In this talk at the LMU workshop "Concrete Causation" (9 July, 2010) Jan-Willem Romeijn (Groningen) discusses probabilistic models of experimental intervention, and shows that such models elucidate the intuition that observations following intervention are more informative than observations per se (due to technical problems about one minute of the recording is skipped)

Causation in Physics

In his study of causation J. L. Mackie once referred back to David Hume, who listed causation among one of the principles that are TO US THE CEMENT OF THE UNIVERSE and thus OF VAST CONSEQUENCE IN THE SCIENCE OF HUMAN NATURE (David Hume, AN ABSTRACT OF A “TREATISE OF HUMAN NATURE”). Yet for example the early endeavours of the developers of the Structural Equation Modelling (SEM) framework, which aimed at embedding causal meaning into the formal treatment, seem to be neglected, and David Lewis' counterfactual analysis of causation based on his possible worlds semantics does not come very handy for application. As Judea Pearl summarises: WE ARE WITNESSING ONE OF THE MOST BIZARRE CIRCLES IN THE HISTORY OF SCIENCE: CAUSALITY IN SEARCH OF A LANGUAGE AND, SIMULTANEOUSLY, THE LANGUAGE OF CAUSALITY IN SEARCH OF ITS MEANING (Judea Pearl, CAUSALITY, 2000). Borrowing mathematical rigour from statistics, one of the most prominent areas of causal modelling today sounds out the interaction of probabilistic and deterministic approaches and is centred around Bayesian Networks, through which causal notions can be identified concretely and utilised for various disciplines eventually.