Perimeter Institute for Theoretical Physics
Tools for causal inference
Scientific explanation often requires one to make inferences about what is unobservable from what is observable. Perhaps the most important example of this pattern is to make inferences about causal mechanisms from observed correlations. This problem is known as "causal inference" and has applications in any scientific field that uses statistical data. Mathematical techniques for solving it, therefore, should be considered to have a status similar to that of standard tools of statistical analysis. Many powerful techniques for causal inference have been developed by researchers in the field of machine learning, but they remain less well-known than they ought to be. It has recently become apparent that a problem that is of central interest to researchers studying the foundations of quantum theory---that of deriving Bell inequalities---can be understood as an instance of causal inference. In this talk, I will describe some of the techniques for causal inference that have been inspired by the foundations of quantum theory and the sorts of applications that these might find in physics.