Doing after seeing


Causal knowledge serves two functions: it allows us to predict future events on the basis of observations and to plan actions. Although associative learning theories traditionally differentiate between learning based on observations (classical conditioning) and learning based on the outcomes of actions (instrumental conditioning), they fail to express the common basis of these two modes of accessing causal knowledge. In contrast, the theory of causal Bayes nets captures the distinction between observations (seeing) and interventions (doing), and provides mechanisms for predicting the outcomes of hypothetical interventions from observational data. In two experiments, in which participants acquired observational knowledge in a trial-by-trial learning procedure, the adequacy of causal Bayes nets as models of human learning was examined. To test the robustness of learners’ competency, the experiments varied the temporal order in which the causal events were presented (predictive vs. diagnostic). The results support the theory of causal Bayes nets but also show that conflicting temporal information can impair performance.

In B.G. Bara, L. Barsalou, & M. Bucciarelli (Eds.), Proceedings of the 27th Annual Conference of the Cognitive Science Society (pp. 1461–1466). Mahwah, NJ: Erlbaum