Have you ever noticed that the word “cause” has different meanings in the present, conditional, and past tenses? I have been discussing Ben-Menahem’s suggestion that causality is a cluster concept and that that might rescue it from Norton’s argument that causality is a folk science. Norton’s argument is that causality is too “plastic” a notion to be of practical use for judging other phenomena. And now, in providing my own take on that issue, I find it informative to discuss an example of that plasticity. That is how the word “causality” has generic, conditional, and specific usages. And one way to make sense of that might be if causality is a cluster kind.
The word “causality” can imply, in the present tense, what is always the case (“heat causes ice to melt”), or it can be conditional (“sneezing on someone might cause a disease to be spread”), or in the past tense it can be limited to a specific instance (“the fall caused me to scrape my arm”). The advantage of the first (that causality by definition—as per Hume—is only about what is always, or at least regularly, the case, like a law) is that it can be used syllogistically and that it can even be made to make it seem that science is the study of causality (the study of regularly occurring events). Taking Hume’s approach enables people to talk of causality prospectively (“If you do that, it will cause the ice to melt”), and so there is great utility in that.
On the other hand, taking such an approach misses how the other usages of causality can be very real and important. Recognizing conditional causality (although it sounds like an oxymoron) can confer survival benefits by serving as a warning (“If you do that, it could cause an accident”). But such a usage does not make for definite syllogistic conclusions. Instead, it might lead us to think of causality as the “probability” that one event will give rise to another event.
Yet definite conclusions are possible when considering specific descriptions of causality in the past tense. That is because such considerations are usually done retrospectively (“ice on the road caused the accident”), rather than prospectively. Instead of deducing from a general statement (such as “heat melts ice”), or instead of speaking of probabilities, we can trace the events that actually happened in the past, how one event led causally to the next. That can even enable us to see a series of “accidents” as acting like “causes.” But that can also give the impression that we ought to be able to work forwards in time, rather than backwards, and extrapolate the process of tracing events to draw conclusions about the future. Yet such impressions of definiteness might not be justified in all such extrapolations. (The future might bring unforeseen circumstances, or if the past cause was an accident, that cannot be extrapolated in a definite way).
So when it comes to the original question of how to define causality—and why that has proven difficult over centuries of trying—we can see how that is complicated by how there are at least three different usages of the word depending on the tense of the verb (causes, could cause, or caused). It can imply using syllogisms, probabilities, or tracing causal chains. Thus any general statement about causality can be confounded by how it must apply to all three usages, or by how it might work in one case but not another. For instance, are we justified in syllogistically making causal inferences, or not? And when can we call an accident a cause? (Or is there no such thing as chance in the world, as Hume argued, because everything has a cause?). We can get different answers depending on the tense of the verb.
That is where there is potentially great value in seeing causality as a cluster kind. In other words, causality can be viewed as a set of various notions not all of which have to apply in every case. Then the point is that, to proceed with an argument, we have to clarify which usage we are now employing. We need to keep in mind how some usages enable, for instance, syllogistic thinking, or the extrapolation of causal histories, and other usages do not. And we can find justification for taking such an approach (instead of thinking there is just one causality that applies in all cases) by understanding causality as a cluster kind. Furthermore, cluster kinds can serve as a manner of bookkeeping to keep straight the various usages. It can be as simple as cautioning ourselves not to mix our tenses when making a causal argument.
There is also precedence for taking such an approach in Bohr’s principle of complementarity. That is where light can be viewed as either a particle or a wave, as long as we do not mix the elements of the two methods together. Likewise, we need to keep straight which manner of causality we are talking about and not mix them, since that can lead to confusion and to false conclusions. (An example of doing it wrong is when we identify “risk factors” probabilistically and then treat the risk factors as if they were generic causes like heat melting ice). All three usages of causality can be valid, but we have to keep them distinct.
In other words, if causality is a cluster kind, that does not mean that it is diminished and that we cannot use it to make causal inferences. Just the opposite, it means that there will be occasions when we are justified in doing so. It is just that we are only justified in making causal inferences under certain usages, which we have to stipulate.
So the bottom line is that we can take Norton’s point that causality is too plastic a notion to be unequivocally reliable, and yet we can still point to the special usages for which it is reliable.
And a good way to keep that straight (how causality can have generic, conditional, and specific usages, each with different qualities) is to view it as a cluster kind, which Ben-Menahem has suggested for her own reasons in surveying the large number of unsatisfactory historical attempts to give causality a single definition.
Even if we are wary of the realist overtones of cluster kinds, we can still think of causality as having its own principle of complementarity. We must keep our usages distinct.
And of course, there are other ways of being plastic besides with the tense of verbs, and there are other reasons why science restricts its use of causality besides plasticity. But we can still point to those usages for which causality is reliable and confine our arguments to the parameters set by those usages. That would be to address the problem of plasticity with causality’s own principle of complementarity, which we can do either with or without also seeing it as about cluster kinds.
Definitions and citations were introduced in the first post, “Is Causality a Cluster Kind?”
Ben-Menahem, Y. 2018. Causation in Science. Princeton, NJ: Princeton.
Norton, J. 2003. “Causation as Folk Science.” Philosopher’s Imprint 3 (4): 1-22.
Slater, M. 2014. “Natural Kindness.” The British Journal for the Philosophy of Science 66 (2): 375-411.