Learning from Interventions
Frederick Eberhardt, UC - Berkley

Abstract: In contrast to the traditional metaphysical approaches to causation I
will use the interventionist account of causation to ask
epistemological questions -- and answer some of them. Using the causal
Bayes net framework I will present different types of interventions
and explore how these can be used to discover causal relations among a
set of variables. I will present results of what can and cannot be
learned about causal structures under different combinations of a
variety of assumptions. Several of the results go against standard
intuition and therefore give us a better understanding of what it
really is about interventions that is useful for causal discovery.