Infants explore the unexpected

See allHide authors and affiliations

Science  03 Apr 2015:
Vol. 348, Issue 6230, pp. 42-43
DOI: 10.1126/science.aab0582

Science can delight us with new and surprising findings. Sometimes, however, a study delights us by confirming something we already believed but could not yet prove. This is the kind of pleasure occasioned by Stahl and Feigenson's report on page 91 of this issue (1). In a series of elegant experiments, the authors show that, controlling for overall attention, 11-month-old infants are more likely to learn a new sound associated with an object if the object previously violated the infants' expectations (e.g., by appearing to pass through walls or roll over gaps without falling) than if the object behaved as expected. Moreover, infants not only selectively explore objects that violate their expectations but also explore in ways specific to the violation. Thus, they bang objects that violate expectations of solidity and drop objects that violate expectations of support (see the figure).

Perhaps the most surprising thing about these observations is that they had not been made sooner. For decades, researchers have known that infants look longer at events that violate their expectations than at events consistent with their prior beliefs (2). The presumption was that such selective attention must support learning, but it was difficult to show this in a way that did not follow trivially from the fact that infants look for a long time at unexpected events. The current study solves that problem by matching infants' initial exposure to the events and then asking whether infants who observe theory-violating evidence are more likely to learn an unrelated property of the objects.

Researchers have also long assumed that children's exploratory play must support learning (35). Again, however, it has been difficult to demonstrate this in a way that does not follow trivially from the fact that the longer children explore an object, the more of its properties they are likely to discover. In the past decade, research has begun to bridge the gap between formal models of learning (6) and children's play. This research has, for instance, shown that preschoolers engage in selective exploration both when evidence violates their prior beliefs (i.e., when evidence is surprising) (7) and when evidence fails to distinguish equally probable possibilities (i.e., when evidence is confounded) (8).

The importance of violating expectation.

Stahl and Feigenson show that babies who have previously seen an object behave in an unexpected way are more likely to explore this object. The results help to understand how babies and young children learn.


Moreover, children spontaneously explore in ways that tend to support information gain. They are sensitive to the degree to which exploration reduces uncertainty (9) and will isolate causal variables and test them one at a time in response to both surprising (10) and confounded (11) evidence. However, such studies have focused on preschoolers or older children. It is both gratifying and astonishing to find not only that infants less than a year old selectively explore objects that violate their prior beliefs but also that they do so in ways directly relevant to the violation they observe.

Stahl and Feigenson frame their study with respect to “core knowledge,” the idea that cognitive representations of objects and agents are present at birth in humans and other species. Constraints imposed by these innate representations could support infants' ability to learn rapidly from sparse data. However, when the authors suggest that violations of core knowledge might signal “a special opportunity for learning” one has to wonder how typical of learning these opportunities really are. Core knowledge is hypothesized to be an innate component of human cognition precisely because it represents aspects of the environment that are stable across evolutionary time scales. It is thus not likely to be violated (outside of developmental laboratories). If infants were to rely on violations of core knowledge as a “wedge into the hard problems of knowing when and what to learn,” they would learn very little. It seems far more likely, as the authors acknowledge in their conclusion, that violations of any expectations—whether innate or recently acquired—provide special opportunities for learning.

One might also wonder how much of a violation of expectation is a good thing. Infants do not always prefer information that is more surprising or more complex: They appear to like things that they think they can learn. Thus, for instance, infants prefer to attend to patterns of linguistic input that are predictable enough to permit learning (12) and to sequences of stimuli that are neither too predictable nor too unpredictable (13). Quantitative models may help to explore the ways in which the predictability of events affects infants' learning. Bringing together the methods and results of the present study with those of other recent exciting experimental results and computational models of infant's attention (14) could pave the way toward a more comprehensive theory of how infants learn and explore.


View Abstract

Stay Connected to Science

Navigate This Article