No surprise: A key to infant learning is surprise


Baby play just got a little less random.

It turns out that infants are natural scientists, spurred by surprises to test primitive hypotheses, according to a study published online Thursday in the journal Science.

Eleven-month-old infants who observed something that defied their rudimentary expectations, such as a ball that seemed to go through a wall, readily learned something new about that object but didn’t learn it after viewing an object acting normally, according to the study.

The babies also sought out the object that defied expectations and played with it in ways that seemed to test its properties. Such directed exploration has been widely chronicled in preschoolers and older children and is a pillar of formal models of learning, but it had not been demonstrated in young infants.


The experiments offer evidence to support what scientists had long assumed, but could never quite elucidate in ways that did not seem trivial: paying extra attention to the surprising probably supports learning.

“The trick was to figure out how to ask babies what they are thinking, since they’re too young to verbally report their beliefs to us,” said study coauthor Lisa Feigenson, a cognitive psychologist who is co-director of the Johns Hopkins University Laboratory for Child Development.

Chronicling surprise reactions, which can be elicited even from newborns, has been a reliable way to try to assess what expectations infants might have of their world and at what ages those rules are present. Hundreds of studies use the technique, and they’ve shown that young infants already have a notion of object solidity and continuity -- balls don’t just pass through walls, or disappear and reappear magically elsewhere. Babies even appear to have numeric and social expectations, studies have shown.

But how surprise mediates additional learning among infants remained a mystery.

Feigenson and Aimee Stahl, a PhD candidate in the child development lab, designed a series of experiments involving a total of 110 babies, to test the hypothesis that violations of core expectations offer a “special opportunity for learning.”

Babies witnessed events that appeared consistent or inconsistent with basic physics. A ball or car rolled down a ramp toward two walls, positioned in tandem and partially obscured by a screen. When the screen was removed, babies saw the rolled object resting against the first obstacle or the second -- as if it had passed through the first. Another group saw a ball put behind one of two screens, which then were lifted to reveal the ball behind in the same place, or behind the other screen.

The experiments and its variations are well-known methods to test whether babies have expectations that solid objects don’t go through other solid objects (object solidity), and that objects don’t disappear or move by themselves (object continuity).


After the ramp experiments, researchers played a squeaking sound from a hidden speaker while moving the target object (a ball or car) up and down. Then they gave the infants a five-second glimpse of the target object and a distractor object.

The researchers were interested in how well the infants would “map” the newly associated sound to the correct, or “target” object. So they moved both objects while the sound played and recorded the proportion of time the babies looked at each object.

Learning scores were based on the difference between the proportions of time babies spent fixating on the target object in the mapping test and the proportion they had spent looking at it when it was presented silently with the distractor.

Learning scores were significantly higher for infants who had previously seen the object behave in a knowledge-inconsistent way than for those that had seen it do the expected, the study showed.

“Even though all of the babies were equally interested in the events that we presented to them, we found that the infants who saw the expected event -- the ball stopped by the wall -- failed to learn this new information,” Stahl said. “However, infants who saw the surprising event -- the ball that passed through the wall – learned this new information very efficiently.”

Researchers ran the ramp and squeak trials again, but played a new sound when they moved both objects -- a rattle instead of a squeak. When the rattle played, infants who saw the knowledge-inconsistent trials didn’t increase the proportion of looking time to the target object. And learning scores were significantly greater in the original squeak test than in the test with the rattle.


That suggested that the infants had actually learned about a new property of the misbehaving object -- the original squeak -- and weren’t just preferring that object.

To test if learning was enhanced only toward the misbehaving object, they ran the continuity violation experiment (the ball magically moving from one hidden location to another), then presented a new object and showed that it had a sound associated with it.

Infants failed to map the new sound to the new object. Infants weren’t just generally aroused and ready to learn anything: They appeared to be focused on learning related to a specific object that defied expectations, the study suggests.

Given the chance, then, would they choose to explore that object over another that behaved normally? The pair ran the ramp experiment and added another scenario, to test object support. They rolled a ball or car across a hard surface, or rolled each object off the edge of that surface, where they appeared to hover, defying gravity.

Infants then got their choice of objects to explore for 60 seconds -- the target or a distractor. Infants who had seen the events that violated expectation had a stronger preference for the target object than those who had seen the consistent event.

What the infants did when they got hold of the target object was perhaps most surprising. The ones who had seen the object seemingly pass through a wall banged the object more than they dropped it; the ones who saw it hover unexpectedly dropped it more than they banged it.


“This suggests that infants were going out of their way to seek information from the very objects that violated their expectations,” Stahl said.

Researchers who have done similar work said the study represented a significant contribution to how play and learning are viewed. Babies are not just passive observers who eventually see a statistical regularity to events, the study suggests. They are little empiricists.

“What’s new here is this idea that when there is some violation, that attracts not just their attention, but also their desire to interact more with the object in principled ways that are consistent with the violation they observed,” said developmental psychologist Scott Johnson, director of UCLA’s baby lab.

“But what does that really do in the real world for them?” Johnson added. “It’s hard to say. That’s not a criticism. That’s just a question.”

Indeed, how much violation of expectation is a good thing for children remains an open question, Laura Schulz, principal investigator at MIT’s Early Childhood Cognition lab, wrote in a commentary published with the study. Infants, she notes, don’t always prefer information that is more surprising or complex, but things they think they can learn.

Feigenson said further research will test less extreme violations of prediction and examine the extent to which children of different ages employ the learning strategy demonstrated by the 11-month-olds.


“The question is when those predictions are wrong -- even when they’re not examples of impossible events, but merely the unexpected or improbable -- are those also opportunities for learning?” Feigenson said. “We think that this is probably a quite general aspect of human learning.”

Follow me on Twitter: @LATsciguy