Learning from Social Informants

The Role of Informant Knowledge in Children’s Causal Inferences
Collaborators: Sophie Bridgers, Elizabeth Seiver, Tom Griffiths, Alison Gopnik

This work looks at how children and adults combine direct observation of probabilistic data with causal predictions provided by a social informant, when making causal inferences. As part of this project, I developed a Bayesian model of causal inference from informant testimony, that takes into account both the informant’s predictions, and their claims about their level of knowledge.

Preschoolers’ Endorsement of Majority and Minority Information
Collaborators: Jane Hu, Fei Xu, Tom Griffiths

This project looks at how children reconcile differences in performance amongst multiple demonstrators. Previous work has found that in an object-labeling task, children trust novel object labels endorsed by a majority over minority labels. However, it is possible that children prefer non-dissenters as social informants only in domains that are strongly socially determined, such as language. In this study, we explore whether children endorse the majority response at the same rates in a less socially determined domain: causal learning.

Preschoolers’ Understanding of Knowledge Source
Collaborators: Jane Hu, Andrew Whalen, Fei Xu, Tom Griffiths

This project looks at how preschoolers balance the size of a majority with the quality of the informants’ knowledge source. Previous research has shown that, beginning at three years of age, children believe that direct experience is valuable. Here we look at what happens when all informants are knowledgeable, but the quality of their information source differs.

Transmission of Information: Multiple Learners and Demonstrators
Collaborators: Andrew Whalen, Tom Griffiths

This project looks at whether children and adults treat multiple demonstrators as independent pieces of data, or whether they assume that these demonstrators themselves learned from others, by comparing their performance to models capturing these two different possible assumptions.