How can irregular causal generalizations guide practice?

In this essay, I shall be concerned with the sort of "irregular" causal generalizations that epidemiologists aim to discover. Examples include "Smoking causes lung cancer," "HIV viruses cause AIDs," or "Low dose aspirin prevents strokes." What do these claims mean and how should they guide action? I argue that philosophers have mistakenly believed that answering these questions calls for a theory of what it is for one event to be a probabilistic cause of another. But this theory fails to provide plausible truth conditions for irregular causal generalizations. Indeed, it implies that most of them are false. I sketch a more plausible view, but point out the risks one takes in relying on these generalizations to guide action.