In general if the claim is made that X is Y this implies that there are some criteria for being Y that one can, in principle, tell if X meets. For example if we say: “that animal is a cat” then there are some features that the animal has which determine if it is or isn’t a cat. Since it seems so obvious I will take it as unproblematic. This rule of thumb, while by itself rather uninteresting, has unexpected consequences (well, for some) when applied to the causal relation. If we say that A is the cause of B then there must be some criteria that allow us to know, in principle, that A and B participate in the causal relation. That knowledge can’t come from a hypothetical “replay” of events, in which A is missing, because there is no way, even in principle, to see what would happen in such a replay. If this is our criterion for causation then causation becomes unknowable, and we would be unable to meaningfully assert that A was the cause of B.
Of course we do speak meaningfully about causation (or at least we think we do), the reason being that the criterion for the existence of a causal relation is not the hypothetical “replay”. Instead we judge that two events participate in a causal relation when, in general, events like A, in some specific way, are followed by events like B, in some specific way, and in general situations like the one presently under consideration without A aren’t followed by B. (Yes, I will provide more details below.) Such a definition of causation does allow us to speak meaningfully, and reliably, about it, since causation is described in terms of what happens (or what has happened) in general in other similar situations, which we can have knowledge of. It also makes the causal relation between two specific events dependent on regularities in general, and thus it makes no sense to say that A is the cause of B in a case where A causing B is a single case (i.e. a special case of causation and not an instance of a regular pattern of events).
Now allow me to return to fill in a bit more of the specifics concerning the criterion for knowing one event to be the cause of another. To begin with, we can’t say that one specific event is the cause of another specific event, since specific events aren’t the kind of thing that can happen repeatedly. What we need to say then is that some features of the first situation caused some features of the subsequent situation. By describing causation between “features” we allow for the possibility that other situations also have these features, giving us the generality required to know that the causal relation holds (assuming that we have reason to believe that a general regularity between situations with those features holds). The second detail I need to fill in is probabilistic causation, specifically where A doesn’t always bring about B, but it makes B more likely. In this case we would say that A causes B if in general situations meeting the A-criteria are followed by situations meeting the B-criteria more often than situations the don’t meet the A-criteria. It is just this sort of probabilistic causation that we are normally acquainted with, since rarely is it the case that one kind of event is always followed another, at least in the macroscopic world. Finally, let me describe how the requirements for two events to stand in a causal relation can deal with the possibility of “causal defeaters”. The best way to define a causal defeater is by giving an example. Consider then the causal relation between “falling from a great height” and “death”. In this situation the possession of a parachute can be considered a causal defeater. Because some people may have parachutes are we unable to say that falling from a great height causes death, must we instead say that falling from a great height and lacking a parachute causes death? If we did have to qualify our causal assertions in this way it would seem to invalidate this criterion for the causal relation, since in any situation there are a vast number of possible causal defeaters. The best way to tackle this problem is to assume that every causal assertion has a ceteris paribus clause attached (all things being equal). The formal way to deal with such clauses is through conditional logics, which I won’t detail here, but I mention them to assure the reader that the existence of causal defeaters don’t undermine the whole project, they simply force us to break out the complex mathematics.
To wrap things up I should probably mention how the account of the causal relation I have given here fits with the formalization of causation I presented earlier, here. At first glance it may appear as though these accounts are in opposition to each other, since in my previous account I based by formalization on counterfactuals, and counterfactuals can be understood as just the kind of “replays” that I claim can’t be the basis of causal knowledge. This would be a misreading, however (although a justifiable one, since I didn’t clarify my original work properly). I would still stand by the formalization based on counterfactuals I gave there, with the understanding that we can only have knowledge about the counterfactual, knowledge about what would have happened if we took away some events, from our knowledge of general regularities in the succession of events.