A computational characterization of domain-based causal reasoning development in children

Abstract

To better understand human intelligence, we must first understand how humans use and learn from stories. One important aspect of how humans learn from stories is our ability to reason about cause and eect. Psychological evidence suggests that when children develop the ability to learn cause-and-eect relationships from stories, they do so in discrete stages where each new stage enables the child to incorporate new kinds of information. In this thesis, I attempt to shed light on the mechanisms that underlie the development of causal reasoning in children. I create a behavior-level model, an explanatory theory, and an explanation-level model that account for the developmental stages. I implement these models on top of the Genesis Story Understanding System. The result is a psychologically plausible explanation-level model that captures the observed causal reasoning behaviors of children at dierent stages of developments. The model also takes the observations from psychological evidence to another level by proposing mechanisms that enable such development in children.