IRCS Conference Room
Rutgers University, Newark
Children's reasoning about evidence: social inferences and sampling
How do children make sense of a causally ambiguous, probabilistic world — rapidly and accurately changing beliefs about it? I will suggest that computational models can provide a precise framework for considering how children should update their beliefs as they encounter new evidence, and they also provide a starting point for considering how ideal learners should interpret ambiguous evidence. Considering how computational models connect to real learners also provides insight into the mechanisms by which child learners approach probabilistic evidence. I will present two projects. The first includes a computational model that takes into account social inferences on the part of the learner and behavioral data from preschoolers that suggest they are sensitive to social information (for better and for worse). In the second half of the talk, I present a simple algorithm for sequentially updating beliefs and show that preschoolers belief updating shows the same signature dependencies of a particular sampling algorithm: win-stay, lose-switch.