CSLI Coglunch: Alison Gopnik, "Intervention, correlation and causal learning: Why children really ARE scientists"

Abstract: Children, like scientists, face the difficult problem of

inferring the causal structure of the world from patterns of evidence.

Scientists classically learn about causal structure by looking at

patterns of correlation (in observational studies) or at the outcomes

of interventions ( in experiments) or, most commonly, at combinations

of interventions and correlations. Recent advances in statistics,

computer science, and philosophy provide a normative account of such

inferences. The causal Bayes net formalism provides a unified

mathematical account of causation, correlation and intervention, and

shows that given a few general assumptions, a wide range of accurate

causal inferences can be made. We propose that even very young children

use similar assumptions and inferences implicitly in their everyday

causal learning. In several experiments we show that preschoolers can

draw accurate causal inferences from patterns of interventions and

correlations even when there are no spatio-temporal, mechanistic, or

associative cues to causal relations.