Dr. Stephen Fleming: So, metacognition literally means cognition about cognition, and the definition comes from the Greek "meta", just means "along with" or "above and beyond", and it's got this flavor of self-reflection, so this ability of the human mind to know that it knows something, and as a psychologist, we also interchangeably use the term "self-knowledge" or "introspection", but the basic idea is that metacognition refers to the brain's ability to reflect on and monitor other cognitive processes.

To take a simple example, you might be watching a film, and an actress comes on the screen, you say to yourself, "I know that I know that person from the film," but you can't quite remember it, and you keep on searching your memory, and you're literally putting in effort searching your memory until you suddenly realize which film or TV program she was in recently.

That's metacognition right in action. It's quite strange when you pause and think about it for a while. Somehow, that phenomenon shows us that your brain, or you, you know something about your memory even though you can't actually know it at the moment when you're realizing that you know that you know it.

Jesse Lawler: It reminds me of the old days of libraries back when they had card catalogs. Before Google, it's like you can find the memory card of this piece of information is in here somewhere, but the actual book isn't on the shelf when you go looking for it.

Dr. Stephen Fleming: Right, exactly. It's got this flavor of what computer scientists refer to as content-addressable memory. So, this idea that there's another part of the system that knows what the system knows, and as your example really nicely highlights, sometimes that goes wrong. So, sometimes, metacognition gets it wrong. Sometimes, we think we know something when we don't, when we're confident in a particular answer, when we've actually got it wrong.

Part of the game of studying metacognition in the lab is to try and tease apart changes in people's metacognitive abilities and understand all the various factions that come together to produce good metacognition.

Jesse: I imagine that, probably, the variance of one person to the next in their metacognitive abilities is as wide as their variance in anything else, that some people are metacognitive superstars, and other people can't tie their shoes, metacognitively speaking. What do we know about that, the degree of variance from one person to the next?

Dr. Stephen Fleming: There's individual differences in metacognition, just like there are individual differences in other aspects of mental function. A central finding in metacognition research over the past couple of decades is that metacognition is partly independent of the cognitive process that it monitors, and as I mentioned before, it can fall foul of various illusions or distortions.

A general rule of thumb from this research is that when a task feels easy, that we're more likely to be confident even if we're not actually doing very well. So, people differ in their ability in which they can discount these various influences on the metacognitive judgments to really tap into the underlying performance. So, people differ in how well they can monitor their objective performance, and this is something that we've done quite a lot of work on in my lab where we've been quantifying in tightly-controlled experiments, such as asking people to tell what they can see on a computer screen. On each judgment in that experiment, we asked people for their confidence in getting the right or wrong answer, and we can then build a picture of how closely that confidence reflects that objective accuracy, and what we've found is that that metacognitive ability, how tight your performance is tracked by your confidence does indeed vary between people. It varies independent of their objective performance on the task, so you can be doing the task quite well, but have poor metacognition and vice-versa.

We've discovered that there's particular reason in the brain particularly in the prefrontal cortex where there are differences in the structure and function of the brain areas in the prefrontal cortex that predicts how well someone would be at this metacognitive ability.

Jesse: I'm picturing somebody who says, "I'm not very good at this," but clearly, they are, and looking at that as an outside observer, your first inclination is to think, "Well, this person's just being modest." Intuitively, we think, "Well, this is something emotional or psychological about the person, not that they might be wrong about it in an objective sense." That must have been difficult to sort of disentangle in your studies.

Dr. Stephen Fleming: It's important to get straight, exactly, what we're trying to measure here. You're absolutely right that someone's overall level of confidence, or whether they have high or low self-esteem, whether they just generally feel confident in doing these kind of tasks is going to feed into these judgments. That's also something we're interested in the lab in a separate line of work where we're trying to think about what produces high or low self-confidence, self-belief.

In these studies, in metacognition, we're trying to actually kind of push that to one side. What we're actually interested in is not your overall level of confidence. We're actually interested how your confidence fluctuates from moment to moment, and whether those fluctuations track your underlying performance. So, you might actually be quite a low-confidence person, but your fluctuations in confidence might track your underlying performance, and that's what we're trying to measure in these kind of studies.

Jesse: What would be an example of something where people have a high level of confidence, they think something is easy, but they are not performing very well?

Dr. Stephen Fleming: There was a really neat set of studies done by the social psychologists, David Dunning and Kruger who asked people to carry out various real-world tasks such as getting up and giving a talk, or they even had one condition which was doing stand-up comedy, and what they found was that people who were actually performing quite badly at those tasks were also the ones who were more overconfident, and they called this the "unskilled and unaware of it phenomenon".

Super interesting result, I think what we're trying to do in our studies of metacognition is trying to kind of pull apart the different reasons that might go into that overconfidence. On the one hand, you have these differences in overall self-belief, which I mentioned before, and then on the other hand, you have this kind of more what we call in the lab, monitoring. So, how can you tell, from a moment-to-moment basis whether you're performing well or badly, and I don't think we know yet whether that Dunning-Kruger effect is mainly driven by poor-skilled people just being generally overconfident in that particular task or whether it's driven by failure of monitoring.

Jesse: I imagine that this is probably one of these cases where people that have gotten brain injuries in various ways could be really informative in seeing, "This piece of the brain is clearly critical in this type of self-monitoring," but, for non-pathological cases or cases where there's not clearly an outside influence on the brain, somebody got a firm whack in the head, what do we know about the learning process involved or developmental things that makes some people maybe more accurate in their metacognition than others?

Dr. Stephen Fleming: It's a really interesting question. It's one that's actively being researched at the moment. One thing we can say is that even if you don't have a brain injury, for instance, one of the key discoveries of cognitive neuroscience over the past few years has been that there are still subtle differences in structure that we can pick up on, for instance, an MRI scan, that are systematically related to subtle differences in behavior and personality, and that's definitely the case for metacognition. That's something we've replicated now in a few studies that people who have better metacognitive abilities have differences in the structure of their prefrontal cortex and the connections to the prefrontal cortex.

What we're trying to do now is use mathematical models to say, "If I was to design, say, a robot to have metacognition," which is a whole other set of issues there, but is I was to try and design a kind of really small toy bit of computer code that could do metacognition, what would I need to put in there for it to solve our tasks? It turns out there are a few pieces that you need to put in there intrusively to monitor your performance, you need to not only know what was the right answer, but you also need to know what you did.

So, it's kind of got this higher order of flavor to it. You need to both monitor the input and the output, and we think that there are areas of the prefrontal cortex that are doing those computations, and the next phase of studies will really be to kind of try to link that back to those individual differences. Do people who have better metacognitions, are they somehow implementing these computations better? That's an open area of study at the moment.

In terms of development, this is really interesting because up until a couple of years ago, people thought that metacognition was this aspect of development that really came in at quite a late stage, that it required this explicit, even linguistic ability to say something about your mind. But, now there's been recent studies, as is carried out in my colleague, Sid Coolidge in Paris. He's been doing very interesting experiments on six and 12-month-old infants using these implicit measures of how confident are they in searching for a particular toy, for instance, in a play scenario, and he finds that he can actually quantify them as cognitive ability and show that they do have some sense that they're confident in making a right or wrong decision.

Another really interesting line to take there is to say, "Okay, are these just initial germ seeds of metacognition that we see in the infants? Do they somehow predict whether you're going to have good metacognitive ability later in life?"

Jesse: It almost seems like there's this top-level metacognition of like do I know this thing or not, versus kind of the deeper level of trying to assess how your thinking process actually works. There are some things that we sort of have intuitions about how our thought process goes where I guess we kind of think more about the actual thought process rather than just, "Does this piece of information exist? Yes or no?" and getting sort of a binary answer out of it.

Dr. Stephen Fleming: I think that's right. I think, often, when we're studying metacognition in the lab, we're studying this monitoring process, which is relatively easy to study because you can just ask people occasional judgments of how confident they are in their answers. But then, you have this other aspect of metacognition, which is knowing about the kind of thinking processes that are going on. We know very little about that.

One thing we do is that it seems to be your metacognition for that, that your thought process itself seems to be even poorer than your metacognition administering of your knowledge. It's saying something.

Jesse: If you thought you were bad before, try this.

Dr. Stephen Fleming: Exactly. An example of that is if you put people in a setting where you get them to make a decision about, say, which face they find more attractive in a pair of pictures, and then you actually perform a trick where you say to them, "Okay, here's the face you chose," and you give them the wrong one, but you actually ask them to justify why they chose it. Some people, some of the time, won't notice the switch, and they'll also go on to justify why they chose it. This is known as choice blindness that you have this kind of blinder to what drove your choice in the first place.

That's an aspect of metacognition that we don't understand very well in terms of how it works in the brain, but it does seem as though we're just not very good at introspecting on what drove our decisions.

Jesse: I guess you want to point the finger at cognitive dissonance there and just say, "The earlier version of you from five minutes ago is still the home team. You're rooting for that person to be right," so whatever they said, you'll find a way of justifying it.

Dr. Stephen Fleming: Right, absolutely. I think there is certainly an element of self-justification to those retrospective defenses of what decision we made.

Jesse: How have your studies looked at things like meditation in other processes where a person is explicitly taking the time to look inward and contemplate rightly or wrongly what's going on within their mind?

Dr. Stephen Fleming: We haven't directly looked at that in our lab, but there have been a couple of interesting papers on meditation suggesting that if you're trained to meditate over a period of two weeks and you haven't done it before, then when we give you the same task again, then your monitoring is improved somewhat. I think there is something to that that this continued engagement, being more self-focused and self-reflective, probably will have plastic effects on the kind of brain circuitry that's involved in metacognition.

Another way that we've been looking at this is by setting up a training protocol where we can get people to perform tasks. We're actually doing this over the web, so people aren't coming into the lab. We give them a task to do over the internet, and then they can do it whenever they like. But, every day, they get a little email saying, "Okay, it's time to log in for your 15 minutes of metacognitive training now," and what we found is that, after a two-week training period, people's ability to monitor their decisions, to know when they're right or wrong is systematically improved.

But, what's really interesting is that then transfers to a new task that we give them after training. We then give them a memory quiz, and we find that people have better metacognition about that memory quiz even though they've never been trained on it. So, that just suggests to us that there's something core about -- there seems to be a kind of common mechanism that might be involved in reflecting on your decisions in lots of different tasks, and that's quite exciting for us because that means that we might be able to devise a setup to improve metacognition in cases where it goes wrong, and do so in a way that is going to help you in the real world.

Jesse: That is really interesting because skill transfer across domains is a famously difficult thing to actually pull off.

Dr. Stephen Fleming: Absolutely.

Jesse: Because metacognition can touch on almost anything, this might be one of the rare instances where learning one specific thing really well could have global benefits.

Dr. Stephen Fleming: I think that's right. I think the fact that metacognition has this domain general flavor to it suggests that if you can improve it at one domain, then it might be one of those abilities, and obviously, as you say, transfers a controversial subject. But, it does seem to be one of those abilities that might have more chance than others of transferring.

Jesse: Let's say in your studies, you find the top 5% of people in their metacognitive abilities, they accurately track their performance with their assessment of their own performance, what do we see as far as how that sort of scales up to how those people live and behave and have results in the real world? Are these people top income earners? Are they happiest in their marriage? What do we know about them demographically?

Dr. Stephen Fleming: That's really interesting. We don't know very much, actually. There's a lot of work done on, for instance, the relationship between your performance on the intelligence test, success in various aspects in life like earnings, and whether you're going to get a good degree and so on. We know much less about what metacognition gives you in everyday life, but I think it's become increasingly realized that, in lots of areas of life, just raw intelligence is not enough.

We know that metacognition in studies in educational settings, we know that metacognition is critical for learning. For instance, a student who reflects that they've not learned enough for an upcoming exam, they are making a metacognitive judgment about strength of their memory in maths, or physics, or whatever it is, and if that judgment is accurate, then that will prompt them to go and study more. The problem comes when someone has poor metacognition, and they think, "Okay, I've studied. I'm all good," but then they just flunk the test because they actually should have studied for longer.

I think, finally, through metacognition, it's clear from the studies in the lab and also in educational assessments, finally student metacognition helps us realize the limits of our knowledge, and should, therefore, link to all these benefits in everyday life, but those studies just haven't been done.

Jesse: Has there been much looking into the social aspects of metacognition versus right or wrong answers, sort of more binary-type things?

Dr. Stephen Fleming: There's not much yet. This is something we're becoming increasingly interested in our lab is how people maybe use their metacognitive abilities to not only reflect on their own thoughts, but also to infer what other people are thinking. There's an area of study called social cognition, and there's quite a lot known about the brain regions involved when we're trying to infer how much someone else is thinking. What we don't yet know is how closely related that is to metacognition about ourselves. It seems like it should involve similar processes, but that's not so well known yet.

What we do know is that having the ability to share confidence, or to share our assessments of our own thinking process is very helpful to combine opinions in a group, for instance. There's colleagues of mine at UCL who have done very interesting experiments on this, precisely quantifying how sharing our metacognitive judgments can then help people make better group decisions.

Jesse: One of the things that I find myself wondering about this is sort of the theory of mind, and one of the questions that always comes up: do we have the ability to think about what others might be thinking because we have the ability to introspect that came first, or did the evolutionary usefulness of being able to predict what other beings in our social world are thinking, then we sort of turn that same gaze inward, and that gave us metacognitive abilities as sort of a nice bonus?

Dr. Stephen Fleming: This is certainly something we're pursuing at the moment. We're analyzing experiments on exactly this issue. One thing we can say is that the brain regions that are involved in metacognition by yourself seem quite different to the brain network that is engaged when you're thinking about other people. So, just on that basic results, that just suggests there are going to be different processes involved. There have been quite a lot of speculation in the literature about, "Wouldn't it all be neat and nice?" it makes a very compelling story if the way that we gain self-awareness back in the days when we were sitting around in caves was we suddenly applied theory of man to ourselves, and we were like, "A-ha, I can think as well."

But, I don't think it's going to be like that. I think the neuroscience on this issue is going to probably turn out to be that we have quite different systems for self-monitoring and inferring the mental state to others.

Jesse: You mentioned the studies that they've done in six and 12-month, year-old human infants. Have there been any animal studies that have had bearing on metacognition?

Dr. Stephen Fleming: There have, and I think they have also benefited from these techniques where we can now measure metacognition without requiring people to actually make a verbal report about what they're thinking. You can set it up such that instead of making a confidence judgment or a rating about your decisions or your memories, you can set it up such that the animal can bet on whether they've made a good or bad decision, so they can make a second choice, which they get trained is rewarded according to the accuracy of their first choice. So, this is kind of like a metacognitive wager on whether they think the first decision was right or wrong.

This has been used to great effect in studies of apes, and monkeys, and even rats can do this effectively. It suggests that there's this kind of core computation or core circuitry that we've all been monitoring and computing confidence in whether things have gone right or wrong that is shared in several animal species. What we don't yet know is to what extent human metacognition is similar to animal metacognition. We know that they're some of the core components we share, but we don't know whether the whole set of components are shared.

Jesse: To what extent do you think our ability with language does or doesn't help us metacognitively speaking? In some sense, that tip-of-the-tongue phenomenon, you just kind of have a feeling. It's not something that you're linguistically thinking your way through. You just kind of have the feeling like, "I know this thing or I don't," whereas other introspections feel self-narratized in a way.

Dr. Stephen Fleming: Yes, I think you touched upon something that is actually quite related to the animal studies, because what most people agree on in this area is that animals have that feeling that something has gone right or wrong. Whether it's conscious or not, that's a whole different issue, but they seem to have this kind of lower-level feeling that feeds into metacognition, and that's sometimes referred to as implicit metacognition. So, it's just kind of this lower level of monitoring.

We also have this ability to create an explicit narrative about what's going on in our own minds, and that's probably something we don't share with animals. There's been this recent proposal in the literature that we can start to divide up metacognition into both its lower-level and higher-level aspects, and that this explicit narrative we create, that's not very well understood, but we think that it might be related to the brain regions that I talked about earlier. These regions towards the prefrontal cortex that are particularly enlarged in humans, actually, compared to non-human primates, like monkeys, and we don't really know why they're so enlarged relative to monkeys.

One hypothesis is that these regions, the contribution they make to metacognition is really facilitating this explicit report, this explicit narrative about our behavior.

Jesse: What proportion of the work done in the brain would sort of be classified overall as metacognition? Checks and balances on what we know and how we're performing, we go back to the library analogy of the books in the library versus the card catalog, what percentage of the brain is the card catalog versus the books?

Dr. Stephen Fleming: That's a really tough question. No one's ever asked me that. We can start to approach that question by thinking about systems in the brain that are involved in primary sensation like vision, and touch, and hearing. They all have dedicated areas of the brain for processing those inputs. So, you can kind of think of those areas as being what we call first order. You can start to divide up processes into whether they're first order or whether they're second order, so they involve this metacognitive aspect. So, there's clearly a lot of sensory areas in the brain that are more first order that you can also think of kind of memory storage as being more first order.

The problem is when you start thinking about higher functions like decision making, it gets really hard because, certainly, in the lab, we talk about first order decisions. You make an initial decision to a stimulus, or you make an initial decision about whether you can retrieve something from memory, but then probably, at the very same time, you're also engaging these second-order processes.

It's very hard to then say, in the brain, whether something is first order and second order, and because another thing that makes doing neuroscience really messy, but also really interesting, is the fact that everything is connected to everything else, and that you also have a lot of top-down connections. Even those sensory areas that I mentioned are also getting a lot of top-down inputs from regions that we would typically think of being involved in higher cognition. I think it's quite hard to say that is definitely not metacognitive or that is definitely metacognitive, but there's a kind of a gradient in the brain, if you like, of more metacognitive or less metacognitive processes.

Jesse: Is there any parallel between metacognition and proprioception, the mental sense of where your physical body is in space? It almost seems like proprioception could be thought of as metacognition for the motor cortex.

Dr. Stephen Fleming: It's interesting. I think most people would think of proprioception as being this first-order sense that you kind of have these inputs which tell you about where your body are, but there is, certainly, a recent suggestion that our ability to sense what's happening in our body may feed into metacognition in ways that we're not really aware of. There's been experiments done to show that changes in, for instance, our heart rates or our pupil size, these things that are controlled relatively unconsciously, they are related to our metacognitive judgment. So, it seems like there are all these things happening in our body that are being monitored, and that monitoring seems to bleed into our metacognitive judgment in ways that we don't really understand yet.

Jesse: Do we know anything about how a person's aptitude at metacognition changes over the course of their lifetime? Have there been any studies following people, tracking them with age?

Dr. Stephen Fleming: There have, and what we know is that metacognition seems relatively well-preserved, actually, in older age. So, what you might imagine based on the neuroscience of metacognition is that because metacognition involves these prefrontal regions that we know decline in function with age, you might imagine metacognition also declines, and there are hints of this. We actually published a paper suggesting that metacognition does slightly decline with age.

But, we've also done more recent studies with many, many more people, and we think that, if that decline is there, it's much less than you might imagine based on a complete drop-off in prefrontal function, for instance. What we think might be happening there is that people have developed solid strategies for monitoring what's happening, and you don't really need to learn those strategies anymore when you're older. So, you're tapping into the right cues that predict whether you're going to be right or wrong, and so you don't need to have this efficient online monitoring anymore. That's one hypothesis about what's happening in aging, but it does seem that, in healthy aging at least, metacognition is relatively unimpaired.

Jesse: It seems like the brain would also want to preserve, as long as it possibly could, its metacognitive abilities. If you had to sacrifice something, that would not be what you want to sacrifice.

Dr. Stephen Fleming: Right, exactly. If your cognitive abilities are declining, it would be quite useful to at least know that that's the case, so you can then take steps to adapt. You can see that being the case in cases of dementia, for instance, where metacognition does take a really big hit, and this can lead to serious problems with monitoring behavior, because if you're not aware that your memory's failing, for instance, then you're not going to take steps to adapt to these changes and so on.

This is what psychiatrists often refer to as lack of insight, and we think that this is partly due to the disease attacking brain regions involved in metacognition, and I think that that kind of work is quite important, because at the moment, that lack of awareness is often ascribed to denial or someone just being difficult, and it can be quite frustrating and heartbreaking for families. But, if we can start to understand what metacognition is and communicate that research to families and say, "No, well actually, maybe these diseases are attacking brain regions involved in metacognition," then they might understand more why people with these diseases are suffering from lack of awareness.

Jesse: Tell me about the current studies that you have going on and also what you see is sort of the takeaway findings, what we've learned about metacognition that people could apply in their own lives.

Dr. Stephen Fleming: The open areas of investigation in the lab at the moment are really focusing on trying to understand more about whether metacognition is really this one central resource that we apply to lots of different tasks. I mentioned earlier that it does seem that training metacognition in one domain allows us to have better metacognition in another. But, we don't know the limits to that, and we don't know, for instance, in the brain, whether the brain regions and circuits that support metacognition in one domain, are they the same or different as the brain regions supporting metacognition in another domain. So, we're doing quite a lot of studies on that issue, and then we're also doing more studies looking at the computations that support our ability to generate confidence in our decisions.

One exciting line of work there is to really treat metacognition as another decision that the brain is making, but it's now not a decision about the outside world, it's a decision about its own processes. So, you can think about it as like a decision about a decision. It's trying to build models of how that might work and why that might go wrong.

Then, in terms of lessons to take away, I guess one is to recognize that metacognition is important in guiding our learning or study in any particular area, and often that's underappreciated. So, people just assume that if they think they've learned something, then they will be fine on the day. But, often our beliefs about what's good or bad for our learning are incorrect. So, just knowing that can then help prompt us to take a bit more care, if you like, self-care in ensuring that we know a particular topic before going into an exam or going into a job interview, or so on.

One really efficient way of doing this, and this is not from the world of metacognition, it just comes from the world of memory research, one really efficient way is to test yourself. Rather than just relying that you think you know the right answer, if you objectively test yourself on a particular topic, then you'll soon find out whether you really do know that material, and people's beliefs about whether testing themselves is good for them are woefully incorrect. That's why I think people who don't think testing themselves is going to help them at all, but it turns out to be a really useful way of finding out whether you know a topic.

One other aspect that we think is going to be important in the future, why it's useful to study metacognition is that we think it might be important for interacting with AI and other technologies. So far, I've talked about how metacognition is applied to yourself, so how we know that we know something. But, we can also talk about metacognitive knowledge about other people's knowledge or even what different pieces of technology you know.

When you start a job in a new place, for instance, it's quite daunting because you don't know the right people to ask to get your ID cards, how to work the coffee machine, so on. Once you've been around for a while, you have this information, and this is what psychologists called "transactive memories". It's the memory that we have about other people's memories. I think that this topic of transactive memory is going to be important for understanding how we interact with technology, because now we don't only need to know what I know and know what other people know, but we also need to know what a particular piece of AI such as Google knows, and if we're confident that it knows, then we don't have to store the actual answer. We just need to store some metadata about where to find that answer.

This is very early days for this area of study, but we think that the way that we store that kind of metadata is a type of metacognition, and it might rely on very similar processes involved in metacognition about your own memory, and if that's the case, I think it's going to be an important area of study for understanding how we can interact better with technology.

Jesse: I'm having a hard time wrapping my head around that idea. I don't see the difference there between I know what I need to dig that I go out and grab the shovel, like a shovel is for digging. Google is for knowing how to look up recipes. I don't see that as being quite the same as metacognition.

Dr. Stephen Fleming: I think it has the same computational structure, if you like. If I take the library card example, you have the ability to kind of index your own memory and know whether you know something, and what we don't know yet, and it could turn out to be -- this is just a hypothesis, and it's something we're actively working on, but it could turn out that the ability to index a memory store is actually quite general, and the ability to index what another device knows, it may rely on similar mechanisms involved in metacognitions. It's not just saying, "Do we know what the device is for?" It's really saying, "How do we index the knowledge that another device has?"

Jesse: I'm seeing that as being a fairly fuzzy line between what something knows and what something can do.

Dr. Stephen Fleming: Yeah, I think that's right. I think what's interesting about this area of psychology is that it kind of got going in the late '80s. A psychologist at Harvard called Dan Bergner who was really interested in this topic of, "What do we know about what other people know?" But, really, it dropped off after he stopped researching that area. I think there's not a lot known about how we store those transacted memories. But, I think it is something that, we as a society, need to study if we're going to be interacting with these machines that know a lot. How do we answer that question of how we know what something else knows, I think, is philosophically quite a deep question, but hopefully, we can get some traction on it by studying how people solved these problems in the lab.