Full auto-generated transcript below. Beware of typos & mistranslations!
Nick Jikomes
Can you start off by just telling everyone a little bit about who you are? And what your lab studies?
Tobias Egner 3:59
Yeah. All right. So I'm a professor of psychology and neuroscience here at Duke University.
And me and my lab were interested in studying cognitive control. Like there's a term that covers the idea of using internal goals to guide behavior. Other terms for this would be executive functions that psychiatrists prefer to use that term. And we're trying to study
how we sort of regulate control attention and things like that by means of behavior. So we do a lot of behavioral experiments where people just are asked to follow certain instructions and push buttons in response to stimuli. We sometimes use a little bit of computational modeling to understand better the underlying cognitive processes that mediate that behavior. And we also do a bunch of brain imaging. So if we have an interesting, you know, phenomenon that we can reliably show behaviorally, we can take participants and usually we put them in an fMRI scanner, and then scan their brains while they're performing the sorts of tasks that we develop with the behavior to understand which brain regions are activated in relation to which cognitive processes that we try to isolate in the behavior. And then additionally to that, we occasionally also use a technique called transcranial magnetic stimulation. So this is a causal intervention, if you will, where you can non invasively briefly mess with a particular brain area and healthy participants while they're performing a task. And that allows you to get at sort of causal evidence for a particular brain region being important for performing that task. So you can briefly mess with this. And the prediction would be that, oh, if this region really is important for say, up regulating attention on a task, and you knock it out for a little bit, then that should be reflected in the behavior.
Nick Jikomes 6:06
And so yes, you study cognition, we're going to talk a lot about things like cognitive flexibility, cognitive control. But I just wanna start off by asking you kind of a big question, what exactly is cognition, as opposed to other things that we might talk about, like, emotion or perception? Yeah, that's good. Good question.
Tobias Egner 6:27
You know, cognition is essentially this study of knowledge of information processing. And most typically studied in human but can of course, be studied on other systems, cognitive science is interested in how information is being represented and computations are performed on those representations. Of in our case, in psychology, that tradition is really that cognition kind of covers, is sort of an umbrella term that really covers most of the mental processes that you would think you find in a textbook, including perception, attention, memory, executive function, and so forth. In many ways. All of this could be concerned Kotori, considered cognitive processing. And so broadly construed cognition covers sort of the whole gamut of psychology, if you will. It's got so you can consider it an approach that is taken in psychology on studying, you know, mental phenomena. So cognitive psychology, take sort of an information processing perspective on asking questions about our mental life, where I say, social psychology might take a slightly a different lens on that.
Nick Jikomes 7:47
I see. So there's not necessarily a hard and fast distinction between things like cognition, and memory and perception. And I guess, you know, another way of sort of coming up this question is, you know, to what extent are these human constructs that we use to make it convenient for us to talk about things, versus, you know, the lines being very blurry between, you know, when we look at the brain, you know, can you say, That's cognition versus that's perception? Because, you know, it is clearly different? No,
Tobias Egner 8:18
I would say not, and cognition would really be a more sort of an umbrella term, you know, information processing in the brain, if you will, that, you know, subsumes things like perception, attention, and so forth. Now, you mentioned the topic of emotion earlier, they, of course, there is a very strong tradition in the fear to distinguish you know, cognitive processing from affective or emotional processing. But even you know, it's sometimes people call it code processing versus hot processing, you know, that the rational cognition versus the irrational emotion. But in reality, most people in the field don't like that distinction. And don't believe in this in the sense that there's a lot of gray area there. And, you know, emotional responses are typically about, you know, analyzing your environment, inferring that you might be confronted by a potential threat, and that will elicit a fear response. And really, strictly speaking, you could call these cognitive processes also, you know, you're using your sensory input, but combine that with prior knowledge to make inferences about your situation, and that then elicits an emotional response on in that sense, you could argue that even emotional processing is a form of, of cognition.
Nick Jikomes 9:43
When we think about things like we might call forms of like higher cognition, executive function, thinking about problems in a very deliberate and conscious way. These things often feel very effortful. Like it takes it I really have to sit Alan and do my calculus homework or something like that, I have to tune out other things, I have to, you know, make a decision to do it, I have to go through all of the operations in my head. Whereas, you know, if we just sort of think about, say, sitting in our chairs, you're passively not even talking to each other. I'm going to see everything sort of passively, but I can't really shut off my visual perception. And I'm wondering if you talk a little bit about sort of like the the demands that things like just passive sensory perception have, as opposed to things like, deliberate cognition when we're thinking about a problem, and I want to actually come up that from the level of like, metabolism, you know, when I'm when I sit down, and I work through a problem, and I'm really thinking hard about something it feels effortful, like it's requiring more energy is is that in fact, the case? Do certain forms of cognition require more ATP than other things that the brain does? Well,
Tobias Egner 10:58
you know, the energy sort of metabolism underlying that, I would say, that's a little less well established, it's not as simple as saying, well, you're doing a more effortful task. Now you have like this additional, you know, glucose consumption, or so there is no doubt, a correlate to that. But there's also a long history of people trying to put their finger on that and not necessarily being successful with this. There's a whole literature and so called ego depletion, where initially that was sort of suggested as a pretty sort of simple underlying account. But that did not turn out to be a very successful. Having said that, no, it's it is absolutely, there's no doubt that we Reese perceive cognition, active cognition is something effortful. And one nice way of measuring this is that people actively avoid more difficult tasks for easier tests, all else being equal, of course, there is a another consideration whereby sort of a minimum level of challenge makes things a little more interesting, right, there's something called desirable difficulty where you might find doing nothing boring, and that sort of aversive tool. But you know, as soon as you come away from that, and you put people in the lab, and you say, give them the choice between picking from different tasks, they will tend to stay away from the tasks that take more cognitive effort. In fact, you can set up experiments that show that you can, they will even forego possible rewards in order to do a slightly easier task. So you can, through clever experiment in manipulation, actually put a monetary value on this and how much they'd rather not do a more difficult task than an easier task. One way of manipulating tasks. And this is something that we'll probably talk about later, when we talk about cognitive flexibility. One way of manipulating effort is, in fact to have people choose between options where they can either do sort of one task at a time for a bunch of trials in a row, and then maybe another one for a bunch of trials in a row versus another condition where they would have to switch a lot between the two tasks. And that, you know, we know that switching is hard, because it takes time, it takes a little extra time, and you have more scope for committing errors when you have to switch more between tasks. But yeah, it also is subjectively perceived as more effortful, and effort in that sense, is a little bit aversive. So people would rather not have to do more switching. And you can quantify that quite nicely with with these kinds of behavioral tasks actually.
Nick Jikomes 13:50
And, you know, one area that I want to spend some time on, is the relationship between memory and perception. And so I'm hoping you can talk about that, especially as it relates to this concept of predictive coding. So can you talk about memory and perception and how they interrelate? And what that has to do? And what predictive coding is?
Tobias Egner 14:11
Oh, that's interesting. Yes. So, the idea is that a perception when we think about it, and often sort of naively think about it, right? The old metaphors for perception are something more along the lines of sort of bottom up style metaphors, we think of like, oh, it's taking photographs or something like that, where you really have to analyze your external world the the stimuli that impinge on your say, RetinA, for example. We'd have to sort of analyze them from scratch. As you go through the world. Everything is computed bottom are the things that surround you. But in reality, this is not the case. And this is not what we have to do. And this is thanks to memory, right. In reality we go through the world, man Accessing external inputs to our internal representations or memories of the things that we've seen before. And that allows us to identify, you know, objects around us very, very quickly. So most of the time we spend our perception really consists of recognizing like recognition, we recognize something by matching something that we've already seen before, and that we have knowledge about to a to that knowledge. And that, of course, makes it much easier to and much faster to recognize things. If I turn around right now, and then turn back to face you, I have my visual system already knows what to expect. And it can use this expectations to save itself a lot of processing, right? If I had to compute this all from scratch, that would take me longer and be more effortful, then relying on expectations or predictions to men, I just have to check, hey, which parts of this current input are deviating from my expectations, right. And it's called prediction error. And in principle, I only have to sort of update my perception as a function of these perhaps quite minor prediction errors, because my expectations are pretty reliable from one moment to the next. And this, you know, is grounded in the fact that our external world is going to the world, our input has a lot of what's technically called, you know, high temporal autocorrelation essentially means that what I'm seeing right now is a really good predictor of what I'm going to see in the next second and the next second in the next second, right. So I can use that knowledge to make generate predictions for future input. And then I only have to really process a small subset of that input. The little bits that are maybe unpredicted and update and I go, Okay, this was a little different than I expected. But most of the time it fits in, I don't have to bother computing everything from scratch. That's sort of the idea of of predictive coding that you can get away with relying on predictions, and only have to update those in proportion to sort of the errors in your prediction.
Nick Jikomes 17:19
And what does this have to do with say, the stability of our memories? So, you know, our memories are not like photographs, where you know, something happens. And there's a, there's a snapshot that's frozen in time completely. But I think many people understand not all people, but many people understand that our memories are updated, and they're malleable. And they change all the time. And you know, you were talking about our need to, you know, update these predictions about what we think we're going to perceive. And so what does this idea of predictive coding have to do with the fact that our memories are not these static images of something that happened, but they're these more changeable representations? Yeah,
Tobias Egner 18:00
that's a good question. Memories, of course, sort of dynamic, or people think of them nowadays as really being constructed rather than we refer to retrieving memories. And that is, you know, correct up to a point. But in a sense, this is a carbon copy of what we experience. And each time that we retrieve a memory, we bring it to mind, and we then re encode it, as well. And each time we do that the memory is slightly altered. So by that logic, if you use memories to get around the world, which we kind of do all the time. So there's really this this perception memory interface. And there's many people that think about the the medial frontal lobe system where the hippocampus sits that is important for memory encoding and retrieval. Many researchers now think of that as sort of really a perception memory interface that allows us to move through the through the word, but yeah, then when you use memories to anticipate your perceptual inputs, and those perceptual inputs deviate from the memory, then then memories get updated and re encoded. And then you have a new memory that you can use, perhaps the next time around to anticipate that sort of minor variation that you've just encountered. The whole predictive coding thing is a really big area. And I have, you know, I've done work in this years ago, a bit at the level of sort of visual cognition where we were just interested in, hey, this theory, it makes these predictions about, you know, prediction, error signals and perceptual in visual cortex, for example, which was not really something that had been anticipated by other views of the brain. We knew about prediction errors in the reward system based on what sort of reward reinforcement learning studies and things like that. So we were intrigued by this and we're testing whether you could really see a visual surprise signal and visual caught texts using neuro imaging. And you can, but yeah, the depth field as such is really rather large. And there's a lot of kind of to and fro about the specifics of, you know, what predictions exactly this theory makes, and so on and so forth. And I'm not necessarily, I'm currently sort of at the forefront of that. But broadly speaking, I think it's interesting that the, the ideas around predictive coding have certainly had a huge impact on on many different fields, the generic idea of the brain being sort of a predictive organ, I think, is now sort of very much a mainstream assumption in, in psychology and neuroscience, I would say,
Nick Jikomes 20:49
you know, one important form of cognition, especially for primates, and humans, is working memory. So what is working memory? And can you give us a sketch of how it works phenomenologically and some of the key areas or networks in the brain that are crucial for working memory? Yeah,
Tobias Egner 21:11
working memory is a really, really important construct in the sort of at the center of cognitive control and executive function is where really, it's notion that we can encode information either from the environment or maybe retrieving information from from long term memory. And we can keep it in in mind, so temporarily, and use it to guide our behaviors. simplest example of this is few, you know, give me a phone number verbally right? Now, I can keep this information in mind for the next 20 seconds or so and then use it to guide my actions to actually type in the phone number and the phone. So even though in the meantime, right, this information is no longer available to my senses, right. So working memory or short term memory, allows us to keep information in mind that is no longer out there in the world. But that's still relevant to us, and use it to guide our actions. Now, this example, I just use with a telephone number, this is something that people would consider declarative memory, or working memory. So memory for items, objects, things, that kind of thing. But working memory is also crucial for keeping the rules of the game in mind. So what's currently what are my current goals? And how can I achieve those goals? And this is sort of where it intersects a lot with the topic of cognitive control. So we use sort of temporary representations of our current context, you know, am I in my office and my home and then somebody else's house Am I in the car, and combine this with our current goals, to then guide how we evaluate stimuli? How we respond to stimuli in software, working memory is very important for that, because it allows us to do this in a flexible way, because it's not the same rules don't always apply, right? We are able you and I to take, you know, a toothbrush and do very different things with it, depending on what our current goals are, is it to clean my teeth? Or is it to clean a little like knock on my stove that I can't reach with another cleaning instrument? Right? And, and working memory has for a long time been associated with the frontal lobe of the brain. So there have been famous really famous very early neuroscience studies, the most influential one or earliest one perhaps was in the 1930s by a guy called Jacobson, who was working with, with monkeys and macaques, I believe, and who did brain lesions lead of the frontal cortex in these monkeys and had them do classic sort of a short term memory task or a delayed response task where the monkey is being shown that through little food wells in front of his cage, so little things were with with snacks in it, and one of them wouldn't have a snack in it, the other would not, then the monkey, there's a covered so the monkey cannot see them anymore. Now in order to pick the right Well, later on, he has to remember which well was the one with the food in it. So that's the working memory component. And then say after five seconds, 10 seconds, 30 seconds or so. The monkey is free to pick one of these worlds and if he picks the right one, he can eat the snack if he gets the wrong one, he doesn't get a snack. So during this kind of experiment, Jacobson found that when he severely lesion the frontal lobes of the monkey, this short term memory function would basically go away so so the monkey would know longer be able to keep in mind where this reward is hidden. And this was the first time that the frontal lobes were associated with this short term memory function. And then in the 1970s and beyond, people started looking at invasive electrophysiological recordings in this area in relation to work in memory. So the famous early study at Hodgkin footstone colleagues where they demonstrated for the first time or if you insert a micro electrode in the macaque lateral prefrontal cortex during a working memory task, just like that, you can find neurons that start firing vigorously when the cue is being shown to the monkey, and then sustain this firing throughout the delay period. So say those 10 seconds or so when the monkey has to keep in mind the location where the food is hidden. And then, as the monkey gets the food that neuron stops firing. And the idea is, was then and many people still believe in that, oh, there are these prefrontal neurons that literally, their sustained firing kind of keeps this information active, even though it's no longer available to the census. And that this allows us to then, you know, bridge this time interval and use this information, in this case to obtain that reward. Now, by now, that's been obviously, decades of much more detailed neuroscience studies been performed. And the overall picture, I think, is still that the front and end the parietal lobe are both really important for visual short term or visual working memory. And most of the literature uses visual stimuli for reasons of ease, you know, very easy to use, and we're very visual species and so the monkey's butt the exact way that that single neurons contribute to how these memories are being sort of kept alive all the time, that view is sort of has changed, or there's, there's a fair bit of controversy about how exactly that works, whether these single neuron sustaining their firing is really the most relevant signature, or whether they are much more complex of dynamics that keep inflammation going, that sort of very much an active ongoing, mine of research,
Nick Jikomes 27:23
and how much you know, how much individual variation is there in, in working memory capacity, say, in monkeys, or in humans?
Tobias Egner 27:34
Quite a bit, I would say. And those, you can kind of probe those in in different ways. So obviously, different people are more or less capable of storing, you know, different amounts of information, how information is defined is actually a bit tricky there, there's a, you know, a long history in cognitive psychology trying to identify, you know, the capacity of working memory, there's a very famous paper by George Miller, famous cognitive psychologist that said that working memory capacity is sort of the you know, the magic number seven, I believe it was back then reset plus minus two or something like that reset, where you know, for most people, if you give them some sort of random stimulus material, they can keep around seven plus minus two items in mind quite nicely. But that really depends on what the stimuli are, for example, we have much better working memory capacity for objects that we know that we have sort of semantic meaning to us, then if I give you some sort of random scribble or some fractal or something like that, that you have no prior memory representation of nothing that you can verbally code about this also, the exact capacity of working memory very much depends on the stimulus material that you use, and it can be probed in a variety of different ways. One type of working memory assessment that tends to be particularly useful in individual different studies. So when people are really interested in say, relating working memory capacity to things like IQ or, or other capacities like that, those are so called complex span tasks. And there. The trick is that you are asked to sort of encode information sequentially and might give you a string of numbers or, or objects to remember, but between the times that are showing an object and the next and the next, you also are being sort of kept busy by having to do little math equations, for example, so that you can easily recode this stuff verbally. You really have to work hard to keep it in mind, and he can't, you know, rehearse it under your breath or something like that, because you do these other things at the same time, and those complex span, so complex plans, then they show quite a bit of individual variation, you know, some people might have a complex span of five and others might have up to 12, or something like that, in that span is, has a pretty reliable association with sort of other, you know, high level cognitive capacities, like IQ, for example,
Nick Jikomes 30:35
how you see a lot of stuff these days, you know, advertised on our phones and on our computers about, you know, training your brain, and, you know, apps that claim to be able to do this, for something like working memory, how trainable is it within an individual? Can you practice it and get better at it? And does it? Does it change systematically over say, our development? Does it sort of peak at some point in our lives and go down? And how much? How much can we do about our working memory? Can we? Can we sustain it for longer if we train it in certain ways? Or how malleable is it?
Tobias Egner 31:12
Yeah, that's a very interesting and important question. Also somewhat contentious. The the overall, I believe this is the accurate the overall picture in these sorts of cognitive training regimes, including working memory is, of course, a prime target for cognitive training, it tends to be the case that you can get better at a particular task, in fact, the one that you're training on working memory, so if you train the working memory span, with a particular paradigm, and particular sets of stimuli, you would probably get you, you will most definitely get a little better that over time. But people have tricky, typically found it very hard to transfer these gains to other tasks. So it tends to be very domain specific, which is, of course, not really what you want, you would like to be able to train somebody to be generally better at working memory, and then ideally, have knock on effects for, you know, all manner of high level cognition. But, you know, I'm now you know, I'm not doing this sort of individual difference research myself much. But I think I know that literature well enough to say that, it is still a case that most most academic researchers say, Yeah, we have not found sort of the the magic approach to generalized training gains beyond the special domain that you're training, and even within that, it might be, you know, restricted to the sorts of stimuli that you've been using, and so forth. So that's, you know, in working memory and related domains, the question whether this sort of cognitive training, really works outside of the exact parameters within which you train is sort of up for grabs, and people are trying to figure out what the parameters might be that would allow for more transfer to other situations rather than less. But in general, I would be very skeptical of apps or other things that people try to sell you that they say, will somehow you know, enhance your general cognitive abilities. It's probably an over selling, what their what they can do.
Nick Jikomes 33:30
And do we have any sense for, you know, neuro physiologically, what is accounting for how well one is encoding information and working memory. So for example, if you take two individuals, and you give them a string of numbers to remember, maybe one of them can reliably remember seven at a time, maybe the other can remember 10 at a time, or maybe within a single individual, right? You know, if I'm sleep deprived, I'm not gonna be able to remember as much as if I'm awake and alert, what's accounting for the difference there? Does it have something to do with the pattern of neural activity in places like the frontal cortex? Does it have something to do with how metabolically efficient the neurons are? are being what is what is accounting for the quality of say, a working memory task performance and neuro physiologically?
Tobias Egner 34:19
Yeah, I don't think that anyone has a conclusive answer to this. And you can most certainly find that, for example, you would find variation in the you know, overall activity level and say, the lateral frontal cortex, you know, where you can you can show for example, if you ask people to encode three versus five versus seven items, right over a delay period, you can show that Oh, in frontal and parietal cortex, the level of activity increases sort of quite lawfully, with just the number of items you're being asked to encode. So clearly, these regions are doing Something to make that happen and then hit they sort of hit a ceiling at the level where you can no longer add more more items. But the actual underlying, you know, nitty gritty, you know, what computations are these neurons doing? And why are they constrained in this manner? That is not conclusively understood, it very likely would have to do yeah, as you said, with, with patterns of groups of neurons that might, for example, be able to form the sort of spontaneous ensembles to keep information available. And there might be a natural limit to the number of these ensembles that varies a bit amongst individuals, you know, how many different in dynamical systems language that could be, you know, how many different sort of space partitions or attractors can you sustain at any one time, but ultimately, that's just really describing the same phenomenon using different language, right, we have some how these clear limits on how many items we can keep in mind, and they differ between, between and within individuals. The exact reason for that we don't know that we know roughly where in the brain, this happens, and thus, most likely which regions or dynamics within which regions will be responsible for this. But to understand there's really at the at the level of computational principles of neuronal ensembles. They are good hypotheses about this, but no conclusive knowledge as far as I know.
Nick Jikomes 36:43
And you know, you do a lot of work on cognitive control and cognitive flexibility. Can you start to talk about those things a little bit more? And give us a sense for how you study them in the lab? What's an example of like a task and up population and an intervention that allows you to understand which parts of the brain are at play here? And what do you actually having people do in some of these experiments? Yes, yes, indeed.
Tobias Egner 37:05
So let's pivot to this. So when I mentioned the the notion that working memory you can also hold in mind, you know, task sets are the rules of the games, what am I doing right now, and am I doing it, this is a an important core construct in cognitive control. So in order for us to evaluate or categorize a stimulus, for example, in a way that it's not the way that we typically do that. So we'll have to override some kind of habitual response. This is when we require cognitive control. A classic example is if I, you know, if you fly to the UK right now, and you rent a car, you suddenly have to change the rules of the game, because now you have to ride a ride on the other side of the road, and you will find this quite effortful. And you will probably even find yourself initially, you know, talking to yourself reminding ourselves of the rules of the games that are driving on the left to drive on the left. And this is sort of at the core of cognitive control, the idea that you can use this temporary rule, you keep it in working memory, and it can guide your actions, and it can guide them in a way that can counter previous sort of habitually, are very much learned responses. And this is what makes us flexible. Now, when we do that kind of thing in the lab, this particular type of control, you know, imposing a sort of an arbitrary task set to override habitual responding is most classically studied in a task called the color naming stroop task, you may have heard of this. But for the benefit of your listeners, let me describe this to you. So this in this task, I will put you in front of a computer screen. And on each trial of the tasks, every couple of seconds or so I throw onto the screen, a color word could be red, blue, green, yellow, and so forth. In those words are printed in different font colors, again, blue, red, green, and so forth. So that means that sometimes you could see the word red printed in red, but other times, you might see the word red, printed in green, or printed in blue. And what I'm asking you to do is something unusual, because I'll ask you to name the font colors rather than read the words. So you would say have to push a button that corresponds to yellow, if red is shown in yellow, even though the word says red. And this sets up a situation where you have to override a very well practiced response, which is to read the words, right, you've been reading for many, many years in our lives. So this is a habitual thing we do with words, very automatic effect, but in this task, we have to override this very habitual response in order to do a much less practice task that is named this this color Okay, so this sets up an interesting situation, it's meant to operationalize the sort of thing that you have to do when you have to ride on the left hand side of the road, for example. And what you find, of course, is that people are very fast and accurate when they have to respond to these colors, and that correspond to the words there's a quote, congruent trance are the word blue, printed in blue, but they're much slower and incur many more errors when they have to contend with incongruent words. So you know, the word blue, printed in green, and so forth. So, let's imagine you performing this task, okay. And the way that we envision says is that you hold in your working memory, these task rules. And in order to carry this out, let's say these are maintained in the frontal cortex, the frontal cortex has to bias how your visual brain is processing this information. Basically, your frontal cortex is trying to enhance color processing and suppress word processing, right, we're trying to not read these words, but we're trying to name the in color. And this, and this is how that process is thought to play out. Now, this is like imposing control, because you have to override this habitual behavior. What I'm really interested in, and many of my colleagues are, how do we sort of regulate that. So we have the ability to do this. But it turns out that we also do this more or less efficiently over time or in different contexts. So for example, if this can be done, unbeknownst to you, I have you do this task. And now I, you know, in one block of trials, most of these stimuli are congruent, so the easy ones, and only the incongruent, you will have, you know, you'll have an easier time on average, but whenever one of these difficult ones shows up, you will do really poorly, you will be sort of caught out. When I compare this with another block of trials, where most of the trials are the difficult ones, the incongruent ones, and only if you're congruent once, you're actually on average, do better. And the lead is, especially for these incongruent trials. So the idea here is that the brain adapts to the level of difficulty or demand of the task. So the level of attentional focus that you have to apply to do the task correctly, gets nudged up or down. In response to the task parameters. If I make it easier, your brain sort of relaxes the the effort that it spends on this device.
Nick Jikomes 42:35
If it's really difficult overall, you're sort of paying harder attention the whole time. And if there's other ones you sort of loosen up.
Tobias Egner 42:42
So essentially, you you adapt to the change in demand. And very interesting thing about this is that people do that without necessarily being aware of it. So you can sometimes you can do these manipulations and ask people afterwards. So do you even notice that there were any changes in these conditions, and they might not even have noticed it? Yet, their performance shows that they adapted to this so that they actually increased their focus on the task. During the times when it gets harder, and they're relaxed, they're focused on the task during the times when it was easier. So this would be an example that we would think of as a sort of at the intersection of learning and control, because the assumption is that it will your interaction with a task, you learn about how much do I have to focus here, and you nudge that up or down in relation to the stimuli that you experience, you can think of this, we can bring this back to the idea of anticipation or predicting inputs as well, because you can you can model this process with so called reinforcement learning algorithms, where the assumption is simply that your current focus of attention or the level of attention that you pay to the task is guided by sort of a recent average of, of the difficulty level that you experienced, based on the assumption that, hey, if it was just difficult in the recent past, it's probably going to be difficult in the near future. So you can generate a prediction and use that prediction to guide how much attention you will pay on the next trial. And of course, if then, you know, over time, suddenly you encounter a lot of these easy trials, then that will nudge that will slowly be nudged down, because that reason, average is now going down, down down. And when you get more difficult trials, again, that gets nudged back up, and we and many others have shown that this the way that people adapt to these changes in difficulty you can capture and simulate that quite nicely with the sort of very simple learning assumption you just adjust to the to the change in the merits.
Nick Jikomes 44:51
And you know, one thing that we mentioned earlier is the idea of task switching, um, you know, you know, very often in life Just in general, we're going around, we're entering into and out of new contexts all the time, we constantly have to sort of update, you know, the rules that we're applying to our behavior, you know, in the context of an experimental setup, like what you've just described in the Stroop test, you know, you could imagine, say, giving people a version of this, where the task is to, say, the color of the letters that the letters are printed in, rather than read the letter letters. But then as the trial goes on, perhaps it switches and then you have to read the letters and ignore the color of the printed and so on and so forth. When you have task switching, experiments like that, how do you measure sort of how difficult the brain finds it to to do task switching? And what are some key variables that influence how good people are at it? You know, for example, age, or whether or not they've got like, a personality disorder or a psychiatric issue. Let's talk a little bit more about tasks, which are, yes,
Tobias Egner 45:55
absolutely. So the the kind of control and of regulation of control that I just described to you can be thought of as sort of happening within a task. So regulating task focus, and complementary to that we have, of course, the ability to switch between tasks, those are usually sort of juxtaposed. And two terms that people like to use for this is that focusing on a current task involves cognitive stability, the notion is simply, hey, I'm keeping a stable goal going in and trying to I can block out distracting information to stabilize my task. But as you say, oftentimes in real life, even when we have engaged stably with a given task, and might be, say, reading a book in a cafe or something, I'm trying to drown out with my attention, that shatter from surrounding patrons. If my phone rings, I want to be able to switch away from reading the book and answer my phone, right. And we are able to do that. And this is what people like to call cognitive flexibility. And in the lab, we measure it with with task switching. So the kinds of tasks that you just described isn't perfect, something that people have done in the past quite a lot. But it doesn't have to be a stroop task. We often for example, we ask people, we show them real world objects, folders of objects on each trial, and might ask them, cue them before the trial, hey, categorize the object in terms of its size, you know, as its smaller or larger than a shoebox, or categorize it, whether it is a natural or a man made thing, you know, I show you a little bird, you say, oh, that small, and it's not so natural, and so forth. And switching between tasks, the way that you assess the actual cost of switching are the reason that we know that OS switching seems to be hard, it's a process that takes cognitive effort and time is that we can compare how long it takes you to respond to a trial, if it is a task repetition. So you just did a, you know, object size trial on the previous trial and dinner the current trial, again is an object size trial. So this is a task repetition. Versus or the current trial asks you to move to the task, we say this is a man made or natural object. So that's a task switch. If we compare the response time between these two trial types, we find that task switching is just reliably slower than task repetitions, and also reliably involves more errors. So sometimes, this updating process to switch to the new tasks that is, you know, imperfect doesn't always work sometimes accidentally still respond according to the old rules that you were just asked to do beforehand. So this is a switch costs are sort of our measure of flexibility. Now switching in you and me Of course, it's something that we do a lot. Some people are more flexible than others, but in the in the sort of like healthy participant population, those individual differences again, they're there, but not not massive, but switching and cognitive flexibility also, more generally is impaired in a variety of, of clinical conditions. That includes conditions like autism or schizophrenia and so forth. And of course, also can be affected by by brain damage this certainly especially to the frontal lobe, so that again, there is a rich neuropsychological literature, for example, that has shown that sort of so called Set shifting which is like task switching, can be very impaired frontal lobe patients are that this is particularly true when the task rule is not explicitly told to somebody but when you have to infer it based on feedback sent as a very famous and classic neuropsychological test called the Wisconsin card sorting test. Sort of a fun task, when each trial, I show you a sort of a playing card, and it has a bunch of shapes in it, they have different numbers of shapes, they can have different shapes and different colors. And the participant is asked to essentially sort this card into one of say, three or four piles. And you can sort this according to the number of shapes, the shapes themselves or the colors. So you could apply three different rules. And in that task, you don't know the rule as a participant, but you try things out, you test the hypothesis, and the experimenter will say, yes, correct or incorrect. And so then you keep sorting, according to say, the color rule. But the tricky experiment, at some point changes the rules. So say after you know, 10 Correct trials are sown in the category, cording to which you should sort these cards changes. And again, you are I will, after a single error feedback will be like, Oh, okay, it's the new rule, let's figure out which one of the other two it is. And we'll test that hypothesis and move on. But patients with frontal lobe damage, find this kind of shift in response to negative feedback, very hard to show and something called preservative perseverate of responding. So where they stick to the old rules, even though they're in repeatedly told that this is no longer valid, they can even sometimes express that verbally, they say, I know I shouldn't, but they're sort of can this disconnect kind of the will from the action, so to speak, that continue to sort under the first month, the wrong words. So that clearly is an impairment of being able to update your, your mindset to understand that this is the new context, and this is how I use it, which we would consider a lack of cognitive flexibility.
Nick Jikomes 52:04
Another thing that comes to mind here is, you know, we've got that, that phrase, you can't teach an old dog new tricks. We all have a notion that, you know, as we age, and we get older and older, cognitive flexibility certainly seems to degrade over time. So to what extent is that true? And are there any general patterns like specific ways in which flexibility tends to get worse, or at least change as we get older and older?
Tobias Egner 52:33
Yeah, interesting question. Sadly, flexibility and stability, like pretty much all cognitive processes do degenerate with aging. And I think the profile, the trajectory of development is probably similar for most of these functions in that you you're at your peak capacity probably in your early 20s or so. And thereafter, there is a, a slow and gradual decline. And these things are both, for example, in keeping out distract us, so the cognitive stability piece, and in the ability to switch tasks, we get a little worse with age. Again, I am not a developmental researcher. So there might be more data in this domain than I am aware of, but the studies that I am aware of just basically show that picture that essentially would expect, you know, sadly, things get worse as you age. And while there are other cognitive domains, where you have these interesting exceptions, where, for example, semantic knowledge, verbal vocabulary, and things like that can be retained, and sometimes even better and older than then younger participants, simple priming effects, they're quite stable over the lifetime. I don't think this is the case for these sort of higher level cognitive functions. So, they generally will, will get worse with age.
Nick Jikomes 54:07
And, you know, in terms of cognitive control and flexibility, generally speaking, which networks in the brain seem to be the most important for, for being able to have high performance here. And in particular, like, like, have you done interventions like, you know, transcranial magnetic stimulation where you're disrupting certain areas and you see performance degrade, what are those sorts of experiments look like?
Tobias Egner 54:32
Yes, a very good question we and others have done those. So, generally speaking, if you when put people in a scanner during something like the stroop task, so this is not you know, regulating on task focus, you find that and you let's say the most simple thing you can do is compare brain activity. When people perform these incongruent trials with it hard versus the easy to add, you very reliably find higher activity in the lateral frontal cortex, lateral parietal cortex. And in the medial frontal cortex, particularly a region called the anterior cingulate cortex, which is, for many years now has been a hotly debated region in in this area. So those are more active for trials, where you will have higher demand on task focus. You can also of course, play the same game I mentioned earlier, we can manipulate over time, the demands, so the proportion of these congruent or incongruent trials and see, hey, we're in the brain do brain regions activity follow this pattern, and as well, and these are the same regions that will show this pattern day we saw the activity was sort of track the level of difficulty over time quite nicely. So one thing that we have done is to fit one of those sort of reinforcement learning models to this kind of imaging data to identify a specific region of the, in this case, it was the left lateral prefrontal cortex that according to our fMRI data, we'd say, Oh, this region seems to be implementing these these predictions that you know, the updating of attentional focus for the next trial, but fMRI data correlation, right. So this is why we took this into the TMS lab transcranial magnetic stimulation, we asked, okay, if this region really does this, have proactive adjustments, meaning, we need to upregulate attention for the next trial or downregulated attention for the next trial. Then, if we messed with this region, just before the onset of the next stimulus, we should be able to remove these adjustment effects that we normally see in behavior. And this is in fact what we what we saw. So it just prior to each stimulus or not each one but so for half of the stimuli, the other ones with control, we briefly delivered a few pulses to this particular region that we identified in the fMRI study. And we found that this messing with this region, it's you can conceptualize that as injecting noise, so to speak in the processes that this region might be doing at the time. And doing that then essentially abolishes these adaptation effects are normally you would see a better attention of focus after an incongruent trial, because he just had a difficult one you upregulating attention. If you that that area, you don't see that effect anymore. And the same is true at the at the block level. So when you have a block with many difficult trials, where this means in many easy trials, that effect goes away if you mess with this particular with this particular area. So clearly supporting the idea that the lateral prefrontal cortex is crucial for implementing these attentional adjustments. Now, I don't know of a comparable study for tasks switching, generally speaking, and this is sort of intriguing, if you do a task switching experiments. And again, you do the simplest thing, you just compare brain activity on switch trials with tasks repeat throughout. So again, the harder ones with easier ones, if you will, but here the hard ones are defined differently when you switch tasks, rather than experience this incongruent stimulus, you find very similar brain regions activated. So again, lateral prefrontal and parietal cortex, the anterior cingulate, what are some sub cortical regions. So teasing out how what regions are doing in this case that might be different to what they're doing in the case of an incongruence tube trial is actually something that's yet to be done in great detail, like how exactly these regions contribute perhaps to somewhat different cognitive processes in both of these cases. For this, the switching there's one sort of a popular model of working memory and working memory updating that argues that in order to allow new information into working memory, you open a metaphorical gate to allow new information in and that gate is in the basal ganglia, these subcortical structures, the striatum and the basal ganglia. And the idea is that this gate is closed when you are focusing on a current task because you don't want to let distractors in. But it has to open up in order to allow a new task set to be to be entered. And by that would make the prediction that you should see very reliable activity in the subcortical regions during updating during switching, and you can find that but what we would really need again, would be more causal data so interfering, for example, with that region while people perform task switching studies, and that to my knowledge has not been done, you could, it's tricky because it's deep inside the brain, you can't use TMS for this because TMS only effects are fairly cortical superficial regions. But you could, for instance, try and recruit Parkinson's patient that have deep brain stimulation devices implanted, which is not so uncommon. And there you can sort of turn on and off their their striatum function, if you will. And people have done this for a bunch of purposes. But I have not seen that done. In a experiment that manipulated task switching and maybe the proportion of switches and stuff like that that would be really interesting thing to do for
Nick Jikomes 1:00:50
the transcranial magnetic stimulation in humans, how exactly does that work? Can you give us a visual of what the setup is, and then how the mediation itself is actually doing what it does?
Tobias Egner 1:01:00
Yes, so the principle is actually quite simple, you have a large sort of wire card, typically, the client takes the form of the figure eight mentioned her figure eight, the size of you know, your two hands, perhaps side by side. And those are sort of two wire coils, put in this particular shape. And you essentially turn a large electrical current on and off, that runs through these coils. Now, as you remember, from high school physics, when you have an electric field like this, perpendicular to this, you will create a magnetic field. And if you bring a conductor into this magnetic field, that will induce an electric current, and the brain or brain tissue is a pretty good conductor. So we can basically induce little electric currents into the brain by having this electric field generated close to the brain. And so the way that's been done is really with pulses. So you turn this on and off, on and off, on and off, and that produces these little electrical pulses inside, inside the brain. Now you have to get the, you know, the magnitude of the stimulation just right that this technique has been pioneered in the mid 80s, originally and just been found that oh, you can do this over motor cortex and make people's fingers Twitch, you know, you can basically activate with a single policy's upper motor neurons that that would then trigger these these muscle twitches. And then over the subsequent, you know, decade or two people have started figuring out, you know, we don't want to turn this on too high, you don't want to induce like an elliptic epileptic seizure, right. And they've also figured out that different patterns of pulses seem to be affecting the underlying neuronal populations differently. So it turns out, for example, that if you stimulate regularly at a very low frequency, like one hertz, for a while, that this seems to inhibit the underlying neuronal population for a while afterwards. Whereas if you stimulate at a higher frequency, like 10 hertz that can excite the underlying population for a while. And through this kind of sort of basic research of figuring out this stimulation parameters, people have, you know, started to use this to more and more in cognitive neuroscience, so to ask these kinds of research questions that we asking, but once you move outside of motor cortex, it's a little bit harder to know exactly what's happening because you don't have this direct readout anymore. We can just measure muscle activity changes as a function of of using these, these pulses. And but yeah, so different kinds of approaches have been developed using TMS and the one I mentioned earlier in our experiment where we messed with this particular brain region, from trial to trial just before stimulus onset, that's just one version. But another version is to use versus pattern stimulation for a while, offline before somebody does a task with the knowledge that oh, this particular regime I'm doing here will inhibit this brain region for the next 10 minutes or so. This is still you know, very much actively being figured out. What exactly is going on how exactly that works, and so on and so forth. So there's an active research area there in doing TMS in modern monitors and like in the macaque, for example, where you can then really look at underlying single neuron responses that will give you a better idea of what exactly is going on.
Nick Jikomes 1:04:53
When I think about cognitive control and flexibility throughout my day A and you know, every day is different. But in the abstract, they're all basically the same in the sense that right, I have multiple tasks I have to do throughout the day, I have a variety of goals that I want to complete, I have to make decisions about, you know, which order I do things in, when I'm at my computer, you know, do I have my email open? Do I have my slack open so that I might see notifications that pull me away from a task that I'm trying to do in the moment? Do I do one task at a time to completion? Do I do a little bit of one and then switch to another one, so on and so forth? You know, we all we all deal with this type of thing every day. So, you know, knowing everything that you know, and have learned about cognitive control and flexibility in these things? How have you how has that influenced how you sort of architect your work days in terms of sequencing and and thinking about how to optimize your workflow?
Tobias Egner 1:05:54
Yes, I wish it would influence me more than it does. But I think one pretty safe recommendation I can give is to do one thing at a time, as it were. So both because there is a cost to switching tasks, but we're also really not that good at it. So the idea that we can multitask is arguably sort of a bit of an illusion, what we really do, is continually switching between tasks, rather than actually doing several tasks in parallel. And because switches are costly, that's not very efficient. So it's a cost each time that you switch. But of course, you also have a cost. Because once you return back to the task you previously did, you really in real life, you have a real sort of restart costs as well, right? In this experiment I described to you right now you have these trials every few seconds, and you constantly switch between these two specific rules, there, you don't have that much of a restart cost. But in real life, if you've just been deeply engaged with writing your your essay or your email, and now you you flip away to, you know, answer phone call, or you switch more likely from paper writing to constant emails to checking your, your Twitter account. And so what that really will incur a big restart cost you getting back into your paper writing will, will really be much harder after you switched away. And the more you do that, the harder it will be to really get back into that that mindset to sort of deeply think about what you were just just jacking there. So in general, I would try and do one thing at a time, set yourself, you know, particular goal would say, Well, the next hour, I will be working on this paper, and I will not be checking my emails, and I will turn off my my phone and not check notifications and so forth. And that certainly will make for a much more efficient way of getting your work done then it is to have all those tabs open at the same time and keep checking and whether you get another message or a like on your social media Max.
Nick Jikomes 1:08:14
What? What's an interesting or big question in the field that you guys are pursuing right now?
Tobias Egner 1:08:25
Well, one big question that I'm particularly interested in right now is the interplay between the stability and flexibility, P piece that we talked about. So why is it that the same brain region seem to be involved in you know, adjusting that focus on the task, as well as in adjusting? how ready you are to switch to another task? Can we sort of parse out what they're doing? And how do people, you know, regulate these things? concurrently? So this is, I think, really interesting, especially at the level of brain mechanism. Another I think, really big topic that hasn't been tackled much. But that's that's really intriguing is something I mentioned to you earlier. And that is that a lot of these effects where we can see in behavior really nicely how people seem to adapt to changes in, in demand. And there's, there's other work, we and others have shown that people can also associate a particular, you know, control or attentional focus state with specific stimuli or cues and they retrieve them when they see that cue. And all of that stuff is often running off without your conscious knowledge of it. And this is kind of really curious, right? Because, by the very definition of cognitive control, we say well, you use these internal goals to guide how you, you know, deal with your surroundings, which Sounds very intentional, right? very deliberate. And it is in the sense that oh, I instruct you to do this task. And you're doing that task deliberately and intentionally. But it turns out that these little the mechanisms that allow you to do this thing efficiently and you know, that tune up or down just how much attention you actually need to pay to perform the task, these mechanisms that that look very clever, and they are clever, seem to be running sort of below our, our level of conscious perception. So we're not necessarily aware of our brain doing that. And I find that pretty intriguing. And there is not a lot of work on, you know, that really tries to dig into this, how come this is, is this really something that people don't know what it is? Does it matter if you notice or not? Why is that? You know, why doesn't that require a conscious awareness to adjust, you know, these very high level cognitive processes in response to changes in the environment. And so but, so that piece, I think, is a really interesting one that's also not been tackled much at all. So how much work goes on under the hood that we're not even aware of? When we just decide to do a particular task, you know, a lot of stuff then runs off by itself, even though it looks very deliberate.
Nick Jikomes 1:11:27
Well, this is fascinating stuff. And I look forward to seeing else what else comes out of your lab and in the field. Doctor to be as Egner thank you for your time.
Tobias Egner 1:11:37
Your very work, it was a great pleasure chatting with you and have a great day.
Comments