The Cognitive Processing of Causal Knowledge
This work connects computational models to human cognition, potentially informing AI and psychology, but it is incremental as it builds on existing theories without new empirical data.
The paper argues that Bayesian networks model human causal reasoning and learning, showing that human judgments in discounting studies match Bayesian network inference algorithms and that human learning aligns with structure-learning algorithms.
There is a brief description of the probabilistic causal graph model for representing, reasoning with, and learning causal structure using Bayesian networks. It is then argued that this model is closely related to how humans reason with and learn causal structure. It is shown that studies in psychology on discounting (reasoning concerning how the presence of one cause of an effect makes another cause less probable) support the hypothesis that humans reach the same judgments as algorithms for doing inference in Bayesian networks. Next, it is shown how studies by Piaget indicate that humans learn causal structure by observing the same independencies and dependencies as those used by certain algorithms for learning the structure of a Bayesian network. Based on this indication, a subjective definition of causality is forwarded. Finally, methods for further testing the accuracy of these claims are discussed.