Guiding Curiosity: How Learning Progress Shapes Young Children's Exploration of New Toys

Previous research has shown that children are drawn to environments that offer learning progress. However, the computational implementation of learning progress varied across studies, potentially overshadowing that there might be two distinct processes at work. In this study, we examined two forms of learning progress-one based on performance increase and the other on information gain-by observing children's exploration as it unfolded during free play, moving beyond the computerized and controlled settings in which this concept was studied in the past. We created a setting that allowed 56 3.5-year-old children to interact freely with their environment, but at the same time allowed us to model their behaviour in a fine-grained fashion. Specifically, we developed three novel toys and estimated the degree to which their play engagement was predicted by the amount of learning progress they made, leveraging the contributions of two existing computational accounts of learning progress. Results showed that performance-based and information-based learning progress were weakly correlated and that both forms contributed to children's exploration behaviour. However, information-based learning progress predicted children's play engagement only for one of the three toys. Overall, we demonstrate that performance increase and information gain capture distinct aspects of learning progress that both guide children's exploration behaviour. This study marks an important step toward research investigating the mechanisms underlying children's exploration behaviour in more naturalistic settings. SUMMARY: We developed three novel toys and created a setting in which 3.5-year-old children were free to explore them on their own pace and without instruction. We test the Learning Progress hypothesis by leveraging two computational accounts of learning progress, one focusing on performance improvement and the other on information gain. Performance-based learning progress predicted children's exploration for all toys, while information-based learning progress only predicted exploration with one toy, highlighting the role of context.