How We Learn
Based on Stanislas Dehaene's book How We Learn (2020), chapters 1, 7, 8, 9 and 10. About 20 minutes to read, all paraphrased in plain, advanced English. You do not need the book itself.
The big idea
Learning is the brain building an internal model of the world, testing it against reality, and correcting it. Dehaene argues that four things decide how fast and how well we learn.
1 Attention
The brain only learns what it amplifies.
2 Active engagement
A passive brain learns almost nothing.
3 Error feedback
Surprise and mistakes are the fuel of learning.
4 Consolidation
Practice and sleep turn effort into automatic skill.
Bring one example from your own life for each pillar: a language, an instrument, a sport, a job skill. We will use them in the session.
Seven definitions of learning
To learn is to form an internal model of the outside world. You already hold thousands of models: a map of your neighbourhood, a feel for which words sound English, the knowledge of how to ride a bike. Even what you see is a model, because your brain decides
that a tilted oval plate is really round.
1 Adjusting parameters
Wear prism glasses and your hand misses the target; after a few tries your brain recalibrates. A baby does the same with language, setting its sounds and word order to match the language it hears.
2 A combinatorial explosion
A few adjustable settings generate a huge space of possibilities. That is how we master tens of thousands of words in about a decade.
3 Minimising errors
Like a hunter adjusting a rifle scope after each shot, the brain uses the size and direction of each error to improve the next attempt.
4 Exploring possibilities
Pure error-correction can get stuck in a local minimum
, like the cheapest stall in the first aisle of the market. Some randomness, curiosity and play help us find better solutions.
5 Optimising a reward
Often nobody tells us the right answer, only whether we succeeded. An inner critic
predicts success and an actor
adapts.
6 Restricting the search
Too much freedom leads to memorising details (overfitting
) instead of grasping a general rule. Simpler models generalise better.
7 Projecting hypotheses
There is no learning from a blank slate. Evolution gave us built-in assumptions, such as the idea that objects persist. In Changeux's phrase, to learn is to eliminate
.
Human learning is not passive recording. It is hypothesis-making plus correction, and that sets up the four pillars.
Attention
The brain is bombarded with information, so it filters ruthlessly. Attention is the set of mechanisms that select, amplify and deepen the processing of what matters. What is attended is boosted many times over in memory; what is ignored barely leaves a trace.
Dehaene, following Michael Posner, describes three attention systems:
Alerting when
Surprise, emotion or passion floods the cortex with chemicals such as dopamine and acetylcholine, a kind of now print
signal. Action video games and gripping teachers can trigger it. Dehaene rejects the claim that today's children have lost the ability to concentrate.
Orienting what
A spotlight that also suppresses everything outside the beam. In the invisible-gorilla experiment we look straight at something and do not see it, and we are unaware of our own blindness. Teachers fall into the same trap and assume that what they see, students see.
Executive control how
The brain's switchboard and the basis of self-control. It is slow and serial, so true multitasking is an illusion: a second task is delayed, often without our noticing. It matures into the twenties and can be trained through music, Montessori-style tasks and some games.
You cannot learn what you do not attend to, and you cannot attend to two things at once.
Adults learning a new script who focused on whole word shapes failed to read new words. Those who focused on individual letters succeeded and used the normal reading circuit. This is why Dehaene favours phonics.
Humans learn by sharing attention: we look where the teacher looks. Babies learn a new word or a foreign sound from a live person, but not from a loudspeaker or a video. The flip side is that we also absorb false beliefs and conform too easily, so we need a critical scientist
mode as well.
Active engagement
In the classic carousel
experiment, two kittens received the same visual input, but only the one that moved freely developed normal vision. A passive organism learns little or nothing. Efficient learning means generating hypotheses and testing them.
Mentally active, not busy
Dehaene was dismayed by a school that put pedals under pupils' desks. What counts is rephrasing ideas in your own words, predicting and questioning.
Deeper processing, better memory
People who judged whether words were animals (meaning) remembered far more than those who judged capital letters or rhymes. A large review found that active methods beat passive lecturing in university science courses.
Active does not mean left alone. Pure discovery learning works poorly. The best results come from structured guidance: demonstrate, then let learners try.
The digital native
who multitasks brilliantly, and learning styles (visual, auditory and so on). People differ in speed, background knowledge and motivation, not in a special preferred channel.
Finding out something new activates the dopamine reward circuit, even in anticipation. It peaks for things neither too easy nor too hard, the Goldilocks effect
. It also needs metacognition: you must know what you do not know.
Three ways school can kill curiosity
1. Work that is too easy or too hard. 2. Punishing questions and mistakes. 3. Purely receptive teaching, where the teacher shows everything. In one study, children shown a toy's function explored it much less than children given no demonstration. Teachers who admit they do not know everything keep curiosity alive.
Error feedback
The mathematician Grothendieck, as a boy, was convinced that pi equals three, because six compass-steps fit round a circle. He felt not shame but the thrill of discovering a mistake. Errors are the natural route to learning.
According to the Rescorla–Wagner theory, the brain predicts what will happen and learns only from the gap between prediction and reality. An animal that already knows a light predicts food learns nothing from a bell added later: no surprise, no learning. You do not even need to be wrong. A correct guess you were unsure about still reduces your uncertainty.
Feedback is not punishment. The best feedback is quick, accurate and informative: it says where you went wrong and what to do instead.
Fear blocks learning
Stress, such as maths anxiety, gets in the way. Grades are a poor substitute for feedback: imprecise, late, and often a stigma. Video games start easy, allow retries and show progress.
Growth or fixed mindset
Carol Dweck's growth mindset (I can improve
) beats a fixed mindset (I'm just bad at this
), although the size of the effect is debated.
Test yourself
Retrieval practice is among the most effective techniques known. Trying to recall an answer before checking it beats re-reading and highlighting. Short-term memory makes material feel familiar, so we underestimate this.
Space it out
Spreading study over several sessions can triple memory compared with cramming. Rule of thumb: review after about 20% of the time you want to remember it, then at increasing intervals.
Consolidation
A beginner reader deciphers a word letter by letter, so longer words take longer. After two or three years the word-length effect disappears: the brain reads all the letters at once. This shift from slow, conscious and effortful to fast, unconscious and automatic is consolidation.
Effortful tasks occupy the brain's executive bottleneck, which cannot multitask. Automating basic skills such as reading, arithmetic, typing and driving frees mental resources for higher-level thinking.
Sleep is part of the learning algorithm
Overnight gains
Material learned just before sleep is forgotten less, and performance can improve overnight without extra practice.
Replay
In rats, hippocampal neurons replay the day's routes during sleep at about twenty times the speed. Humans show similar reactivation, for example after hours of Tetris.
Insight
Sleep moves knowledge from episodic memory towards general knowledge in the cortex. In one experiment it doubled the number of people who found a hidden shortcut.
Cues, not recordings
A smell or sound linked to learning and replayed in deep sleep can strengthen memory. Learning a language from recordings while asleep does not work.
Children's sleep is especially effective, and naps help toddlers and preschoolers. Teenagers' body clocks shift later, so later school start times improve sleep, attention and grades.
Practise a little every day, review a lesson before bed, and protect your sleep.
The book's four slogans
The conclusion lists thirteen take-home messages, including: do not underestimate children, enrich the environment, keep learners active but guided, correct errors without punishment, and let students sleep.
Questions to think about
Jot down a few notes. You will not be asked to present them, only to use them in conversation.
- Think of a skill you learned well. Which of the four pillars did the most work? Which was missing when you struggled with something else?
- When is your attention at its best? What distracts you most, and have you ever missed something obvious, like the invisible gorilla?
- Dehaene says pure discovery learning fails, but passive lectures also fail. Where would you draw the line between guidance and freedom?
- How do you react to being corrected? Do you prefer immediate or delayed feedback, and from whom?
- Do you cram, or do you space your study? Has sleep ever helped you solve a problem?
- Which important idea about learning, for instance learning styles, did you believe before reading this? What changed your mind?
- Is any pillar missing from the list? Motivation, emotion, other people, or something else?