Starting at the end of a book is typically a bad idea, but in this case I think it’s enlightening to begin at the end, as it were. For one, Pearl makes some of his most dramatic claims and I want to draw attention to those claims and treat them as promissory notes to be cashed in by the final substantive chapter. I also want to ensure that these claims are capable of holding up to philosophical scrutiny.
My (newest) position concerning causality is one that I’m fairly confident Pearl shares, which is good because, for one, it’s an unpopular position and, two, it’s always nice to have friends. Lately, I’ve been entertaining the notion of an “epistemic” conception of causality, which means that causation need not be thought of so much as a feature of the world but as a feature of the way we interpret the world. These are very different interpretations of causation since one implies that causal relations are things in the world (whether we observe them or not) and the other implies that causal relations are items of our understanding of the world (which may not be there if we’re not here to observe them). I think Pearl holds this position and I’ll give you some reasons to back up my claim.
First, when discussing the problem Bertrand Russell has regarding the problem of causation, Pearl is quick to point out that the asymmetry of causal hypotheses is not due to the structure of the universe as such, but is instead due to the ways in which researchers go about studying some small part of any physical system. Pearl says:
The lesson is that it is the way we carve up the universe that determines the directionality we associate with cause and effect. Such carving is tacitly assumed in every scientific investigation (p.420).
Now, while not a definitive acceptance of this epistemic conception of causality, it is still telling, I think. Pearl is responding to Russell’s criticism that in advanced physics causal language never comes up by arguing that the language of causality only comes into the picture when researchers focus on areas of the physical universe that are not maximally inclusive. In other words, it is only in isolated subsystems that causality appears: take a subset of the interrelations between physical objects and you will find causal relations; take the system as a whole and those causal relations disappear and the system seems symmetric.
There is something very interesting going on here with the related notions of “boundary conditions,” “input/output relations,” and “background conditions.” When we first isolate some subsystem and then characterize it in terms of inputs and outputs we draw a boundary around some phenomenon and, in so doing, create this asymmetry between what was designated as the input and the output. This idea is worth exploring further and, I’m sure, will become a much-debated item of discussion on this blog. It’s also in my dissertation, so there’s that.
Now, to be fair to the realist and a possible realist-reading of Pearl, declaring something to be an input does not mean that one must be an anti-realist regarding the nature of that element (inputs can be just as real as anything, after all). Now whether Pearl is or is not a realist about causation is going to be one topic of this multi-part discussion, but I am going to try to defend this rather extreme position I’ve outlined above and argue that, in fact, it is the labeling of certain phenomena as “input/output relations” that provides the directionality and the asymmetry of our causal hypotheses about those phenomena. So I agree with Pearl when he says that it is the act of investigation of an aspect of the universe that brings about causal language and I’m going to take him literally and defend the position that causation is just that; namely, the way in which we characterize systems in terms of asymmetric relations between elements of that system.
Like I said above, this is a highly unpopular and idiosyncratic position to hold, so I’m very interested in hearing your views.