Expectations of Causal Structure Influence Preschoolers’ Choice of Interventions.

• In Event: 2-193 - Information Gain as a Guide for Decision-Making in Early Exploratory Learning

Integrative Statement

In everyday reasoning, it is often impossible to infer causal structure from observation alone. Instead, learners must pair their observations with informative interventions. However, even for a simple causal system, there are often a large number of possible interventions that a learner might take, not all of which will turn out to be informative. How do learners decide where to intervene on an unknown causal system in order to learn more about it? Here we explore the hypothesis that even very young learners’ expectations about the possible causal structures of a system may help them to resolve the problem of selection by constraining and guiding their choice of intervention.

Forty-eight preschoolers (range: 46-82 mos., M: 64 mos.) were introduced to a novel causal system: a machine with two interlocking gears on top and a switch on the front (Schulz, Gopnik, & Glymour, 2007). Children were taught that the gears rotated when the switch was flipped to the ‘on’ position. They then observed a pair of gears that remained inert when the machine was turned on, and were provided with illustrations of two possible causes for this unexpected outcome (Figure 1). One possibility was that both gears were ‘broken’ (i.e., stuck) and unable to spin. The other possibility was that only one of the two gears (either the left or the right, counterbalanced) was ‘broken’, thus preventing its mate from spinning. Children were then offered a choice between two interventions to figure out which of the two possibilities accurately described the state of the world. Both interventions involved removing one of the two gears to observe the remaining gear operating in isolation. Critically however, only the intervention that isolated the broken gear in the ‘both-broken’ structure, but not in the ‘one-broken’ structure, would provide the information necessary to disambiguate the two. Children were also randomly divided into one of two conditions: scaffold and no-scaffold. Prior to making a selection, only children in the scaffold condition were asked to explicitly consider the potential outcomes of each intervention. If the consideration of possibilities is key to selecting informative interventions, scaffolding should improve children’s decision-making.

Across conditions, the majority of children (81.25%, p < .001) chose the intervention that was informative about the true causal structure of the system. There was also a significant effect of scaffolding on children’s intervention choice (p = 0.02; φ = 0.37, Fisher Exact), with 95.83% of children in the scaffold condition selecting the informative gear (p < 0.001, two-tailed binomial), compared with 66.67% of those in the no-scaffold condition (p = 0.15, two-tailed binomial). Regardless of condition, all 39 of the children who selected informative interventions made subsequent causal inferences that were consistent with the evidence they generated. These findings suggest that a learner’s choice of interventions on an unknown causal system is sensitive to the informative potential of those interventions from an early age. Ongoing and future work further examining the relationship between expectation and intervention will also be reviewed.

Authors

• Elizabeth Lapidow, University of California, San Diego
  Presenting Author

• Caren Walker, University of California, San Diego
  Non-Presenting Author