UC Berkeley

A major challenge children face is uncovering the causal structure of the world around them. Previous research on children's causal inference has demonstrated their ability to learn about causal relationships in the physical environment using probabilistic evidence. However, children must also learn about causal relationships in the social environment, including discovering the causes of other people's behavior, and understanding the causal relationships between others' goal-directed actions and the outcomes of those actions. In addition, many of the causal relationships children experience do not occur in the physical world at all, but instead occur in richly causal imaginary worlds.

In this dissertation, we argue that social reasoning and causal reasoning are deeply linked, both in the real world and in children's minds. Children use both types of information together and in fact reason about both physical and social causation in fundamentally similar ways. We suggest that children jointly construct and update causal theories about their social and physical environment and that this process is best captured by probabilistic models of cognition. We also argue that causal pretense may serve as a form of counterfactual causal reasoning, allowing children to explore causal "what if" scenarios in alternative imaginary worlds, and suggest a theoretical link between the development of an extended period of immaturity in human evolution and the emergence of powerful and wide-ranging causal learning mechanisms.

We investigate the complex and varied ways in which children learn causal relationships through three primary lines of research, each of which extends probabilistic models beyond reasoning about purely physical causes, while also characterizing the distinctive aspects of causal pretense and social causal reasoning. In the first set of studies, we examine how causal learning can influence the understanding and segmentation of action, and how observed statistical structure in human action can affect causal inferences. We present a Bayesian analysis of how statistical and causal cues to segmentation should optimally be combined, as well as four experiments investigating human action segmentation and causal inference. We find that both adults and our model are sensitive to statistical regularities and causal structure in continuous action, and are able to combine these sources of information in order to correctly infer both causal relationships and segmentation boundaries.

The second line of work examines how the social context influences children's causal learning, focusing particularly on children's imitation of causal actions. We define a Bayesian model that predicts children will decide whether to imitate part or all of an action sequence based on both the pattern of statistical evidence and the demonstrator's pedagogical stance. We conducted an experiment in which preschool children watched an experimenter repeatedly perform sequences of varying actions followed by an outcome. Children's imitation of sequences that produced the outcome increased, in some cases resulting in production of shorter sequences of actions that the children had never seen performed in isolation. A second experiment established that children interpret the same statistical evidence differently when it comes from a knowledgeable teacher versus a naive demonstrator, suggesting that children attend to both statistical and pedagogical evidence in deciding which actions to imitate, rather than obligately imitating successful action sequences.

The final line of work explores the relationship between children's understanding of real-world causal structure and their pretend play. We report a study demonstrating a link between pretend play and counterfactual causal reasoning. Preschool children given new information about a causal system made very similar inferences both when they considered counterfactuals about the system and when they engaged in pretend play about it. Counterfactual cognition and causally coherent pretense were also significantly correlated even when age, general cognitive development and executive function were controlled for. These findings link a distinctive human form of childhood play and an equally distinctive human form of causal inference. We speculate that during human evolution computations that were initially reserved for particularly important ecological problems came to be used much more widely and extensively during the long period of protected immaturity.