Knowing when to learn. Infants’ sensitivity to confounded information in a causal reasoning task.

• In Event: Recognising and shaping opportunities for learning: neurocognitive insight into active learning in early infancy

Abstract

Identifying causes, by distinguishing spurious correlations from unconfounded evidence, is a challenge in every day reasoning as well as fundamental to the scientific process. Sensitivity to confounded information is of particular importance when one cannot generate data for oneself, as is generally the case for infants. Before the age of 2 years, infants have limited means of acting on the world and much of their learning depends on observing others’ interventions and drawing the correct inferences. We investigate whether infants recognize when the information they are expecting or observing is confounded or not, whether they use this distinction to correctly ascribe features to novel objects, and predict future events based on these inferences. 

Specifically, infants are familiarised with an adult placing objects on a box (‘blicket detector’), which produces an audio-visual effect when contacted by some of the objects (‘blickets’), and not others (e.g. Sobel et al., 2014). The objects are then sorted into 2 possible boxes on either side of the scene, depending on whether they have produced an effect. Following this, pairs of sorted objects are placed on the box, providing further information on the causal structure. Infants learn that a single ‘blicket’ is sufficient to activate the box, and two also activate the box.

In test trials, infants see a new unknown object placed on the box, paired with one of the known (and visually marked) objects, and the box activates. This provides infants with either confounded (when paired with a ‘blicket’) or unconfounded information about the novel object. We investigate whether infants distinguish between the two scenarios by measuring infants’ expectation of information as indexed by EEG theta activity, a neural rhythm shown to predict the degree of learning (Begus et al., 2014; Guderian et al., 2009) and to index an active and selective preparation for encoding information, in infants and adults (Begus et al., 2016; Gruber et al., 2013). If infants recognise which of the two situations offers the opportunity to learn about the functionality of the unknown object, a greater increase in theta activity would be expected in the unconfounded compared to confounded trials. Furthermore, we investigate whether infants correctly infer which category the new object belongs to (when such prediction is possible, i.e. in unconfounded trials), by measuring anticipatory looks towards either of the two boxes into which the objects are sorted. We expect infants to demonstrate preferential anticipation towards the correct box when the information they observed was unconfounded, and chance performance when that was not the case.

Following the results from a pilot experiment, suggesting a relationship between infants’ sensitivity to confounded information and making the correct causal inferences, a second experiment is currently conducted with infants between 15-17-months (planned N=30), the age range during which the ability for such causal inferences is believed to emerge. By relating the accuracy of infants’ inferences to the preceding measures of information expectation, we hope to offer a unique insight into the mechanisms of, and individual differences in, early causal reasoning.

Authors

• Katarina Begus, Harvard University
  Presenting Author

• Elizabeth Bonawitz, Rutgers University - Newark
  Non-Presenting Author