Children’s Exploration During Play Indicates Both Rational and Irrational Strategies for Learning

• In Event: Developmental Change in the Form and Function of Exploratory Learning

Abstract

Numerous findings suggest young children have precocious causal reasoning abilities. This view posits children as “little scientists” who explore the environment rationally to learn causal structure (e.g., Gopnik & Wellman, 2012). For instance, preschoolers select unconfounded interventions over confounded ones to resolve ambiguous data (Lapidow & Walker, 2020). Such findings contrast with research suggesting children lack the ability to design unconfounded experiments (Kuhn, 2007). Instead, preschoolers act like “little engineers”, focused not only learning, but on making as many outcomes occur during their exploration (e.g., McCormack et al., 2016).

We hypothesized that children might use both strategies, but are influenced by their initial action. If their first action is efficacious, children act more like a “little engineer,” and subsequently try to produce effects, regardless of their value for learning. In contrast, if their first action is inefficacious, they act more like a “little scientist”, and are more likely to generate the data necessary to learn. Further, we hypothesized that generating those data, not their initial action, predicts learning.

We presented 4-7-year-olds (N=207) with a “blicket detector” and a block that could hold two Lego bricks. As long as one brick on the block was effective, the block would activate the machine, and it would only activate the machine if two bricks were on it. Children were shown that when the block had a Red (A) and Black (B) Lego attached, it activated the machine. Children were told that only one of A or B was actually efficacious, and then were asked to play with the machine and Legos. Children were always allowed to play with the B brick and two distractor (C and D) bricks. The B and C bricks were always ineffective. Using a 2x2 design, we varied whether children also were given the A brick (which was effective) and whether the D brick was effective or not.

Controlling for the conditions and age, children whose first action during play was efficacious had a greater frequency of trials (not counting the first) on which they activated the machine, 84% vs. 24%, B = 2.41, SE = 0.44, Wald Chi-Squared(1) = 29.76, p < .001. Also controlling for conditions and age, children whose first action was efficacious were less likely to ever generate the data necessary to learn the causal structure, 44% vs 88%, B = -1.58, SE = 0.49, Wald Chi-Squared(1) = 10.29, p = .001. The conditions, age, and whether the first action was efficacious did not predict whether children learn the causal structure, but whether they observed the data necessary to do so during their play did so, 66% vs. 46%, B = 0.90, SE = 0.36, Wald Chi-Squared(1) = 6.31, p = .01. Age was significantly related to children’s strategies during play, but not their ability to learn.

These data suggest that children can learn from play, but the results of their self-generated actions during play affects whether they recognize that exploration for learning isn’t always its own reward.

Authors

• David Marc Sobel, Brown University
  Presenting Author

• Deon Benton, Carnegie Mellon University
  Non-Presenting Author

• Zoe Finiasz, Cornell University
  Non-Presenting Author

• Yuyi Taylor, Brown University
  Non-Presenting Author

• Deena Skolnick Weisberg, Villanova University
  Non-Presenting Author