| At present, most theories of human causal learning and inference slice up the world into a set of variables and the contingent relations among them. Most prominent among these theories is the causal Bayes nets framework. Many proponents of this theory argue that the strength of this model is demonstrated by its usefulness for learning from and interacting with the world. However, a careful consideration of what causal thinking does for us will lead us to discard the assumption that causal inference involves only contingency relations.;I make the case that we can perform the tasks of causal inference in ways that go beyond contingency-based models like Bayes nets. We also use representations of physical mechanisms, which allow us to make inferences that cannot be captured by contingency models. The results of six experiments demonstrate that we do form representations of physical structure and use those representations in causal reasoning. |
