Proceedings of the Annual Meeting of the Cognitive Science Society

We build a computational model of how humans actively infer hidden rules by doing experiments. The basic principles behind the model is that, even if the rule is deterministic, the learner considers a broader space of fuzzy probabilistic rules, which it represents in natural language, and updates its hypotheses online after each experiment according to approximately Bayesian principles. In the same framework we also model experiment design according to information-theoretic criteria. We find that the combination of these three principles -- explicit hypotheses, probabilistic rules, and online updates -- can explain human performance on a Zendo-style task, and that removing any of these components leaves the model unable to account for the data.