Causal Imitation

Using Social Knowledge and Statistical Cues in Imitation
Collaborators: Alison Gopnik, Tom Griffiths, Pat Shafto

This line of work looks at children’s imitation of causal action sequences, examining how children decide which of the actions they see another person performing are the ones that are necessary to bring about a desirable outcome. These studies look especially at how this decision is informed by different sources of information, including contingencies between action sequences and outcomes across repeated demonstrations, information about the actor’s knowledge state and pedagogical intentions, and the actor’s direct testimony about the causal relationship.

Recent studies of imitation have shown that while children sometimes selectively reproduce the most obviously causally effective actions, at other times they will “overimitate”, reproducing apparently unnecessary parts of a causal sequence, or copying an actor’s precise behavior, when a more efficient action for accomplishing the goal is available. The evidence on children’s use of intentional and pedagogical cues to inform their imitation is similarly varied, with studies showing that in some contexts children use information about the demonstrator’s intentional and knowledge state to aid their causal inferences, while in others these cues can lead children astray

My research suggests that these different results reflect the multiple sources of information that contribute to a rational statistical inference about causally effective action sequences. My colleagues and I developed a Bayesian analysis of causal inference from repeated action sequence demonstrations, that makes behavioral predictions based on both information about statistics and about the demonstrator’s knowledge and pedagogical stance. Through a series of experiments investigating children’s imitative behavior and causal inference, we compare children’s performance to our model’s predictions.

The Role of Intentions and Physical Knowledge in the Origins of Causal Reasoning
Collaborators: Amanda Seed, Emma Tecwyn, Alison Gopnik, Tom Griffiths

How and why do the minds of different species differ? What mental abilities (if any) are unique to humans and why did they evolve? An issue of current debate is whether humans are unique in their general ability to form abstract concepts, including a notion of cause, or specifically in their aptitude for reasoning about agents and intentional action: both differences would impact causal learning from demonstration. This project, a collaboration with Dr. Amanda Seed at the University of St. Andrews, aims to investigate how non-human primates and children combine different sources of information to make causal inferences, including how differences in the intentions and pedagogical stance of a social demonstrator change their choices of which causal actions to imitate.