's not how things actually work, But you can think of metacognition as, in some sense, being like the index of a book. And the index usually points you to the right page in the book. But if the index maker has got things wrong and the book's self-knowledge has, in some sense, failed, then sometimes there will be an index entry that does not correspond to the actual text in the book. And I think we can start to build up a picture of how metacognition works by thinking of the brain as effectively a hierarchical system and that it does not only encode information in memory, it does not only perceive and represent things at a first order level, but it also has what we think of as higher order representations. And we think

parts of association cortex like prefrontal and parietal cortex are important for this, that it builds representations at a more abstract level of how the system is working. And I think that's probably the best way of conceptualizing metacognition at a cognitive systems level that we have at the moment. And then we can obviously take this in many different directions in terms of specific topics within that broader umbrella term of metacognition. And how does it interact, if it interacts at all, with the variable of intelligence? So I think there is an initial intuition that we have that intelligence is in some sense allied with having good awareness of what we know and don't know, but as ever, it turns on our definition

of intelligence. And empirically, what we've found, perhaps surprisingly in many of our studies, is that when we measure metacognition in the lab, and maybe it's useful to say a few words about how we actually do that. So typically the way we can quantify your metacognition and put a number on it in a particular task is by asking you to assess your performance on a number of trials of the task. So we might give you a memory task and after every decision about whether this object was on the list that you were asked to remember or not, we'd ask you how confident you were about that choice. Or we might give you a task involving perceptual judgments and then ask you how confident you are about each choice. And the key thing we're interested in there is not only your performance on the memory and the perception task, but also how

your confidence tracks your performance. So intuitively, if I have high confidence when I'm right and lower confidence when I'm wrong, that's what we call having good metacognitive sensitivity or metacognitive ability. And what we found in those studies, now we've done studies of thousands of people, is that performance on classical IQ tests is not a great predictor of metacognitive ability. IQ test is not a great predictor of metacognitive ability. And this lines up with some other work in using other measures of metacognition, like whether we tend to be fooled by initially intuitive answers without reflecting on them. So these are things like the cognitive reflection test that taps into more system two than system one thinking. And again, there in the literature on that kind of test, it does seem to be independent of classical IQ

measures. And I think one way of thinking about this on a very broad brush basis is that the kind of neural and cognitive resources that we bring to the table to solve reasoning problems, which is effectively what an IQ test is tapping into, is that those are distinct to, or somewhat distinct, to the kinds of neural and cognitive resources involved in reflecting on our performance in those tasks, including potentially even in an IQ test. So you, in theory, and we have done a little bit of this, you can measure someone's metacognition about their performance on a test of intelligence. So in a sense, both on a theoretical basis, but also on an empirical basis, we think metacognition and IQ come apart in interesting ways.

Yeah, you can see that metacognition and performance have to break apart because you would have perfect metacognition if you were confident that you had utterly failed to perform, if in fact you had utterly failed to perform. You could just go through life failing again and again, and as long as you're aware that it's just one failure after another, well, then your metacognition score is perfect. That's right. That's exactly right. And I say, I kind of make the throwaway line in the book that metacognition is often most useful when we're doing stupid things, because that's when we need to be aware of making errors. So no, that's absolutely right. What picture do we have based on evolutionary psychology of metacognition? How do we think this might have evolved? And what are the benefits of being able to represent to oneself

the likelihood that one has made an error? I mean, that's obviously only one slice of metacognition, but this second order reflection what how does this fit in in the context of evolutionary psychology so one starting point for getting at that question is to look at how and whether we share metacognitive capacities with other species and there has been an interesting line of work for many years in comparative psychology, looking at tests of confidence and uncertainty and error monitoring in animals. And the general picture there is that in many species, you can have pretty sophisticated tracking of confidence, tracking of errors and so on. So there's been some lovely work in dolphins and monkeys and rats

showing that they pass confidence tasks similar to the ones that we use with humans. But that, I think, is a type of metacognition that occupies a different space to explicit self-awareness in humans. And the reason that we think that's the case is because when we look at child development in humans, that kind of implicit metacognition, the ability to track confidence and monitor errors, that seems to be in place relatively early in life. lab in Paris showing that babies as young as 12 months, I think even younger than that in some of their studies, show signatures of error monitoring both in EEG activity and also in their persistence for searching for particular objects. So when you use their persistence of searching for a toy, for instance, as a marker of confidence, then you get all the same metrics of metacognitive sensitivity that you can get out of the adult data. Now, that seems to be that kind of lower

level ability to self-monitor seems to be in place quite early in life in humans. But when you actually, when kids become verbal and you then ask them about their confidence and about whether they know something or don't know something, then as I'm discovering at the moment with my two and a half year old, their metacognition is terrible. They think they know things they don't know. They fail to realize they need to ask you about something and so on. So it's not until the age of around three or four that children start to gain this explicit self-awareness of what they know and don't know.

And we think that in studies in adult humans, that kind of more explicit level of self-awareness is related more to theory of mind or the ability to think about other people as well as to think about ourselves. So I'm not sure if that answered your question question but hopefully it got us started along that line yeah well that does um neatly differentiate us from other animals even other primates when you um imagine that the that an awareness a comprehensive awareness of our own mind is of a piece with what we call theory of mind. It goes by other names like mind reading and mind sight, but it's the ability to represent the mental states of others such that you can recognize that other people can have, rather often, different beliefs and desires and expectations

than you do, and they can be at odds with what is in fact true of the world. Obviously, the famous test of this is to set up a little playhouse with some dolls and ask kids around the age of four. One doll leaves the room, and then another doll hides a cherished object somewhere in the playhouse, and then you ask the kid, you know, when this other figure comes back, where is he or she going to look for the object? It's only once they can develop the concept of another person holding a false belief that they can give the correct answer, which is he's going to believe it was where it last was before

he left the room. So yeah, I mean, remind me, I think while there's some possible basis for very rudimentary theory of mind in other primates, I mean, I think there's something like deception. It's still somewhat controversial to call it deception, right? I think we still don't think that other primates have a proper theory of mind. Is that correct? Yeah. I mean, it's an evolving field. And in fact, only in the past two or three years have there been studies suggesting that chimpanzees can represent false beliefs, at least to the extent of being able to shift their gaze towards where the object is actually going to be? Sorry, to where they think the object will appear from the perspective of the other person. But so far, at least, and I was reading a review on this recently from Laurie Santos and colleagues.

And so far, at least the picture is that even though if you use clever experimental techniques, you could get some hint that they can track false beliefs, at least in behavior, in terms of being able to act upon those and use those to guide behavior. It seems like there is a gulf there from the best experiments on chimps to humans. There's quite a gulf. And this is not human adults. This is, as you say, kids age around four. And what's really interesting there is that if you go back, that field of research, that field on animal theory of mind was kicked off by this famous paper back in the seventies, which just had the title of, does the chimpanzee

have a theory of mind and going back to that paper you what what the authors of that paper meant by theory of mind was the ability to think about other people's mental states but also the ability to think about your own mental state that term theory of mind has kind of got used most often in the literature to be about other people. What's interesting now, I think, with this rise of work on metacognition is that we're starting to think, okay, maybe this is just a more general computational capacity that subserves not only thinking about other people, but also thinking about ourselves. really interesting. This is a place where it does at least make a point of contact with

the self that I often denigrate as illusory. I mean, there's this sense that our sense of our representation of ourselves in social space and in the world is of a piece with our concrete representation of others as

others, right? That this really indelible sense of self and other emerges together, kind of a single cognitive brushstroke. And when, as many people can attest in, you know, experiences in meditation and, you know, with psychedelics, when that boundary between self and other erodes, you know, it erodes, again, it's kind of a single boundary where if you're not really reifying self, you're not quite reifying other in quite the same way. In the normal course of events where we feel like ourselves and surrounded by other minds, it does seem intuitive to me that we're doing something quite similar when we're representing other minds and reflecting on our

own. I mean, it's just we're thinking about the same kinds of things, and it's the angle of our gaze that is different. But this goes to many other results in neuroscience when you think of the mirror neuron research and just how is it that we interpret the behaviors of others when you see someone reaching for an object, you understand their intention in a way that maps on to what it's like to be you doing more or less the same thing, reaching for objects of that kind. There's a kind of mirroring component here in the way we understand other people's behaviors, and it is the research thus far, I think it's appropriate to be somewhat skeptical of just how much has been made of the mirror neuron research. But it certainly seems that there is a

kind of self-mapping that is the basis for our understanding the behavior of others. Yeah, I think there seems to be a lot of circumstantial evidence surrounding that linkage. It's really hard to pin it down. And what I find fascinating and somewhat frustrating is, you know, can we cash that out in a more computational terms? Like what is that system really actually doing? Assuming it is a system that is, as you say, building a model of someone else and also building a model of ourselves. But it does seem like that similar brain networks are involved. And we recently did a meta-analysis of all the

studies of brain imaging studies of metacognition and compared that to classical theory of mind networks. And there was interesting overlap in regions of the medial prefrontal cortex. And we know, for instance, in neurodegenerative diseases like dementia, decline in self-awareness is often accompanied by a decline in social cognition as well. And developmentally, they seem to go hand in hand in children. So there's a lot of kind of, there does seem to be a symmetry there, and I'm attracted to that symmetry. I just think it's hard to find a good way, and we are thinking of trying to do this, but it's hard to find a good way of directly comparing the kind of computations that might

underpin self and other evaluations yes we've just discussed that theory of mind is the the necessary precursor for deception because it's not until you understand that other people have beliefs and and representations uh that you can then manipulate those beliefs and representations strategically with an awareness that this is a likely way to produce a desired effect in their behavior. But then there's this question of self-deception, which again is a somewhat controversial topic scientifically. There are paradoxes that await us when we try to think of self-deception as being truly analogous to the deception of others, because then you're left with this quasi-Freudian picture of part of you consciously deceiving some other part of you. So the part of

you that is in the deception business must know the truth in order to strategically hide it or distort it for the rest of you. How do you think of self-deception or the phenomenology of being flagrantly wrong about one's inner life or outer behavior in ways that invite this analogy to deception? We do often summarize it as self-deception or willful ignorance. I mean, the willful part of it is perverse and inscrutable from a cognitive point of view. Where does that fit into the discussion of metacognition? Yeah, it's interesting in terms of how that might connect to this notion of belief decoupling from accuracy or confidence decoupling from performance because i think that is something we do see routinely

in many studies people's metacognition isn't very good they are sometimes confident that they've got the right answer even though it's clearly. And we know that there are all these biases in belief and confidence that people like Daniel Kahneman have famously documented. I think that one place it connects there to the discussion we were just having on theory of mind is that we model or we create narratives to explain the behavior of others. That's part of the depth of mental state inference that we can do that. We can say, well, they must have ignored me in the street because of what I did yesterday or something like that. There's a narrative that we create about the thought processes going on in other people's heads. And we seem to create a similar self-narrative and that can cohere more or less

with reality. And when it decouples completely, then we're in the realm of psychosis or confabulation. So I think that there are, you know, we can start building up a story about why beliefs or narratives might decouple from what is the ground truth of our behavior or how we appear to others. What I think is really interesting about your question is that I hadn't really thought about before is that does that then in some sense require a system to also know the truth internally? And it's not clear to me that that is the case, although I think it could be possible that that is the case in some circumstances.

So we've done a bit of work. This was work led by a former postdoc of mine, Dan Bang, who has been really interested in this problem he calls private- mapping, which is effectively how do we take our private beliefs and convert them into what we say to others. And so his example of this is, you know, what do you say to a kindly aunt who's given you a terrible Christmas present? And, you know, you don't want to hurt their feelings, so you say an untruth, but you do this strategically. And we studied that in the context of metacognition by being able to track using brain imaging, the confidence that was being formed at any given moment, because we have a fairly good understanding now of the neural correlates of confidence in

individual decisions. But then we required subjects to strategically adjust the confidence they communicated to their partners in a collaborative game. And what we found was that there were distinct networks involved in this private sense of confidence. How do I feel about my performance now? And another part of the prefrontal cortex was engaged when they had to strategically adjust that to communicate to the other person. So that would be, it's not quite deception, but it's some kind of strategic mapping between this kind of private feeling of what's going on and what we're trying to communicate to others for the purposes of strategic manipulation. So it'd be super interesting to know whether we're at some level doing that to ourselves, that at some sense, that same general circuit for strategic

manipulation of others is also working under the hood for ourselves. And I don't know of any work on that. Yeah, well, when you look at the structure of much of our thought, it is conversational. I mean, we are talking to ourselves much of the time as though there's someone in us who is listening, who needs to be told certain things. Otherwise, much of our discursive thought is totally superfluous. Why does part of you say anything to the rest of you as though the rest of you isn't aware of the thing that's being said. You know, if I'm looking for an object on my desk and when I spot it, I might say, oh, there it is, right, to myself, you know, silently with the voice of the mind.

But if I'm the one to see it, right, who am I telling, oh, there it is, right? Who needs that further linguistic information when I, the one who is in possession of the eyes that have seen it, is looking at it in that moment? And so, so much of our thought is dialogical that one could imagine a similar process is happening. The thoughts are tumbling out our mouths when we're speaking to others. process is happening we know we're the thoughts are tumbling out our mouths when we're speaking to others and then when we shut our mouths we keep talking to ourselves about more or less everything yeah and i'm i'm very attracted to the position that chris frith holds on this that in a sense and this comes back to the conversation about an evolutionary story of metacognition that why did we start building this if you'd like to continue listening to this conversation you'll need to subscribe at

samharris.org once you do you'll get access to all full-length episodes of the making sense podcast along with other subscriber onlyonly content, including bonus episodes and AMAs and the conversations I've been having on the Waking Up app. The Making Sense podcast is ad-free and relies entirely on listener support, and you can subscribe now at SamHarris.org. you