CHAPTER 3

Babies’ Invisible Knowledge

On the surface, what could be more destitute of knowledge than a newborn? What could be more reasonable than to think, as Locke did, that the infant’s mind is a “blank slate” simply waiting for the environment to fill its empty pages? Jean-Jacques Rousseau (1712–78) strove to drive this point home in his treatise Emile, or On Education (1762): “We are born capable of learning, but knowing nothing, perceiving nothing.” Almost two centuries later, Alan Turing, the father of contemporary computer science, took up the hypothesis: “Presumably the child brain is something like a notebook as one buys it from the stationer’s. Rather little mechanism, and lots of blank sheets.”

We now know that this view is dead wrong—nothing could be further from the truth. Appearances can be deceiving: despite its immaturity, the nascent brain already possesses considerable knowledge inherited from its long evolutionary history. For the most part, however, this knowledge remains invisible, because it does not show in babies’ primitive behavior. It therefore took cognitive scientists much ingenuity and significant methodological advances in order to expose the vast repertoire of abilities all babies are born with. Objects, numbers, probabilities, faces, language . . . the scope of babies’ prior knowledge is extensive.

THE OBJECT CONCEPT

We all have the intuition that the world is made of rigid objects. In reality, it is made up of atoms, but at the scale on which we live, these atoms are often packed together into coherent entities that move as a single blob and sometimes collide without losing their cohesiveness. . . . These large bundles of atoms are what we call “objects.” The existence of objects is a fundamental property of our environment. Is this something that we need to learn? No. Millions of years of evolution seem to have engraved this knowledge into the very core of our brains. As early as a few months of age, a baby already knows that the world is made up of objects that move coherently, occupy space, do not vanish without reason, and cannot be in two different places at the same time.¹ In a sense, babies’ brains already know the laws of physics: they expect the trajectory of an object to be continuous in space as in time, without any sudden jump or disappearance.

How do we know this? Because babies act surprised in certain experimental situations that violate the laws of physics. In today’s cognitive science laboratories, experimenters have become magicians (see figure 5 in the color insert). In small theaters specially designed for babies, they play all sorts of tricks: on the stage, objects appear, disappear, multiply, pass through walls. . . . Hidden cameras monitor the babies’ gazes, and the results are clear-cut: even babies a few weeks old are sensitive to magic. They already possess deep intuitions of the physical world and, like all of us, are stunned when their expectations turn out to be false. By zooming in on the children’s eyes—to determine where they look and for how long—cognitive scientists manage to accurately measure their degree of surprise and infer what they expected to see.

Hide an object behind a book, then suddenly crush it flat, as if the hidden object no longer existed (in reality, it escaped through a trapdoor): babies are flabbergasted! They fail to understand that a solid object can vanish into thin air. They appear dumbfounded when an object disappears behind one screen and reappears behind another, without ever being seen in the empty space between the two screens. They are also amazed when a small train rolling down a slope seamlessly passes through a rigid wall. And they expect objects to form a coherent whole: if they see two ends of a stick moving coherently on both sides of a screen, they expect them to belong to a single stick and are shocked when the screen lowers and reveals two distinct rods (see below).

Babies therefore possess a vast knowledge of the world, but they don’t know everything from the start, far from it. It takes a few months for babies to understand how two objects can support each other.² At first, they don’t know that an object falls when you drop it. Only very gradually do they become aware of all the factors that make an object fall or stay put. First, they realize that objects fall when they lose their support, but they think that any sort of contact suffices to keep an object still—for example, when a toy is placed at the edge of a table. Progressively, they realize that the toy must not only be in contact with the table, but on top of it, not under or against it. Finally, it takes them a few more months to figure out that this rule is not enough: in the end, it’s the center of gravity of the object that must remain above the table.

Babies possess extremely early intuitions of arithmetic, physics, and even psychology. To assess them, researchers evaluate whether babies look at a surprising scene for a longer time than an unsurprising one. When a box contains a majority of black balls, babies are surprised to see a white one come out (intuition of numbers and probabilities). If two ends of a stick move coherently, babies are dumbfounded when two different rods are revealed (intuition of objects). And if babies see a ball move autonomously and jump over a wall before escaping to the right-hand side, they deduce that the ball is a living being with an intention of its own, and they are amazed if it keeps jumping once the wall has disappeared (intuition of psychology).

Keep this in mind the next time your baby drops his or her spoon from the table for the tenth time, to your great despair: they are only experimenting! Like any scientist, children need a whole series of trials to successively reject all the wrong theories, in the following order: (1) objects stay put in the air; (2) they must touch another object to not fall; (3) they must be atop another object to not fall; (4) the majority of their volume must be above another object to not fall, and so on and so forth.

This experimental attitude continues all the way into adulthood. We are all fascinated with gadgets that seem to violate the usual laws of physics (helium balloons, mobiles in equilibrium, roly-poly toys with a displaced center of gravity . . .), and we all enjoy magic shows where rabbits disappear in a hat and women are sawed in two. These situations entertain us because they violate the intuitions that our brain has held since birth and refined in our first year of life. Josh Tenenbaum, a professor of artificial intelligence and cognitive science at MIT, hypothesizes that babies’ brains host a game engine, a mental simulation of the typical behavior of objects similar to the ones that video games use in order to simulate different virtual realities. By running these simulations in their heads, and by comparing simulations with reality, babies discover very early on what is physically possible or probable.

THE NUMBER SENSE

Let’s take a second example: arithmetic. What could be more obvious than that babies have no understanding of mathematics? And yet, since the 1980s, experiments have shown quite the opposite.³ In one experiment, babies are repeatedly presented with slides showing two objects. After a while, they get bored . . . until they are shown a picture with three objects: suddenly, they stare longer at this new scene, indicating that they detected the change. By manipulating the nature, size, and density of objects, one can prove that children are genuinely sensitive to the number itself, i.e., the cardinal of the whole set, not another physical parameter. The best proof that infants possess an abstract “number sense” is that they generalize from sounds to images: if they hear tu tu tu tu—that is, four sounds—they are more interested in a picture that has a matching number of four objects in it than in a picture that has twelve, and vice versa.⁴ Well-controlled experiments of this sort abound and convincingly show that, at birth, babies already possess the intuitive ability to recognize an approximate number without counting, regardless of whether the information is heard or seen.

Can babies calculate too? Suppose that children see an object hide behind a screen, followed by a second one. The screen then lowers—lo and behold, only one object is there! Babies manifest their surprise in a prolonged investigation of the unexpected scene.⁵ If, however, they see the two expected objects, they look at them for only a brief moment. This behavior of “cognitive surprise,” in reaction to the violation of a mental calculation, shows that, as early as a few months of age, children understand that 1 + 1 should make 2. They build an internal model of the hidden scene and know how to manipulate it by adding or removing objects. And such experiments work not only for 1 + 1 and 2 − 1, but also for 5 + 5 and 10 − 5. Provided that the error is big enough, nine-month-old babies are surprised whenever a concrete display hints at a wrong calculation: they can tell that 5 + 5 cannot be 5, and that 10 − 5 cannot be 10.⁶

Is this really an innate skill? Could the first months of life suffice for a child to learn the behavior of sets of objects? While children undoubtedly refine the accuracy with which they perceive numbers⁷ over the first months of life, the data show, equally clearly, that the starting point for children is not a blank slate. Newborns perceive numbers within a few hours of life—and so do monkeys, pigeons, ravens, chicks, fish, and even salamanders. And with the chicks, the experimenters controlled all the sensory inputs, making sure that the baby chicks did not see even a single object after they hatched . . . yet the chicks recognized numbers.⁸

Such experiments show that arithmetic is one of the innate skills that evolution bestows unto us, as well as many other species. Brain circuits for numbers have been identified in monkeys and even in ravens. Their brains contain “number neurons” that behave in a very similar way: they are attuned to specific numbers of objects. Some neurons prefer to see one object, others two, three, five, or even thirty objects—and, crucially, these cells are present even in animals that have not received any specific training.⁹ My lab has used brain-imaging techniques to show that, at homologous locations of the human brain, our neuronal circuits also contain similar cells attuned to the cardinal number of a concrete set—and recently, with advances in recording techniques, such neurons have been directly recorded in the human hippocampus.¹⁰

Incidentally, these results overturn several tenets of a central theory of child development, that of the great Swiss psychologist Jean Piaget (1896–1980). Piaget thought that young infants were not endowed with “object permanence”—the fact that objects continue to exist when they are no longer seen—until the end of the first year of life. He also thought that the abstract concept of number was beyond children’s grasp for the first few years of life, and that they slowly learned it by progressively abstracting away from the more concrete measures of size, length, and density. In reality, the opposite is true. Concepts of objects and numbers are fundamental features of our thoughts; they are part of the “core knowledge” with which we come into the world, and when combined, they enable us to formulate more complex thoughts.¹¹

Number sense is only one example of what I call infants’ invisible knowledge: the intuitions that they possess from birth and that guide their subsequent learning. Here are more examples of the skills researchers have demonstrated in babies as young as a few weeks old.

THE INTUITION OF PROBABILITIES

Going from numbers to probabilities takes only one step . . . a step that researchers have recently taken by wondering if babies a few months old could predict the outcome of a lottery draw. In this experiment, babies are first presented with a transparent box containing balls that move around randomly. There are four balls: three red and one green. At the bottom, there is an exit. At some point, the container is occluded, and then either a green ball or a red ball comes out the bottom. Remarkably, the child’s surprise is directly related to the improbability of what she sees: if a red ball comes out—the most likely event, since the majority of the balls in the box are red—the baby looks at it for only a brief moment . . . whereas if the more improbable outcome occurs, that is, a green ball that had only one chance in four to come out, the baby looks at it for much longer.

Subsequent controls confirm that babies run, in their little heads, a detailed mental simulation of the situation and the associated probabilities. Thus, if we introduce a partition that blocks the balls, or if we move the balls closer to or farther away from the exit, or if we vary the time before the balls exit the box, we find that infants integrate all these parameters into their mental calculation of probability. The duration of their gaze always reflects the improbability of the observed situation, which they seem to compute based on the number of objects involved.

All these skills surpass those of most current artificial neural networks. Indeed, infants’ surprise reaction is far from trivial. Being surprised indicates that the brain was able to estimate the underlying probabilities and concluded that the observed event had but a tiny chance of occurring. Because babies’ gazes show elaborate signs of surprise, their brains must be capable of probabilistic calculations. Indeed, one of the most popular current theories of brain function views the brain as a probabilistic computer that manipulates probability distributions and uses them to anticipate future events. Infant experiments reveal that even babies are equipped with such a sophisticated calculator.

A series of recent studies further shows that babies come equipped with all the mechanisms to make complex probabilistic inferences. Do you remember the Reverend Bayes’s mathematical theory of probabilities, which allows us to trace an observation back to its probable causes? Well, even babies a few months old already seem to reason according to Bayes’s rule.¹² Indeed, not only do they know how to go from a box of colored balls to the corresponding probabilities (forward reasoning) as we just saw, but also from observations back to the content of the box (reverse inference). In one experiment, we first show babies an opaque box, whose contents are hidden. Then we bring in a blindfolded person, who randomly takes out a series of balls. The balls appear one after another, and it turns out that the majority are red. Can babies infer that the box must contain an abundance of red balls? Yes! When we eventually open the box and show them that it contains a majority of green balls, they are surprised and look longer than if the box turns out to be full of red balls. Their logic is impeccable: If the box is filled mostly with green balls, how do we explain that the random draw yielded so many red balls?

Again, this behavior may not seem like much, but it implies an extraordinary ability for implicit, unconscious reasoning that works in both directions: given a sample, infants can guess the characteristics of the set from which it was drawn; and, vice versa, given a set, they manage to guess how a random sample should look.

From birth on, thus, our brain is already endowed with an intuitive logic. There are now many variations of those basic experiments. They all demonstrate the extent to which children behave like budding scientists who reason like good statisticians, eliminating the least likely hypotheses and searching for the hidden causes of various phenomena.¹³ For example, the American psychologist Fei Xu showed that if eleven-month-olds see a person draw a majority of red balls from a container, and then find out that the container holds a majority of yellow balls, they are surprised, of course, but they also make an additional inference: that the person prefers the red balls!¹⁴ And if they see that a draw is not random but follows a specific pattern, say, a perfect alternation of a yellow ball, a red ball, a yellow ball, a red ball, and so on, then they deduce that a human, not a machine, made the draw.¹⁵

Logic and probability are closely linked. As Sherlock Holmes put it, “When you have eliminated the impossible, whatever remains, however improbable, must be the truth.” In other words, one can turn a probability into a certainty by using reasoning to eliminate some of the possibilities. If a baby can juggle with probabilities, she must also master logic, because logical reasoning is only the restriction of probabilistic reasoning to probabilities 0 and 1.¹⁶ This is exactly what the philosopher and developmental psychologist Luca Bonatti recently showed. In his experiments, a ten-month-old baby first sees two objects, a flower and a dinosaur, hide behind a screen. Then one of these objects exits from the screen, but it is impossible to tell which one because it is partially hidden in a pot, so that only the top can be seen. Later, the dinosaur exits from the other side of the screen, in full sight. At this point, the child can make a logical deduction: “It is either the flower or the dinosaur that is hiding in the pot. But it cannot be the dinosaur, because I have just seen it come out from the other side. So, it must be the flower.” And it works: the baby is not surprised if the flower comes out of the pot, but she is if the dinosaur comes out. Furthermore, the baby’s gaze reflects the intensity of her logical reasoning: like an adult, her pupils dilate at the precise moment when deduction becomes possible. A true Sherlock Holmes in diapers, the baby seems to start with several hypotheses (it is either the flower or the dinosaur) and then eliminates some of them (it cannot possibly be the dinosaur), thus moving from probability to certainty (it must be the flower).

“Probability theory is the language of science,” Jaynes tells us—and infants already speak this language: way before they pronounce their first words, they manipulate probabilities and combine them in sophisticated syllogisms. Their sense of probability allows them to draw logical conclusions from the observations they make. They are constantly experimenting, and their budding scientist brains ceaselessly accumulate the conclusions of their research.

KNOWLEDGE OF ANIMALS AND PEOPLE

While babies have a good model of the behavior of inanimate objects, they also know that there is another category of entities that behave entirely differently: animate things. From the first year of life, babies understand that animals and people have a specific behavior: they are autonomous and driven by their own movements. Therefore, they do not have to wait for another object to bump into them, like a pool ball, in order to move around. Their movement is motivated from within, not caused from the outside.

Babies are therefore not surprised to see animals move by themselves. In fact, for them, any object that moves by itself, even if it is in the shape of a triangle or a square, is immediately labeled as an “animal,” and from that moment on, everything changes. A small child knows that living beings do not have to move according to the laws of physics, but that their movements are governed by their intentions and beliefs.

Let us take an example: if we show babies a sphere that moves in a straight line, jumps over a wall, then heads to the right, little by little, they will get bored of it. Are they simply getting used to this peculiar motion? No, in fact, they understand much more. They deduce that this is an animate being with a specific intention: it wants to move to the right! Moreover, they can tell the object is highly motivated, because it jumps over a high wall in order to get there. Now let’s remove the wall. In this scenario, babies are not surprised if they see the sphere change its motion and move to the right in a straight line, without jumping—this is simply the best way to attain its goal. On the other hand, babies open their eyes wide if the sphere continues to jump in the air for no particular reason, since the wall has vanished! In the absence of a wall, the same trajectory as in the first scenario leaves the babies surprised, because they do not understand what strange intention the sphere might have.¹⁷ Other experiments show that children routinely infer people’s intentions and preferences. In particular, they understand that the higher the wall is, the greater the person’s motivation must be in order to jump over it. From their observations, babies can infer not only the goals and intentions of those around them, but also their beliefs, abilities, and preferences.¹⁸

Infants’ notion of living beings does not end there. Around ten months, babies start attributing personalities to people: if they see someone throw a child to the ground, for example, they deduce that this person is ill-intentioned, and they turn away from her. They clearly prefer a second person who helps the child get back up.¹⁹ Long before they are able to pronounce the words mean and nice, they are able to formulate these concepts in their language of thought. Such a judgment is quite subtle: even a nine-month-old baby can distinguish between someone who intentionally does harm and someone who does it by accident, or someone who intentionally refuses to help another person and someone who does not have the opportunity to help.²⁰ As we will see later, this social skill plays a fundamental role in learning. Indeed, even a one-year-old child understands if someone is trying to teach him something. He can tell the difference between an ordinary action and an action with the goal of teaching something new. In this respect, a one-year-old child already possesses, according to the Hungarian psychologist György Gergely, an innate sense of pedagogy.

FACE PERCEPTION

One of the earliest manifestations of infants’ social skills is the perception of faces. For adults, the slightest hint suffices to trigger the perception of a face: a cartoon, a smiley, a mask. . . . Some people even detect the face of Jesus Christ in the snow or on burnt toast! Remarkably, this hypersensitivity to faces is already present at birth: a baby a few hours old turns its head more quickly to a smiley face than to a similar image turned upside down (even if the experimenter ensures that the newborn has never had the chance to see a face). One team even managed to present a pattern of light to fetuses through the wall of the uterus.²¹ Surprisingly, the researchers showed that three dots arranged in the shape of a face () attracted the fetus more than three dots arranged in the shape of a pyramid (). Face recognition seems to start in utero!

Many researchers believe that this magnetic attraction to faces plays an essential role in the early development of attachment—especially since one of the earliest symptoms of autism is avoiding eye contact. By attracting our eyes to faces, an innate bias would force us to learn to recognize them—and indeed, as early as a couple of months after birth, a region of the visual cortex of the right hemisphere begins to respond to faces more than to other images, such as places.²² The specialization for faces is one of the best examples of the harmonious collaboration between nature and nurture. In this domain, babies exhibit strictly innate skills (a magnetic attraction to face-like pictures), but also an extraordinary instinct to learn the specifics of face perception. It is precisely the combination of these two factors that allows babies, in a little less than a year, to go beyond naively reacting to the mere presence of two eyes and a mouth and to start preferring human faces to those of other primates, such as monkeys and chimpanzees.²³

THE LANGUAGE INSTINCT

The social skills of small children are manifest not only in vision, but also in the auditory domain—spoken language comes to them just as easily as face perception. As Steven Pinker famously noted in his best-selling book The Language Instinct (1994), “Humans are so innately hardwired for language that they can no more suppress their ability to learn and use language than they can suppress the instinct to pull a hand back from a hot surface.” This statement should not be misunderstood: obviously, babies are not born with a full-blown lexicon and grammar, but they possess a remarkable capacity to acquire them in record time. What is hardwired in them is not so much language itself as the ability to acquire it.

Much evidence now confirms this early insight. Right from birth, babies already prefer listening to their native language rather than to a foreign one²⁴—a truly extraordinary finding which implies that language learning starts in utero. In fact, by the third trimester of pregnancy, the fetus is already able to hear. The melody of language, filtered through the uterine wall, passes on to babies, and they begin to memorize it. “As soon as the sound of your greeting reached my ears, the baby in my womb leaped for joy,” said the pregnant Elizabeth when Mary visited her.²⁵ The Evangelist was not mistaken: in the last few months of pregnancy, the growing fetus’s brain already recognizes certain auditory patterns and melodies, probably unconsciously.²⁶

This innate ability is obviously easier to study in premature babies than in fetuses. Out of the womb, we can equip their tiny heads with miniature electroencephalography and cerebral blood flow sensors and peek into their brains. With this method, my wife, professor Ghislaine Dehaene-Lambertz, discovered that even babies born two and a half months before term respond to spoken language: their brain, although immature, already reacts to changes in syllables as well as in voices.²⁷

It was long thought that language acquisition does not begin until one or two years of age. Why? Because—as its Latin name, infans, suggests—a newborn child does not speak and therefore hides its talents. And yet, in terms of language comprehension, a baby’s brain is a true statistical genius. To show this, scientists had to deploy a whole panoply of original methods, including the measurement of infants’ preferences for speech and nonspeech stimuli, their responses to change, the recording of their brain signals. . . . These studies gave converging results and revealed how much infants already know about language. Right at birth, babies can tell the difference between most vowels and consonants in every language in the world. They already perceive them as categories. Take, for instance, the syllables /ba/, /da/, and /ga/: even if the corresponding sounds vary continuously, babies’ brains treat them as distinct categories separated by sharp borders, just like adults.

These early innate skills become shaped by the linguistic environment during the first year of life. Babies quickly notice that certain sounds are not used in their language: English speakers never utter vowels like the French /u/ and /eu/, and Japanese locutors fail to differentiate between /R/ and /L/. In just a few months (six for vowels, twelve for consonants), the baby’s brain sorts through its initial hypotheses and keeps only the phonemes that are relevant to the languages that are present in its environment.

But that’s not all: babies quickly start to learn their first words. How do they go about identifying them? First, babies rely on prosody, the rhythm and intonation of speech—the way our voices rise, fall, or stop, thus marking the boundaries between words and sentences. Another mechanism identifies which speech sounds follow each other. Again, babies behave like budding statisticians. They realize, for example, that the syllable /bo/ is often followed by /t^l/. A quick calculation of probabilities tells them that this cannot be due to chance: /t^l/ follows /bo/ with too high a probability; these syllables must form a word, “bottle”—and this is how this word is added to the child’s vocabulary and can later be related to a specific object or concept.²⁸ As early as six months of age, children have already extracted the words that recur with a high frequency in their environment, such as “baby,” “daddy,” “mommy,” “bottle,” “foot,” “drink,” “diaper,” and so forth. These words become engraved in their memory to such an extent that, as adults, they continue to hold a special status and are processed more effectively than other words of comparable meaning, sound, and frequency acquired later in life.

Statistical analysis also allows babies to identify certain words that occur more frequently than others: small grammatical words such as articles (a, an, the) and pronouns (I, you, he, she, it . . .). By the end of their first year, babies already know many of them, and they use them to find other words. If, for example, they hear one of their parents say, “I made a cake,” they can parse out the small function words “I” and “a” and, by elimination, discover that “made” and “cake” are also words. They already understand that a noun often comes after an article and a verb usually comes after a pronoun—to such an extent that, around twenty months of age, babies react with utter surprise if they are told incoherent phrases like “I bottle” or “the finishes.”²⁹

Of course, such a probabilistic analysis isn’t entirely foolproof. When French children hear “un avion” (an airplane), which is pronounced with a liaison (the n of “un” melds into the a of “avion”), they improperly infer the existence of the word navion (“Regarde le navion!”). Conversely, English speakers imported the French word napperon (place mat) and, due to incorrect parsing of the phrase un napperon, invented the word apron.

Such shortcomings are rare, however. In a few months, children quickly manage to surpass any existing artificial intelligence algorithm. By the time they blow out their first candle, they have already laid down the foundation for the main rules of their native language at several levels, from elementary sounds (phonemes) to melody (prosody), vocabulary (lexicon), and grammar rules (syntax).

No other primate species is capable of such abilities. This very experiment has been attempted many times: several scientists tried adopting baby chimpanzees, treating them like family members, speaking to them in English or sign language or with visual symbols . . . only to find out, a few years later, that none of these animals mastered a language worthy of the name: they knew, at most, a few hundred words.³⁰ The linguist Noam Chomsky, therefore, was probably right in postulating that our species is born with a “language acquisition device,” a specialized system that is automatically triggered in the first years of life. As Darwin said in The Descent of Man (1871), language “certainly is not a true instinct, for every language has to be learnt,” but it is “an instinctive tendency to acquire an art.” What is innate in us is the instinct to learn any language—an instinct so irrepressible that language appears spontaneously within a few generations in humans deprived of it. Even in deaf communities, a highly structured sign language, with universal linguistic characteristics, emerges from the second generation on.³¹