What are children learning in their long childhood? How should they learn?

Cognitive control is not a mental function, but rather emerges from a highly interactive and complex system that links thought and action. To fully understand cognitive control, it must be viewed as a dynamic function rather than a static ability.

This article is excerpted from "Cognitive Control" (May 2022 edition, International Culture Publishing Company), with some deletions.

Written by David Budd

Translation | Fang Qinghua

If you have children yourself, or at least spend a lot of time around them, you probably have some idea of ​​where the concept of “children’s cognitive control disorders” comes from. Cameron might keep repeating the same word for 20 minutes straight, Joan might decide to let everyone in the restaurant know that she just went to the bathroom, and Ellroy might not realize that he’s been wearing two different shoes all day and his shirt is inside out.

These moments make us laugh to ourselves, but as comedian Ray Romano mentioned in his comedy routine, "Grandpa is like this, but not as cute as a kid. Right? That's your double standard." Ray is absolutely right. Neuroscientists have long observed that adults with frontal lobe damage behave very much like children, which is a major obstacle in their lives.

A case that can be used for reference appeared in 1936, that is, patient A behaved very childishly. For example, the following is a description of patient A getting dressed in front of his doctor, wife, and mother:

(Patient A) is washing his hands. "Why should I wash my face? The barber will wash it for me. He puts a hot towel on my face and that's enough." (It is pointed out that soap cleans better than a hot towel.) A replies, "Nonsense!" ... A wanders from room to room, teasing his mother and mocking his wife. He whistles, sings, grins, dances, and claims, "I'm a bit of a dancer, too. I bet you can't dance." (When no one pays attention to him, A starts to make fighting gestures and pushes L with his fists.) A often mentions the Stock Exchange and the trading floor. A puts on his shirt and then his trousers, starting with his right leg and buttoning them only partially. He puts on his shoes but does not tie them. Then he stands up, holding his slippers in his hand, when his mother appears to take them from him.

I imagine my wife and I have experienced the exact same mornings as we try to get our kids ready for school, except that our kids may have substituted Minecraft for the New York Stock Exchange. These behavioral similarities suggest a common system underlying the behavior of these two groups. In other words, these commonalities have led to the hypothesis that cognitive control is the primary extension site of developmental change in children. But it’s just an analogy.

Just as cognitive control develops slowly, the brain systems that are critical to cognitive control also undergo a long process of change from infancy to adolescence. However, this is not to say that the frontal cortex is waiting to develop or becomes active only in late childhood. Instead, its development begins in utero, and by the time we are born, the brain already has the regions and neural network subdivisions of the prefrontal cortex.

"Cell migration" refers to the process by which young cells move around the embryo and place themselves in the appropriate position in the developing organism. During the fetal period, cell migration in neurons is important for the development of the nervous system because it determines how neurons are positioned, grouped, and connected to each other. Notably, neuronal cells migrate primarily from front to back within the frontal lobe, so that cells in the rostral prefrontal cortex differentiate earlier than cells in the caudal prefrontal cortex.

In contrast, the thalamus is connected to the prefrontal cortex in a posterior-to-anterior fashion. As you may recall, thalamocortical drive is regulated by the striatum and supports working memory gating. Thus, this caudal-to-rostral pattern of cortico-thalamic connectivity may be the source of asymmetries in gating circuits supporting hierarchical cognitive control, as discussed in Chapter 4.

Furthermore, the pattern of thalamic innervation in the frontal cortex from caudal to rostral is in stark contrast to the pattern of cell maturation from rostral to caudal, which may have important organizational consequences. The thalamus, as a brain structure, is actually a large central station. Inputs from outside the brain must first pass through the thalamus before they can contact the brain. The thalamus is the main stop for sensory information from the posterior areas of the neocortex to the frontal lobe. Since there is no initial input from the thalamus, the rostral prefrontal area will mature in the absence of sensory input from the posterior part of the brain. Therefore, the early differentiation of rostral prefrontal neurons depends mainly on their own local frontal input. This local integrated processing is another feature of the prefrontal hierarchical control structure discussed in Chapter 4. Surprisingly, these features are already present when we are born.

Other changes in the brain continue after birth and extend to the prefrontal cortex, a region of the brain that is thought to have evolved over time. Although the entire brain increases in size early in life, the prefrontal cortex grows twice as fast as other regions, a fact that reflects the evolutionary expansion of the human forebrain.

The prefrontal cortex is also one of the last areas of the brain to mature. As all cortical areas mature, cortical thickness increases and then declines to a stable level in adulthood. However, the timeline of this process varies between different brain regions, and progress along this timeline can be used to measure the maturity of a particular cortical area. Studies using these measures consistently find that while most basic sensory and motor cortical areas reach a stable stage of maturation around 3 to 6 years of age, the prefrontal cortex continues to mature during adolescence and does not stop until the early 20s. The figure below shows this timeline. White matter measurements show that the maturation timeline for the frontal regions is very slow relative to other areas of the brain.

Figure 1. Grey matter maturation on the cortical surface. (A) Top view, adapted from Shaw et al. (2008); (B) Side view, redrawn from Gogtay et al. (2004). The color scale corresponds to the grey matter volume, and the thinner the grey matter, the more mature the development.

One reason the cortex changes thickness during maturation is that the density of neurons communicating with each other through synapses changes. Throughout the brain, new synapses form after we are born, a process developmental biologists call synaptogenesis. Synaptic development begins with an initial increase in number, during which the number of synapses increases dramatically, followed by a period of synaptic pruning, during which many synapses disappear. This pruning process is essential for efficient neural network processing. Synapses that are not used disappear, while groups of neurons that fire together develop stronger synaptic connections. This use-dependent change in the cortex is critical because it is the first clue we have that brain development is not immutable but is driven by use. What a person experiences in life determines how the brain is used. This process of synaptic change is protracted in the prefrontal cortex, peaking later and taking longer than in other brain regions.

What drives the development and variation of cognitive control? This is a fundamental question, the answer to which may tell us why individuals differ in their cognitive control abilities, and how we can intervene to ensure the healthy development of the brain and cognitive control. As you might expect, given its importance and complexity, this is also a controversial question in the scientific community.

Cognition and brain function are determined by both genes and the environment, and most importantly, by the interaction between them. Environmental factors include the biological environment, such as hormones and molecules that we are exposed to all the way back to when we were in the womb. Environmental factors also include the influence of information processing through our senses. Our experiences influence the development of most cognitive functions, and cognitive control is no exception.

To understand the influence of environment and genetics on cognitive control, we first need to explore how scientists measure differences in cognitive control between people. I probably don't need to tell you that people's cognitive control varies a lot. I'm a typical professor who is always forgetful. For example, I recently went abroad for work and lost all of my international power adapters. I lost more than one, and more than once. In fact, I just texted my wife in frustration about how I lost one adapter at the hotel and lost another at the airport. It's not easy to be so sloppy!

So cognitive control varies from person to person. Yet scientists trying to measure these differences face a challenging problem. We want to know how various abstract mental abilities, such as inhibition, vary from person to person. But we have no way to measure them directly. We can test them with tasks, such as the stop-signal test, which are designed to tap into these abilities. But the tasks we do in the lab are not very pure. When people perform a given task, multiple cognitive and brain systems may be involved, interacting with each other in complex ways to produce the behavior we observe.

For example, the stop-signal task measures inhibition, but it also involves vision and hearing, spatial attention, motor preparation, language, memory, etc. We account for these influences by controlling for them, but even these controls are not pure, and we have to make assumptions about how to separate inhibition from all the other factors listed.

To address this problem, scientists hypothesize that while no task is pure, multiple tasks performed simultaneously can be impure in different ways. So, instead of just using the stop-signal task, we test multiple tasks that share the same hypothesized inhibitory component. We can then look at the similarities in how subjects perform during these tasks. For example, people who are particularly good at inhibition tend to be good at any task that has an inhibitory component relative to tasks that have other components. Of course, the main limitation of this procedure is that we assume we know which tasks involve inhibition or whatever process we're interested in measuring, which is not a simple assumption. Still, this approach yields some general and consistent patterns in the variability of human cognitive control functions.

Akira Miyake and Naomi Friedman of the University of Colorado conducted a landmark study of individual differences in cognitive control. They hypothesized a distinction between three constructs of cognitive control ability: inhibition, updating, and working memory. Roughly speaking, they envisioned “inhibition” essentially corresponding to what we called termination inhibition in Chapter 6, while “updating” and “working memory” approximated the two dimensions of flexibility and stability of working memory gating that we framed in this book. Each of the three constructs was tested multiple times. For example, inhibition was tested by three tests, such as the stop signal test, the Stroop task, and the action-no action test.

The results were both compelling and contradictory. First, people’s performance could be explained in part by different architectures of inhibition or updating. In other words, people’s performance on a given inhibition test was more related to their performance on other inhibition tests than their performance on a working memory test, for example. So, as we would expect, different aspects of cognitive control determine very different performance patterns.

Importantly, however, although the tasks performed exercised specific control functions, there was a general component that predicted performance across all tasks. Thus, if a person excelled at one task involving cognitive control, then it would be expected that they would also perform well on other tests of cognitive control to some extent.

Miyake and Friedman called this paradoxical set of findings the “unity and diversity of executive function.” In other words, control functions do not operate as separate organs, like the heart and liver, but rather as inseparable units. It is likely that there are some common aspects of brain function that affect all cognitive control performance, and there are also some systems or factors that favor certain types of control performance.

Given this complexity, we can look at how genes and environment influence the development of cognitive control. Studies of twins provide the most powerful investigations of the influence of genetic and environmental factors on the development of common and specific cognitive control architectures.

Twin studies include identical twins, who share 100% of their genes, and fraternal twins, who share 50% of their genes. By comparing twins, we can estimate three factors that influence performance. The first is the effect of genes, with identical twins estimated to be more similar to each other than the similarities observed in fraternal twins. The second is the effect of the twins' shared environment, which refers to the similarities between the twins without taking into account their genetic similarity. Finally, there is the effect of the non-shared environment, which is assessed by the differences between identical twins who are exposed to the same environment.

Astute readers may notice that this calculation leaves out the interaction between co-living and non-co-living environment and genes. It is difficult to assess this interaction without a large sample of twins raised together or apart. Even if we had such a sample, the groups were not truly randomly assigned. Given what we know about genetic and epigenetic effects, this interaction may be an important factor in individual differences, so its absence leads to a significant limitation on the conclusions drawn from human behavioral genetics studies. However, despite this limitation, the evidence obtained from twin studies is still important.

Twin studies of cognitive control have found that common genes explain nearly all individual differences in a common cognitive control component, and this finding holds true for all cognitive control test performances. Multiple studies of children and adolescents, and accounting for socioeconomic status, education, race, and other demographics, have found this component to be about 99% heritable. Common and unique living environments have little effect. Furthermore, unlike other putative traits, such as general intelligence, whose heritability increases with age until adulthood, the common cognitive control factor appears to be highly heritable and equally heritable in children, adolescents, and adults.

Importantly, it would be a mistake to interpret this high heritability as indicating that the environment has no effect on the development of cognitive control. First, this high heritability applies only to the common cognitive control component. As we will discuss later, environmental factors may have a large influence on more important specific control components. Second, although some studies used diverse samples, most individuals in these studies were still subject to a limited range of environmental influences. Genetic studies of IQ have consistently found that heritability increases with socioeconomic status, so similar phenomena may affect cognitive control results. Third, as mentioned earlier, the present analysis did not assess the key interaction between genes and environment. Fourth, it is unclear whether and how abnormal environmental factors such as extreme neglect, abuse, and malnutrition affect the common control component. Nonetheless, these observations do suggest that part of our common performance on cognitive control tests will be based on stable individual differences in our genes.

It is not entirely clear what this common cognitive control capacity corresponds to biologically. However, a recent genome-wide analysis of 427,037 people in the UK Biobank identified 299 loci associated with the architecture of an estimated common cognitive control capacity. Broadly speaking, these loci are associated with biological features of the brain, related to the formation of fast synaptic pathways and the prevalence of the neurotransmitter GABA (gamma-aminobutyric acid). How or why features like fast neural dynamics or GABA are so broadly important for cognitive control remains unknown. These factors are too general to change during tasks that impose inhibition or switching demands, and are difficult to study. Therefore, they are unlikely to explain the diversity of individual difference paradoxes.

While shared cognitive control components may be highly heritable, this is clearly not the case for more specific cognitive control architectures or performance on individual tasks faced in daily life. For example, in a study of 7- to 12-year-old twins, nonshared living environment was the main determinant of performance on a stop-signal task, followed by shared living environment. In terms of order of effect, genetics came last. Part of the reason for reduced heritability in these specific tasks is the task purity issue we discussed above. However, task imprecision is not the only reason. Individual control architectures such as updating or inhibition that arise from multitasking also have low heritability. In the study mentioned above, nonshared living environment between the twins was the main contributor to inhibition, updating, and working memory.

Given the importance of learning in building control systems, this environmental influence on individual control functions makes sense. Remember the example of working memory gating? In Chapter 3, we discussed how acquiring a gating strategy that is appropriate for a given task is key to performing that task perfectly. We need to learn not only the rules of the game, but also how to execute these rules through working memory based on the relationship between inputs and outputs. Gating plays a crucial role in managing multi-level goals as hierarchies become more complex, and the correct gating strategy can be learned from experience.

In collaborative work, Dima Amso, postdoc Kerstin Unger, and I found that 7-year-olds seemed to be more likely than 10- to 12-year-olds to choose the wrong gating strategy in the first/last task discussed in Chapter 3. This failure to choose the right gating strategy was partly responsible for their poorer performance than the older children. So it may not be that children are always out of control. Instead, they just don’t find the right way to break down the task into smaller pieces and control the input and output of working memory in order to perform the task efficiently.

Thus, during the long and critical middle childhood, children may be constantly learning what to control and when to control it. They are busy developing increasingly abstract gating strategies that apply to more and more situations, adapting themselves to increasingly complex tasks. They are also learning how to exercise internal control. Of course, while doing this, they are also constrained by perceptual, conceptual, linguistic, motor, and other systems.

This view of cognitive control development therefore places great emphasis on learning and experience, and especially on the diverse experiences we have during childhood. In order to build useful, abstract gating strategies that are applicable to many situations later in life, we need to try to control ourselves in many different contexts.

Computer models of cognitive control in neural networks exhibit this fundamental property. Take, for example, the gated cortico-striatal model discussed in Chapters 3 and 4. Many training runs allow the model to learn which inputs to gate into working memory and when to gate out of it, based on dopamine prediction errors. Similarly, presenting these models with multiple different tasks allows them to generalize and create abstract representations of situations that are not only useful in a specific task but are reusable components of tasks. Thus, these models provide an existence proof for our hypothesis, showing us that building a gating system for cognitive control requires learning and diverse experience to ensure that it is gated correctly.

In the real world, this view fits with data showing that environmental enrichment is key in the development of cognitive control systems that work well in a wide range of new contexts. Enrichment builds on children’s diverse experiences and learning environments. Enrichment has long been associated with active learning, including cognitive control. Therefore, one explanation for these observations is that an enriched environment allows children to develop abstract gating strategies that are broadly applicable to new contexts. Enrichment is of great benefit to children later in life, as most children discover that the adult world is very different from the world of their childhood. They already have a large repertoire of gating strategies that they can use to synthesize solutions to a wide range of problems, depending on the goals they face.

Another striking thing is that the necessity for diverse learning provides an explanation for the long process of developing cognitive control. It is necessary to collect as much data as possible from experience, which will optimize the control system to adapt to the world we live in. In essence, the brain assumes that the first 15 years of life are a "model" for how life will be for the next 65 years, and the brain optimizes control on this basis. This means that the effectiveness of cognitive control will only be as good as the validity of this assumption, that is, the quality of the model the brain builds. Like any statistical model, it is likely to produce better results if it is given a lot of data - a useful sample of the kinds of demands that will be encountered later in life.

This emphasis on using cognitive control to learn and experience warns against a growing trend toward interventionist parenting. The “helicopter parenting” of the early 21st century has evolved into today’s “lawnmower parenting.” Lawnmower parenting refers to parents trying to remove any obstacles in their children’s path, both at school and at home. This extreme parenting style deprives children of the opportunity to choose their own path to success or failure.

There are many reasons for this trend, most of which are justifiable. Parents’ concerns for their children’s safety, their love for their children, and their desire for their children to express confidence and succeed are important driving forces, in addition to social pressures. Parents are well aware that their children need to succeed in an increasingly competitive learning environment. It is impossible for children to remember to pack their bags for school or do their homework. For many parents, even if they want to encourage their children to be more independent, it is difficult to do so in today's world. Even if children are allowed to go to the park by themselves, who can they play with?

Of course, children are given a great deal of autonomy. Most children of previous generations are unfamiliar with this kind of upbringing; they have been able to play freely, to do whatever they want, to wander aimlessly through parks, streets, and woods. Children of different ages can form groups and come up with their own goals, games, rules, and solutions to problems. Many of these ideas are undoubtedly bad, even terrible, and inevitably lead to some failure. But when there is no danger, failure is also a wonderful way to learn, especially for developing cognitive control.

Giving children the autonomy to succeed and fail is a learning principle that good coaches of children’s soccer teams already understand. As beginners, children do not position themselves effectively on the field and create spaces. They wait too long for the ball to be passed, and the ball does not arrive when they should have. In these situations, parents try to tell their children where to run, when to kick, who to kick to, and so on. However, unlike frustrated parents, good coaches wait for their children to make their own decisions—or, if they fail to make a decision—and then praise or correct them accordingly. They point out what the child should do in that situation, but importantly, they give the child a chance to make his or her own decision first. Coaches do this because if children are constantly told what to do, they will never learn to interpret for themselves how to take the right action in their situation. They will never develop the right control strategy—the strategy that interprets the dynamic system of the sport and selects the appropriate action. They only learn one control strategy: listen to the coach.

Life is not a soccer field. So how can we, as parents, create a safe and effective learning environment for our children? One approach that seems promising: carve out time between piano lessons, gymnastics, and homework for unstructured activities. In fact, preliminary research on unstructured or semistructured learning supports its benefits for developing cognitive control. In these studies, unstructured play refers to play in which children can decide on their own goals and tasks, make plans and organize activities, and find solutions to their own problems. Therefore, having these opportunities may be particularly important for children to learn self-directed cognitive control, as opposed to following direct instructions from others.

So, if children are given opportunities to face new problems, struggle with them, fail, and solve new problems, especially when they do these things independently, their brains also get the opportunity to develop abstract and effective control strategies. Child caregivers should resist the temptation to "take over" and seek out opportunities for children to be autonomous and successful, and to experience real failures while ensuring safety. With these experiences, control systems can be optimized and prepared to adapt to a range of new situations in later life. This is particularly important because, as we will see in the next section, control systems become a major source of challenges and support for us as we age.

About the Author

David Badre is a professor of cognitive science, linguistics, and psychology at Brown University and a member of the Carney Institute for Brain Science. He has made groundbreaking contributions to the neuroscience of cognitive control and executive function.

↓↓Go to the "Fanpu" public account and click "Read original text" in the lower left corner to purchase↓↓

Special Tips

1. Go to the "Featured Column" at the bottom of the menu of the "Fanpu" WeChat public account to read a series of popular science articles on different topics.

2. Fanpu provides a function to search articles by month. Follow the official account and reply with the four-digit year + month, such as "1903", to get the article index for March 2019, and so on.

Copyright statement: Personal forwarding is welcome. Any form of media or organization is not allowed to reprint or excerpt without authorization. For reprint authorization, please contact the backstage of the "Fanpu" WeChat public account.