Probabilistic Causal Structure

Abstract

The concept of causation is critical to our understanding of how we can understand our world, because causal relations just are the rules by which our world operates. Bertrand Russell supposed early in this century that the concept could be done away with since the product of science appeared in every case to be no more nor less than a set of functional relations — an axiomatized theory in which no primitive predicate of the form ‘Cause(X, Y)’ appears. He later abandoned this view and with excellent reason: even if causation does not figure as a subject matter of any science,² it figures essentially as a predicate in the metatheory of science and particularly in any plausible accounting of how scientists may come to learn theories that are true of the world. The unhappy consequence of this fact is that in order to come to an understanding of epistemology — how anyone can learn anything about the world in which one finds oneself — we must come to some understanding of the nature of causation, an understanding which has proved highly elusive. Here I shall review some of the recent philosophical literature on the subject of causation, with special emphasis on what appears to be clearly the most promising avenue of analysis: the probabilistic conception of causality. In the end I will offer a criterion of causality which improves upon all those reviewed and which may serve to clarify the role of controlled experimentation in science and which may also serve in attempts to automate causal reasoning within an artificial intelligence and, in particular, to automate reasoning to causal models from data (causal inference) and reasoning with causal models during prediction or planning.

“The word ‘cause’ is so inextricably bound up with misleading associations as to make its complete extrusion from the philosophical vocabulary desirable.”—B. Russell (1912)

I thank Wesley Salmon, Gurol Irzik, Michael Dickson, Chris Wallace, James Forbes and Honghua Dai for discussions helpful in the writing of this paper. I also thank Howard Sankey for organizing the symposium on laws and causation at the 1996 Australasian Association for the History, Philosophy and Social Studies of Science, where this paper was presented. Finally, I especially thank Linda Wessels for her seminar on the subject.