Backwards Explanation C.S. Jenkins and Daniel Nolan

At first glance, ‘backwards’ explanation is ubiquitous. For our purposes, a backwards explanation of a token event e1 which occurs at time t1 is an explanation in which another token event e2 which occurs at some later time t2 plays a key role.1 We shall only look at backwards explanation of particular events here, though for all we say there may also be backwards explanations of types of events, facts, states, and other things. Here are some candidate backwards explanations of events:

1. I am tidying my flat today because my brother is coming to visit tomorrow. 2. The scarlet pimpernels closed up because it was about to rain. 3. The volcano is smoking because it is going to erupt soon. 4. Sam is upset because his mother will die within the next month. 5. The magnets are moving faster now because they are about to touch each other. 6. Dr Bostrom appeared in Framley in 1905 because his time travel experiment in 2005 was successful. 7. The acorn has sprouted here because the mature oak tree will need sunlight.

Unless the many prima facie cases of backwards explanation can be discounted for some principled reason, their existence is an important piece of philosophical data which any acceptable theory of explanation should be able to accommodate.

Not everybody agrees. Sheffler (1957, p. 300) says explicitly that “[f]or explanation [of b by a], we require ... that a must not temporally follow b”. He argues that:

1

since no temporal criteria are to be found among [Hempel and Oppenheim’s] four requirements, they define a wider class of inferences than simply explanatory ones. A non-explanatory instance which fits the pattern is afforded by any case where b precedes a.

In a similar vein, here is Strevens (2004, §2.2):

The principal motivation for the causal approach is the recognition ... that the direction of explanation runs parallel to the direction of causation. The height of a flagpole explains the length of its shadow, but the length of the shadow does not explain the flagpole’s height. This explanatory asymmetry mirrors a causal asymmetry.

Though such explicit statements are comparatively rare, many share the intuition that there is no backwards explanation. In our experience, philosophers are often keen to explain away apparent cases of backwards explanation by arguing that in these cases ‘the real’ explanation is something else. We shall argue, however, that such attempts are unmotivated in full generality.

Apart from its intrinsic interest, the existence of backwards explanations which cannot be explained away is an important constraint on philosophical theories of explanation which has not always been appreciated. If we are right about backwards explanation, then various extant theories of explanation have a challenge on their hands, or at least have to recognize a class of explanations which their accounts as they stand do not handle. For instance, there are some theories according to which explanations of an event e are events in e’s causal history.2

Some theories of explanation seem well placed to allow for backwards explanation. For example, unificationist approaches (see e.g. Friedman 1974, Kitcher 1981, 1989) and familiarity approaches (see e.g. Bridgman 1968, p. 37, Dray 1964, pp. 79-80) face no obvious difficulty if it turns out that there are backwards explanations. The same goes for deductive-nomological approaches.3 Our 2

purpose here, however, is not to argue in favour of these approaches at the expense of those mentioned in the previous paragraph. Sophisticated versions of the latter group of views may be able to accommodate the data we are presenting (see our discussion on pp. 17-18 below). Our point is merely that this data exists, and should be taken seriously in philosophical theorizing about explanation. To emphasize its importance, we shall mention at the end of the paper a few places where it does not seem that enough attention has been paid to backwards explanation, and the theories on offer are less attractive as a result.

We would like to focus here on a number of central cases which we think are particularly hard for theorists to discount. Cases that appeal to exotic metaphysical possibilities, while interesting, are not our concern here. So for example, time travel cases, like our example six, will be put to one side. We also want to put to one side any other cases of backwards causation there might be. Some cases of apparent backwards explanation might be thought to involve primitive teleology: the postulation of irreducible purposes in nature (see Hawthorne and Nolan 2006). Even if that is a coherent possibility, it is too far from the usual contemporary way of looking at things to be relevant for our purposes. Likewise with postulated “advanced action”, or lawful correlations of future states with past ones that seems to allow the future to constrain the past in ways that suggest explanations of past or present events by future ones. This may be ubiquitous, if for example Huw Price’s explanation of quantum phenomena like EPR results in terms of advanced action is correct (Price 1996, especially chapter 9). But we do not want to have to rely on any of these metaphysical speculations. Those who defend accounts of explanation which cannot easily account for backwards explanation might, without too great a loss of plausibility, make special exceptions for these cases, or may deny that our intuitions about them are to be relied upon, or may even deny that we have any strong-enough intuitions at all about these cases. Backwards explanation appears to arise in much more mundane areas of our understanding of the world.

3

We begin, therefore, by thinking about case 1 on our list. At first glance, citing my brother’s future visit is a perfectly good way of explaining my tidying behaviour today. In uttering 1 I have said enough to enable other people to understand why I’m tidying my flat. Nonetheless, you might feel tempted to say that ‘the real’ explanation of my behaviour in this case cites not the future event but rather my belief in it, which predates the tidying behaviour it explains. If so, this is not really a case of backwards explanation.

It is certainly true that citing my belief that my brother is coming to visit tomorrow is one way of explaining why I am tidying my flat today, in many cases.4 But there are generally various different explanations of a single explanandum; this does not mean we need to regard one of them as ‘the real’ explanation and the rest as non-explanations. For example, one explanation of the outbreak of a fire can cite the arsonist’s striking of a match, while another cites her desire to start a fire and yet another cites both these things. It may be that a particular one of these explanations is best in any given particular context or for any particular audience. But it does not follow that the others are not explanations; some of them might even be better explanations in other contexts or for other possible audiences. It would be a mistake to pronounce one of these ‘the real’ explanation in such a way as to imply that the others are not really explanations at all.

It may sometimes be that one explanation x of event e is true, or is a good explanation of e, in virtue of the obtaining of another explanation y of the same event e, and this might be a source of motivation for the claim that ‘the real’ explanation is y rather than x. There are a variety of things that this kind of ‘in virtue of’ claim might amount to. Perhaps y is somehow more fundamental than x; for instance, it may be felt that citing my hunger explains my moving towards the fridge, but that an explanation of the same event in terms of the state of the microphysical particles which comprise

4

me is more fundamental. Or it might be that x and y both explain e but y also explains x.5 Or it might be that the information given by x is properly included in that given by y.6 But none of these motivations for describing y as ‘the real’ explanation of e casts any doubt on the claim that x is also a respectable explanation of e.

Other reasons for regarding one explanation y of an event e as ‘the real’ explanation of e, despite the existence of other candidates x, may include the thought that y is the explanation that best focuses on the things we are – or should be – most interested in right now, or that y provides a larger quantity of relevant information than x (but see footnote 5). Again, however, neither of these provides any motivation for doubting that x is an explanation of e (at least in other contexts and/or for other people).

Perhaps it will be felt that intentional behaviours such as tidying one’s flat can never be properly explained except by citing the propositional attitudes of the agent. This seems false to us; we think people sometimes dive into rivers because children are drowning, leave the room because they’ve been asked to, and buy new shoes because their old ones are worn through.

A closely related type of backwards explanation appears to occur when we are dealing with artifacts. The washing machine started to drain water because it was about to spin the clothes. The submarine was watertight because it was going to be submerged under water. The smoke alarm beeped because the battery was about to go flat. Explaining the current action of the washing machine by the future spin cycle or the submarine’s properties in terms of its future contact with water seem like cases of backwards explanation. What seems to be underwriting these explanations are the artifacts’ functions. And what an artifact is for, in turn, seems to be a matter of human intentions, though perhaps in a complicated way. (The actual maker may be sitting in a sweatshop

5

with no idea what the device is for, the inventor’s intention may be superceded by future adaptations, the user may use an artifact with an intention irrelevant to its function, and so on.) If human intentions can give rise to backwards explanations in the flat-tidying case, presumably they can give rise to backwards explanations when an artifact is functioning as it is supposed to. There are many interesting questions that arise about explanations in terms of function, and we shall deal with some more of them in the section on biological function.

What matters for our purposes at this point in the paper is that one might be tempted to think backwards explanations that are underwritten, directly or indirectly, by the intentions of agents, are a special class. Even if we admit them, and admit that they are common, perhaps one could try to quarantine these as a special case and insist that, apart from these, explanations can never be backwards. We think this cordoning-off strategy is a mistake. For one thing, acknowledging the existence of some backwards explanations puts pressure on the claim that all the other apparent cases ought to be explained away. For another, there are many prima facie cases of non-intentional backwards explanation, to which we will now turn.

Biological Function

Consider example 2 on our list. The scarlet pimpernel is a wild plant with a small red flower which often closes up when it is going to rain. (This protects the pollen in the flower.) It seems possible, in certain situations, to explain why a scarlet pimpernel flower is closing by saying that it is about to rain: someone unfamiliar with the pimpernel might feel satisfied just with that bare explanation if they are observing the flower. Or consider someone concerned about their new scarlet pimpernel plant, worried that perhaps something has gone wrong. They can surely be reassured by being told “your flower is only closing because it is about to rain”. Again, some will feel the temptation to 6

deny that the claim about the future rain is literally part of the explanation even in these cases: they may feel that in this case ‘the real’ explanation is the forwards explanation citing the fact that it became darker just before the flower closed up, or some other specific kind of forwards explanation.7

One problem with many of these proposals is that an explainer might not have the relevant forwards explanations available. A horticulturalist might know that scarlet pimpernels always close up before rain without having any idea whether it is increasing moisture levels, or decreasing barometric pressure, or registering loss of light due to stormclouds, or whatever, that they are proximally responding to. (In fact, we understand it is sensitivity to light.) A hearer might not be able to work out any particular current or past cause to fill in this slot either. Of course, one might say that the real explanation is just unknown in these cases, but it would be odd that an apparently successful act of explaining occurred even though both parties were ignorant of anything that counted as a ‘real’ explanation. Or one could postulate a very general current or past fact about the pimpernels to explain it: that the petals were closed by some-mechanism-or-other which is such that it tends to be triggered before rain. To insist something like this is the ‘real’ explanation – the one the explainer meant all along – is of course possible, but not immediately appealing. (One thing which the original explanation provides but this proposed replacement does not is the information that it is about to rain.)

The second problem, already mentioned above, is that the fact that there are good explanations of a pimpernel’s flower closing that cite only past and present conditions does not preclude there being another explanation in terms of the future rain. Somebody wanting to argue that all the ‘real’ explanations invoke past or current conditions, and perhaps even that all of them are silent about future conditions, cannot rest content with identifying some explanation of the event that fits her

7

scheme - she needs to tell us why our backwards explanation is not one of the ‘real’ explanations.

The pimpernel case is not one where we explain via propositional attitudes - it is not as if the pimpernel believes it is going to rain, or rationally acts in the anticipation of rain. But perhaps it could be argued that the pimpernel case is part of a rather limited class of cases - cases where an explanation of behaviour can be given in terms of biological function. Furthermore, it is often claimed that biological function is itself a matter of how things went in the past, in the selectional history of an organism (see Millikan 1984, 1989 and Neander 1991a and 1991b). On this view, a biological system has a proper function when it has a property with a certain history. Roughly, the property must have contributed to an ancestral organism’s fitness, and its function is to do what it did in order to contribute to that fitness. For example, a cockatoo’s wing is for flight because the property of having a wing caused ancestors to be more fit for survival and reproduction, and did so by being causally responsible for ancestral cockatoo flight. Presumably the pimpernel has a system with the function of closing it before rain, because pimpernel ancestors caused to close before rain thereby did better than they would have without that property. Today’s pimpernels close before rain because of natural selection of many generations of pimpernels in the past.

All of those closures, patterns of damage and lack of damage by rain, and so on, cause today’s pimpernels to close before rain. How does this affect the issue of backwards explanation? Here are three alternatives that could be tried.

The first is that this account of proper function shows that there is no backwards explanation going on when we are told that a scarlet pimpernel closed because it was going to rain. Instead, we are referred to the biological function the pimpernel has to close when it rains. And that, in turn, is just to have the pimpernel’s current behaviour explained by a certain feature of the selectional history of

8

pimpernels, which is all entirely in the past. There is no backward explanation after all. Or so this alternative holds.

We do not think this first attempt is an appealing diagnosis of the case. For one thing, it is not clear how “its flowers are closing because it is going to rain” could be referring to a function - no function is explicitly mentioned, and the only things explicitly mentioned are things like the flowers and the rain. Furthermore, even if the function is invoked in that claim, the further claim that the grounds of that function are what are talked about is even stranger. A horticulturalist might be perfectly reasonable in telling us that the closure is due to immanent rain without having the foggiest notion of how to spell that out in terms of selectional histories of traits - not every botanist has read their Neander and Millikan, after all. Postulated explanations that are supposed to be given by a speaker to a hearer, even though they seem unavailable to either, are uncomfortable things to be stuck with. Furthermore, even if we are told about the sweeping history of pimpernel evolution, it looks like the explanation in the particular case needs something else, since pimpernels with the same evolutionary history often do not close (because the weather is set to stay fine). And what else is needed seems to be the fact that it is about to rain. But if we need that future fact for an adequate explanation, as well as a selectional history, then it looks like the explanation is still backwards – a key role is still played by a future event – so we have not made much progress.

A second attempt to use the notion of biological function does not require that the ‘real’ explanation being offered is one in terms of a past selectional history, even though it may be agreed that the past selectional history is part of the truth-maker for the function claim. Rather, we are explaining the pimpernel’s behaviour in terms of a current property it has - the functional property of closing before rain. It has that function now, and we could claim that citing this functional property is what provides the ‘real’ explanation of the closure. This second attempt avoids some of the problems of

9

the first - while a botanist may not implicitly have a selectional history story in the background, it is more plausible that they might realise that the pimpernels have a biological function. Maybe they wouldn’t put it in such high-falutin language, but if they are prepared to admit the closing is for protecting the pollen, an implicit belief in that much teleology might be reasonable to attribute. And in mentioning the function, and its activation, we might hope to avoid referring to the future rain altogether. But it still faces problems. There is still the problem that the original explanation did not seem to refer to a function. And there is the problem that in explaining the activation of the pimpernel’s function (to distinguish it from plenty of unclosed pimpernels with that function not being activated), we might need to mention the fact that it is about to rain. (Replacing this with mention of other factors, such as the reduction in light, raises the same problems as before: the botanist might not know that this is what the pimpernel is responding to.)

The third way that proper functions could be appealed to in response to the pimpernel case is to admit that there is a good backwards explanation in this case - the rain’s being about to arrive does explain the pimpernel’s closing. The appeal to proper function then shows us that this explanation is not mysterious - not requiring backwards causation, primitive teleology, or any other such metaphysical extravagance. This would be to concede our point that there is backwards explanation, of course, but someone in favour of it might also use it to narrowly restrict the range of backwards explanations. Suppose someone held that there were good backwards explanations, but they were only to be had when backed by a proper function account or an account that made appeal to intentionality, to cover the cases of plans and intentions covered before, as well as to cover the teleology associated with artifacts. Perhaps functional explanation is one of the fundamental genuses of explanation (as Aristotle thought), and the rule against backwards explanations only applies to non-functional ones. Is it plausible to restrict the domain of backwards explanation to only these cases (and, perhaps, the intentional cases)?

10

We think not. The first reason to think not is that backwards explanations seem available in teleological cases that are not explained by appeal to proper function, or to designer’s (or maker’s, or user’s) intentions. Consider the case of malfunctioning systems. For example, suppose there is an alarm system that was designed to detect authorised and unauthorised personnel entering an area – perhaps by detecting a signal emitted by an electronic tag that comes in authorisation documents, perhaps by face recognition, or whatever. The system was designed to set off an alarm whenever unauthorised people are about to enter the area. (So when it is working properly, backwards explanations of the form: “The alarm went off because an unauthorised person was going to enter the area” are in order.) Unfortunately, something went wrong because of a random accident, and now the system stays quiet when unauthorised people are about to enter the area, but sets off the alarm when authorised people are about to enter the area. Once we are aware of this malfunction, it seems just as good an explanation of why the alarm goes off on a given occasion to say that the alarm went off because an authorised person was going to enter the area. That would be a backwards explanation (the current sounding of the alarm is explained by the authorized person’s future entry), but now an appeal to the function given to it by its designer will not back that explanation up, since no designer intended the alarm to go off before authorised people entered. And likewise for any function given to it by the installers, users, and so on, since the malfunction might be entirely against what anyone intended or wanted.

Similarly with a biological case. A toad might be designed to jump for safety when a predator is about to get too close (before the predator gets within a certain distance). Now suppose the toad is malfunctioning, maybe due to neural damage, or eating a hallucinogen, or whatever, so that it reacts to toads of its species as normal toads do to predators. Once we know this, we can explain the toad’s rapid movement away by saying it started to move away because another toad was going to

11

become too close to it. But the toad was not selected for a trait of running away from approaching toads - it is not exercising the relevant proper function, but rather a malfunction. Examples can be multiplied - malfunctions can provide the basis of backwards explanations as well as proper functions underwritten by selectional histories or artificial functions underwritten by the intentions of designers.

The second reason to think that backwards explanation is not only limited to cases where we can appeal to function and/or intentionality is that there are backwards explanations for things in the natural world that do not seem to involve functions or intentionality. A planet which is slowing down because it is going to reach its apogee soon does not have the function of orbiting its sun, at least not in the biological or intentional sense. A volcano which is smoking because it is about to erupt does not have the function of erupting or of smoking before eruption. We will now consider these non-teleological cases.

Non-Teleological Cases

A large class of non-intentional, non-teleological cases is associated with the existence of regular patterns in nature. Smoke regularly appears above volcanos when they are about to erupt. Planets generally slow down when they are close to their apogee. Magnets move together faster when they are about to touch each other. Correspondingly, it looks like we can explain certain events by citing the later occurrence of an event of the type with which the explanans is regularly linked. (We are not claiming that all regularities correspond to apparently good backwards explanations, but merely that many do.) Such cases can be generated in the absence of teleology and intentionality.

In these cases, there will generally (perhaps always) be some explanations of the explanandum 12

events which refer only to earlier events. But that doesn’t mean that ‘the real’ explanations in these cases are ones which refer only to earlier events. We can generally (perhaps always) find several different explanations of the same explanans; but as we have already argued above, that doesn’t mean one of these is ‘the real’ explanation and the others are not really explanations at all.

We do not propose to offer an account of why backwards explanations can work in these cases where there is a corresponding natural regularity. However, we do have a couple of suggestions about the way such explanations might often work. We can illustrate these with the volcano case. When we say that the volcano is smoking because it is about to erupt, we give information that rules out other possible explanations of the smoking - for instance, that someone has lit a fire in the crater, or that it has just been struck by lightning. Some explanations seem to work primarily by excluding other rival explanations. (For instance: “Why did John send me this email about his holidays?” “It was a mistake.”) So even if this is all backwards explanations of this kind can ever do, they may to be able to do enough to count as explanations.

But they often seem to be able to do more than that. Maybe part of what we’re conveying with the backwards explanation of why the volcano is smoking is that volcanic eruptions tend to follow a certain pattern, which includes pre-eruption smoking, and that pattern is what’s in progress now. That certain backwards explanations convey this kind of information plausibly goes some way towards accounting for the fact that they appear to be good explanations.

This suggestion gives rise to a certain kind of response to the main claim of this paper, however. The response we have in mind says that there aren’t really any backwards explanations; what a ‘backwards’ explanation does is draw our attention to a certain regular pattern of events in nature (which exists concurrently with the explanandum event). The ‘real’ explanation is that this pattern

13

exists.

One nice feature of this response is that, unlike the divide-and-conquer strategies that might be applied to particular backwards explanations (the intentional and teleological ones), this strategy might, at least at first blush, seem to be available for all kinds of backwards explanations (excepting, perhaps, recherché cases such as those involving backwards causation, if indeed there are any such). People regularly tidy up when they have visitors coming; scarlet pimpernels regularly close up before rain; the malfunctioning alarm regularly goes off when an authorized person is about to enter the building. Offering general reasons for doubting that backwards explanations need to be taken seriously would seem much less ad hoc than offering a different reason for each category of backwards explanations.

This strategy faces a difficulty similar to one discussed in the previous section. Even if we are told about the relevant regularity in the volcano case, it looks like the explanation still needs to mention something else, since volcanos to which the same regularity facts pertain often do not smoke (because they are not about to erupt). And what else is needed seems to be the fact that the volcano we’re interested in is about to erupt. But if we need to rely on this key future fact to render the explanation adequate, then the explanation is still backwards.

In addition, despite appearances, it may not in fact be able to deal with all the cases. Some backwards explanations do not seem to be trading on the existence of regularities or tendencies. It may be that volcanoes very rarely smoke before they erupt, but nonetheless the fact that this one is about to erupt explains why it is smoking on this occasion.

In an attempt to avoid these kinds of problems, someone might argue that what apparently-

14

backwards explanations do is draw attention to some relevant (present) disposition or dispositions. For instance, in the volcano case, we could say that ‘the real’ explanation is that the volcano is (presently) disposed to erupt.

We are not convinced of the plausibility of this move. For one thing, backwards explanations seem to be given and understood by those to whom it is not terribly plausible to attribute any understanding of what these dispositions are. For another, the existence of an explanation in terms of dispositions does not as yet make it plausible that this is ‘the real’ explanation and the backwards explanation is no explanation at all. While the dispositional explanation of the volcano’s eruption may be a good explanation, why should this cast doubt on the existence of another, backwards, explanation in terms of the fact that the volcano is about to erupt? Thirdly, we doubt whether the dispositional response can cover all the cases; it is unclear how, for instance, it will apply to the apparent backwards explanation of the slowing of a planet which mentions only the fact that it is about to reach its apogee.

Fourthly, it seems that the required dispositional explanations are not equivalent in strength to the corresponding backwards explanations. This makes it prima facie unlikely that the latter are ‘really’ the former. To see the difference, suppose I try to explain a volcano’s smoking by saying that it is about to erupt. We then observe the volcano, but in fact it doesn’t erupt. Then my explanation attempt was mistaken; my claim that the volcano was about to erupt was false, and there must have been some other reason why the volcano was smoking. Now suppose, for comparison, that instead of saying that the volcano is going to erupt I say that it is disposed to erupt. And we observe it as before, and it doesn’t erupt. Still, my explanation may have been perfectly correct; it may have been true, and we are not compelled to look for another explanation of the smoking. (We may suppose that the facts which underlie the disposition - facts about whether there

15

is magma heading towards the volcano’s crater, for instance - are also responsible for the smoking, so that citing the disposition can serve as a good explanation of the smoking, at least for those who know how these things work.)

Some readers might wonder if this is really fair. Sometimes, they might think, we use phrases like “x is going to φ” in such a way that they are not falsified if it turns out that x does not φ. On these uses, it is perfectly acceptable to say things like: “Sarah was going to go to the party, but then she didn’t”, or: “The elastic band was just about to break, but then I let it go”. What is meant by these uses of “x is going to φ” is that x is disposed to φ, or planning or intending to φ, or likely to φ, or something similar.

A better response to our claim that backwards explanations appear to be an important class of cases (better, that is, than the dispositional response just described) holds that apparent backwards explanations are ‘really’ explanations in terms of one or more of these various things which can be meant by expressions like ‘x is going to φ’. Furthermore, it holds that the reason these things are ‘the real’ explanations, rather than merely one among many, is that they are what is meant when the apparently-backwards explanations are given.

One source of dissatisfaction with this line of response is that, even if in each case of apparent backwards explanation we were prepared to acknowledge the existence of some explanation of the explanandum event in terms of dispositions, likelihoods or whatever, we can as yet see no good reason to regard these replacement explanations as ‘the real’ explanations in the sense of the ones which were really intended. We doubt whether all the everyday examples of apparently-backwards explanation are plausibly interpreted as invoking dispositions, intentions, likelihoods or whatever.

16

But more crucially, regardless of what people ordinarily intend when they appear to give backwards explanations, it seems to us that backwards explanations intended as such can be perfectly good explanations in all sorts of situations. 1-5 on our list, for example, when read as backwards explanations and not anything else, seem perfectly kosher to us, at least prima facie.

Conveying More Than You Say

It is important to make clear that admitting that backwards explanations are a significant group does not commit one to any view about where the explanatory action is in these cases. One might hold (with Lewis 1986) that the explanatory action is located in the provision of causal historical information about the explanandum. Lewis’s sophisticated causal historical account of event explanation can handle backwards explanations, because he does not require that information be explicitly given about causal histories: it can be enough, for example, that causal historical information be supplied “on condition that the recipient is able to supply the extra premises needed” to derive this information given what is said (p. 222). So it could be that citing a volcano’s imminent eruption explains its present smoking because it enables the hearer to work out that the present smoke is caused by the sort of process which precedes an eruption. An explanation can convey more information than it explicitly contains, just as a police informant can provide the police with information about who the criminal is without explicitly telling them by saying “Let’s just say she’s older than any of us.”

So, as we mentioned in footnote 2, the claim that there is a significant class of backwards explanations may not undermine sophisticated causal historical accounts of explanation. Still, we do think that careful consideration of this class puts some pressure on the thought that causal historical information provision is always where the explanatory action is. Some backwards 17

explanations seem to work well even when their recipients do not seem to be ‘able to supply the extra premises needed’ to derive much relevant causal historical information from them. I can know an alarm sounds just before unauthorised people enter the building, but have no idea how the people are detected, how their authorisation is determined, etc. However, this theory can always be saved if we allow the causal information to be unspecific enough (“something, I know not what, precedes this smoking and the eruption” “something about unauthorised people about to enter the building is detected and causes the alarm to sound”). Be that as it may, our central point is merely that a theorist attempting a general theory of explanation should either count backwards explanations as among the data she needs to accommodate, or come up with a very good reason not to.

Some theorists of explanation are careful not to claim that they are giving theories of explanation in general. Lipton 1991, for example, offers a “causal model” of explanation, but allows that this model “cannot be complete” (p. 34), on the grounds that it does not cover some sorts of explanations, particularly mathematical and philosophical ones. Woodward 2003, likewise, offers only an account of causal explanations, not of explanations in general (Woodward 2003 pp. 4-5), and Salmon 1984, while more ambitious, appears to only claim his causal theory of explanation is “reasonably adequate for the characterization of explanation in most scientific contexts–in the physical, biological, and social sciences–as long as we do not become involved in quantum mechanics” (Salmon 1984 p. 278). This caution is often motivated by concerns about mathematical explanation, or worry about the mystery of quantum mechanics, or tasks such as explaining the meaning of an expression. If we are right, there is another good reason for this caution: the apparent existence of backwards explanations of the varieties we have been considering. None of Lipton’s, Woodward’s or Salmon’s models of explanation can be applied very plausibly to our cases, even though our cases concern contingent empirical matters of fact (unlike mathematical and

18

some philosophical explanations), and include explanations found in the sciences. We are inclined to believe that theories of one special class of explanations that do not seem to be able to be usefully generalised to explanations across the board should be suspect, as with any other theory that responds to apparent counter-examples simply by restricting its scope so as to cover only cases where there are no counterexamples. But a general theory of explanation goes beyond the scope of this paper, and probably beyond the extent of these authors’ agreement.8

19

References

Bridgman, P. (1968). The Logic of Modern Physics. (New York: Macmillan)

Dray, W. (1964). Laws and Explanation in History. (Oxford: Oxford University Press)

Friedman, M. (1974). Explanation and Scientific Understanding. Journal of Philosophy, 71, 5-19

Hawthorne, J. (2006). Metaphysical Essays. (Oxford: Oxford University Press)

Hawthorne, J. and Nolan, D. (2006). What Would Teleological Causation Be? (In Hawthorne 2006, pp. 265-84.)

Hempel, C. (1965.) Aspects of Scientific Explanation And Other Essays In The Philosophy of Science. (New York: The Free Press)

Kitcher, P. (1981). Explanatory Unification. Philosophy of Science, 48, 507-31.

Kitcher, P. (1989). Explanatory Unification and the Causal Structure of the World. (In Kitcher and Salmon 1989, pp. 410-505.)

Kitcher P. and Salmon W. (eds). (1989). Scientific Explanation, Minnesota Studies in the Philosophy of Science volume XIII. (Minneapolis: University of Minnesota Press)

Lewis, D. (1986). Causal Explanation. (In his Philosophical Papers, Vol II (pp. 214-40). Oxford: 20

Oxford University Press.)

Lipton, P. (1991). Inference to the Best Explanation. (London: Routledge)

Millikan, R. (1984). Language, Thought and Other Biological Categories. (Cambridge, MA: MIT Press)

Millikan, R. (1989). In Defence of Proper Functions. Philosophy of Science, 56, 288-302

Price, H. (1996). Time’s Arrow and Archimedes Point. (Oxford: Oxford University Press)

Neander, K. (1991a). Functions as Selected Effects: The Conceptual Analyst’s Defence. Philosophy of Science, 58, 168-84

Neander, K. (1991b). The Teleological Notion of ‘Function’. Australasian Journal of Philosophy, 69, 454-68

Salmon, W. (1984). Scientific Explanation and the Causal Structure of the World. (Princeton: Princeton University Press)

Salmon, W. (1989). Four Decades of Scientific Explanation. (In Kitcher and Salmon 1989, pp. 3219.)

Scheffler, I. (1957). Explanation, Prediction and Abstraction. British Journal for the Philosophy of Science, 7, 293-309

21

Strevens, M. (2004). The Causal and Unification Accounts of Explanation Unified – Causally. Noûs, 38, 154-79

Woodward, J. (2003). Making Things Happen: A Theory of Causal Explanation. (Oxford: Oxford University Press) 1

We are not presupposing here that events are the only, or fundamental, explanatory relata, only that events can be

explanatory relata (or at least, can play whatever role it is that they play in explanation-giving sentences like 1-7). Analogues of our arguments phrased in terms of processes, facts or states may work equally well. 2 On inspection of the literature, there are not as many of these as one might think. Wesley Salmon (Salmon 1984), Peter Lipton (Lipton 1991) and James Woodward (Woodward 2003), as we will see below, do not claim that explanations in general are causes. One of the few to defend a more general causal-historical theory of explanation, intended to cover at least all explanation of events, is David Lewis (Lewis 1986). However, as we shall see below (p. 17), Lewis’s sophisticated account seems to be able to meet the challenge of accounting for backwards explanation. Salmon, Lipton and Woodward are, however, among those who must allow that there is a class of explanations their accounts do not handle, if indeed there are backwards explanations. 3 Hempel himself remarks on this point: 1965, pp. 353-4. He seems content with this consequence of his account of explanation, but offers no arguments in response to those who disagree, merely pointing out that ‘it is not clear ... what reason there would be for denying’ that there are backwards explanations, except perhaps for reliance on some ‘notion of factors “bringing about” a given event’, and he thinks ‘it is not clear what precise construal could be given to [that notion]’. 4 Not all though. It could be, for instance, that my partner, knowing of the impending visit which I’ve forgotten about, has asked me to tidy the flat. The existence of such possibilities casts doubt on the idea that some forwards explanation is the ‘real explanation’ in such cases. For one might proffer the backwards explanation without knowing which of the forwards explanations – the belief-involving one, the partner-involving one, and so on – are true. This being so, it would be odd to claim that one of these forwards explanations is what they really proffered. 5 This seems unlikely to help with example 1, since it is more plausible that the future fact explains my belief in it than vice versa. 6 But note that this tendency will make it seem, rather implausibly, that ‘the real’ explanation of e is some maximal

22

explanation containing as much explanatory information as there is about e. 7 Note that biological-functional, or otherwise teleological, explanations need not be backwards explanations. Explanations which cite present goals to explain present behaviours, for instance, are teleological but not backwards. This point is not always made fully clear; Salmon (1989), for example, says that ‘the issue of final causation’ is ‘the problem of explaining present facts in terms of future goals’ (p. 28, emphasis added), and also that ‘the basic philosophical problem’ of teleological explanation (in cases where there is no conscious intent) ‘involves the question of whether a future state can legitimately be said to explain a present fact’ (p. 26). 8 We would like to thank members of the ANU Explanation Reading Group, as well as audiences at the Australasian Association of Philosophy Conference 2007, the Arché Core Seminar and the Bellingham Summer Philosophy Conference 2007. We also particularly thank Trenton Merricks and Alyssa Ney. Daniel Nolan’s work on this paper was supported a Philip Leverhulme Prize awarded by the Leverhulme Trust.

23