How We Learn — Stanislas Dehaene on Education and the Brain

AuthorAthenarium / Posted on

How We Learn — Stanislas Dehaene on Education and the Brain

From roundworms to humans

Why do humans need to learn? Why are we not genetically programmed from birth with the inbuilt knowledge we need to survive in the Darwinian jungle? This is a simple yet profound question. One that speaks directly to our long heritage and natural history. And as the neuroscientist Stanislas Dehaene explains in How We Learn, the answer pertains in part to adaptability and efficiency.

True, evolution via natural selection is a reliable and powerful algorithm for producing well-adapted creatures. But it is also a slow and toilsome process, for genes are rather rigid instructors. Many creatures have to be snuffed over geologic time before new innovations come about. What’s more, the environment in which animals live is full of novelties and surprises. Writing genetic instructions for every permutation would be an impossible feat. There just wouldn’t be enough space.

Genetic rules and programs for learning, however, may endow living creatures with an informationally practical mechanism for navigating their open-ended world. “Even simple organisms devoid of any cortex, such as earthworms, fruit flies, and sea cucumbers, learn many of their behaviors”, Dehaene notes. He points, for example, to nematodes—a species of roundworms that rely on habituation and association to tailor their behavior to the chemistry and temperature of aquatic environments.

“Darwinian selection is, in effect, a learning algorithm—an incredibly powerful program that has been running for hundreds of millions of years, in parallel, across billions of learning machines (every creature that ever lived). We are the heirs of an unfathomable wisdom.”

Stanislas Dehaene. (2018). How We Learn.

Interior models and Bayesian brains

But we’ve been somewhat coy. What exactly does it mean to learn? We might say in the first instance that learning refers to the acquisition of some experience or knowhow. But at its most distilled, Dehaene argues that “to learn is to form an internal model of the external world.”

Where roundworms are reacting only to a few variables in a roundworm world, the human brain in a human world contains thousands of interlocking and interacting internal models, which range from facial recognition systems to the structures of language.

Of course, any learning algorithm or internal model must begin with some rules and assumptions about the problem. Dehaene says the construct is similar to the concepts in Bayesian probability, where prior hypotheses are revised through education, experience, and other accumulated inferences.

In this way, learning pertains to the manner in which we fine-tune the settings and rules that characterize our interior models. The baby who listens to the babbling of her parents, for instance, is formulating the parameters of her language program to the sounds, constructions and associations of her native tongue.

Reward functions and error minimization

Learning is not a blind adjustment, mind you. While there is trial and error as we explore the combinatorial explosion of our domains, the process also involves a restricting of our search spaces—a process that is driven in part by the reward functions and error correction procedures that prescribe our models.

I think this is well expressed, for instance, in the game of chess. Through wins, losses, practice, and study, young players will incorporate and discard the respective principles that do and do not work. This exploration and winnowing repeats and iterates itself as the player refines her parameters for strategizing, calculation and evaluation. What results, after decades of experience, is a sophisticated internal model for chess.

For another example, Dehaene points to shared attention rules in language acquisition. A baby quickly realizes, for instance, that when her mom speaks to her, it is usually in reference to some object in her visual space. Hence, the baby needs only to follow her mom’s gaze or finger to narrow the possibilities of intent, and to find association and meaning in the sounds and scenes she experiences.

Symphonies and infinities

Of course, there is a lot about the brain that we do not yet understand. Dehaene notes, for instance, that “there is no truly satisfactory model of how synaptic changes in neural networks underlie language acquisition or mathematical rules.” 

I myself cannot help but share Dehaene’s fascination for these small, squishy contraptions. As we speak, the millions of neurons encased inside our skulls are busy at work, inputting and outputting signals to one another. Somehow, in this hierarchical, self-organizing symphony of interactions, we are able to conceive of concepts as bewildering and unending as infinity. What’s more, these abstractions, once made, can be generalized to disparate domains and shared with other fellow brains.

A statistical, lingual genius

Eons of selection pressures have endowed brains with the curiosity and circuitries for generating and selecting abstract rules. Babies in particular will pick up many such formulas in their early life. They quickly learn, for instance, that most objects “occupy space [and] do not vanish without reason”, Dehaene writes. This is well observed when “babies act surprised in certain experimental situations that [appear to] violate the laws of physics.”

A lot is made as well of a child’s instinct for language, as popularized by the cognitive psychologist Steven Pinker. But “what is hardwired in them is not so much language itself as the ability to acquire it”, Dehaene writes. And when you stop to think about it, the child’s brain is something of a statistical lingual genius. From all the adult jibber-jabber they hear, babies and children are somehow able to sift and sort the sounds to arrive at a language system that they can share with their parents and peers.

Neuronal hurricanes

We do know that many of the neural architectures responsible for these feats are constructed during the final trimester of pregnancy. Chemical messages guide the growth of excitatory neurons as “the brain self-organizes into a network of crisscrossed connections”, Dehaene explains. He likens a baby’s cortex at birth to “a forest after a hurricane”. During the first six months, what begins as a sporadic scattering of neural trees turns quickly into an “inextricable jungle” of connections.

The synapses, which enable neurons to pass signals to one another, are made initially in excess. The circuits that our brain then keeps and prunes depend on the learning, experiences, and feedback it receives. Of course, the brain is not a perfect recorder. It has to be selective with what it does and does not change. As Dehaene explains, our “synaptic plasticity is modulated by vast networks of neurotransmitters, [like] dopamine and serotonin, that signal which episodes are important enough to remember.”

Similarly, procedural memory functions because the relevant neurons and circuits self-modify themselves to enable more efficient information flows where repetition is concerned. This is why “learning to play music, read, [or] juggle results in detectable improvements in the thickness of the cortex and the strength of the connections that link cortical regions”, Dehaene writes. “The highways of the brain improve the more we use them.”

“Such is the new vision of the brain: an immense generative model, massively structured and capable of producing myriad hypothetical rules and structures—but which gradually restricts itself to those that fit with reality.”

Stanislas Dehaene. (2018). How We Learn.
Figure 1. The neuronal trees of a child’s brain
“In the first two years of life, neuronal trees grow wildly until they form inextricable bushes… In the course of development, dendritic trees are progressively trimmed under the influence of neuronal activity. Useful synapses are preserved and multiplied, while unnecessary ones are eliminated.” Source: Stanislas Dehaene. (2018). How We Learn.

Congregations and societies

It can help, Dehaene adds, to think of the brain as a congregation of inner experts, critics and actors that are forever collaborating and competing on various activities. Our systems for attention, for example, are frequently engaged in “biased competition” as sensory inputs from our noisy environment fight for our brain’s finite resources. 

In fact, his analogy applies just as well to human society. After all, what is human society but an agglomeration of brains in ceaseless cooperation and competition with one another. We must wonder then if our personal and collective approach to education is appropriate for the times. And if it is not, we ought to hurry with corrective action, for the brains of children are among the most precious, miraculous, and malleable of things.

Forgetting to learn

Indeed, the brain is literally and metaphorically shaped by learning. Unfortunately, as most of us know today, the sensitive period for synaptic plasticity is limited. The ease of learning declines gradually with age. Dehaene notes, for instance, that “sensory areas reach their peak plasticity around the age of one or two years old, while higher-order regions such as the prefrontal cortex peak later in childhood.” While the brain can exhibit plasticity at an advanced age, societies would do well not to squander their children’s learning potential during these crucial years.

Of course, most of us will intuitively understand the dangers of delayed learning, the value of a general education, and the elements that make for a conducive environment. Few would disagree likewise with the importance of nurturing healthy motivations, focus and attention, active engagement, timely feedback, a sense of wonder and curiosity, good nutrition and rest, and a stimulating classroom. 

So it is ironic that aspects of our education system continue to inhibit the learning of children and young adults. The college megafactories of recent times, I think, are especially guilty of this. Many high-paid professors today continue to deliver droning monologues as scores of sleep-deprived students nest and nap in the back rows. What we have instead is a learning culture that incentivises kids to cram and forget. Indeed, why should they bother when the final grades they receive tell them little about the nuances of what they do and do not understand. 

The greatest difficulty

Indeed, learning and the sharing of learning is the cultural ratchet that enables society to retain and build upon good ideas. It is why we have survived for so long. It is the very “triumph of our species”, Dehaene writes. Conversely, a failure to learn will be the source of our undoing and degeneration. When education and culture closes one too many minds, we risk shuttering an entire generation.

I stress again that humanity would do well to preserve and nurture its inner childhood—for in the black box of every child’s mind is an immense and latent potential that our education and cultural institutions are yet to fully realize at scale. As the philosopher Michel de Montaigne wrote more than four hundred years ago: “the greatest and most important difficulty of human science is the nurture and education of children.” 

Sources and further reading