Explanatory Relevance

logical model into a mechanistic one, and thereby explaining rather than just summarizing empirical results. This case study was intended to show that mathematical models like the Hodgkin-Huxley equations do not provide explanations without those additional details. That what was needed for explanation in this case was more mechanistic detail does not necessarily generalize to all of neuroscience though. The examples I give in Chapter 4 will illustrate that in some cases more abstract models provide better explanations than highly detailed ones.

4_{BothSchaffner(1993) andStrevens(2004) have offered accounts of explanation that intertwined or uni-}

fied causal-mechanistic explanation with generalization, but neither is a mechanistic account in the manner of MDC.

Intuitively the problem with taking it as a general rule that adding details always im- proves explanations, is that explanations are then no longer responsive to the explanatory context. That context includes the specific question the explanation is supposed to an- swer, any background information that is being assumed, and the audience it is intended to convince. There are usually, if not always, multiple explanations available for any given phe- nomenon, all of which are correct, in that all of them account for the phenomenon’s being so. This is clearly true of explanatory texts, but I maintain it is also true of ontic explanations. These multiple alternative explanations might differ in terms of how much detail they go into, and whether they depend on particulars or on general principles, to name just two axes of difference; there may be more. I’m calling this constraint that explanations be pitched at the appropriate level given the explanatory context the norm of explanatory relevance.

An anecdote Salmon(1981, 1990,1992) relates about a helium balloon moving forward as a plane starts moving demonstrates the second sort of difference. In Salmon (1992), he uses the example to make the point that there are often several explanations of the same phenomenon, and that these are not necessarily in competition in terms of which is correct. We choose which is more appropriate based on the context in which we’re explaining. A child might better understand a story about the back of the cabin pushing the air forward, which pushes the balloon forward, because it is lighter than the air, while a university physics class might better understand an application of the equivalence principle. We might further distinguish between a generic explanation about air pushing against the balloon, which would be expected, perhaps in most airplane designs, to result in the balloon moving forward, and a specific explanation of why in this particular plane, with this particular arrangement of air molecules, and these particular acceleration forces acting, this particular ballon moved forward. Which of those is more appropriate depends partly on the question being asked. We might be asking why helium balloons on planes move forwards when the plane does, or why it happened in this particular case.

The equivalence principle provides a covering law explanation. Even if we throw out the DN apparatus because of its many faults, we can still explain some things by citing truths along the lines of ‘things of type T behave as B’ and ‘t is a thing of type T’. In some cases we might still want to ask why things of type T behave as B, but once we have an explanation

of that connection, a good explanation of the event will usually just cite the rule. We use this form of explanation even when the rule isn’t a law. To apply such a rule when there are exceptions involves a further claim that this is one of the cases where the rule applies, or that there are no interfering factors in this case.

The explanation where the back of the plane’s cabin pushed the air forward is described by Salmon as a causal-mechanical explanation. The specific causal story where the back of this plane’s cabin pushed this air forward I will call a causal narrative, by which I mean a causal story specific to a particular instance. Bogen’s insistence that mechanisms might operate only once suggests that he sees mechanistic explanation as providing causal narra- tives. In cases of one-off causal chains, it may be necessary to go into a lot of detail about the specific circumstances, because these cases are exceptional. This does not mean that in all mechanistic explanations one needs a detailed causal narrative specific to a particu- lar instance. There is always a finer grain of detail one could go into, even in the case of a mechanism that operates only once, but more detail than is required to account for the explanandum is superfluous; it makes an explanation worse, not better.

It is easier to see how this norm can be upheld in explanatory texts than in ontic ex- planations. An explanatory text can include or leave out details as required by the context, but explanations as things in the world can’t be partial in the same way, or so people seem to think. One possible response to this difficulty is to set aside explanatory relevance as being of less importance than accounting for phenomena in their entirety, including any rare or exceptional cases. In order to account for all cases, Craver claims explanations must go into the gory details. Since he takes explanations to be things in the world, perhaps he thinks that as a thing in the world, the explanation must remain the same thing in the world, regardless of the explanatory context. This is intuitively tempting. Since the various explanatory contexts might require all the weird little details, the explanation must have to include these if there is to be just one explanation. An unstated assumption, however, is that there is just one explanation for any given phenomenon, so that same explanation needs to work in any explanatory context. I think we can have multiple explanations that are all in the world, without having to have a strange metaphysics where there are multiple objects or systems in the world that supply these explanations. This is one of the constraints on

explanation that needs to be heeded if we’re to have both ontic explanation and explanatory relevance.

The downside of not upholding the norm of explanatory relevance is that the explanation for something like why a wheel rolls can’t just be that the wheel is round. The explanation would have to be something along the lines of the wheel being made of 3 billion iron atoms and 5 million copper atoms, with various forces acting between the atoms such that they form a solid, in a particular crystal structure such that only certain wavelengths of light are reflected making it appear blue, organized in a particular spatial arrangement such that it is round, and so on, and so on. I think it’s fair to say that an explanation like this fails at its job. One might assign to the explanatory text the task of picking out which parts of the explanation are relevant to the context, but this makes the explanationo _{nothing less than a}

slice of the world in all its complexity. That notion of explanation does not seem to do any useful work. It also makes one wonder why going down just to the molecular layer should be good enough for neuroscientists. If all the details that might be relevant in any explanatory context are needed, there is no bottoming-out. Another constraint on explanation then, is that it should be able to do the work of picking out which parts of the world are relevant to the phenomenon being explained.

An additional problem is that it is unclear how explanations can count as looking upwards to higher levels if they have to go into so much lower-level detail. What I’m trying to carve out space for are explanations where more abstract or higher-order features of phenomena can explain either particular phenomena, or general patterns. If higher-order features cannot do this explanatory work, then we haven’t accounted for the types of information-processing explanations from the previous chapter.

3.3.2.2 Types and Tokens Earlier in Section 3.2.4 I reviewed several commentaries