The Challenge
We see a carelessly kicked football head right towards a window, and moments later, the window is broken. We see the traffic lights change from green to red, and this is immediately followed by a cessation in the flow of traffic. We hear someone in an office building shout ‘Fire!’, and there is subsequent stampeding of feet down the corridor. What we want to say in these, and similar, cases, is that we perceived one event cause another. This is certainly how we report them: ‘I saw the ball smash the window’, ‘I saw the lights halt the traffic’, ‘I heard people responding to the fire warning’. But is there anything more to our perception of causation than perception of temporal sequence? We seem inclined to think so. Certainly, we want to distinguish between the cases above and what we see as mere temporal sequence, for example the striking of a church clock being followed by a sudden cloud burst. But is this to do with what we perceive, or just to do with our background assumptions about what kinds of thing can be connected?
This is the challenge that David Hume poses, first in his Treatise of Human Nature, the first two books of which were published in 1739 when he was just twenty-six, and again nine years later in the Enquiry Concerning Human Understanding, written under the impression that his first book had been largely ignored. It leads him to propose a view of causation which is some distance from our ordinary assumptions. Let us follow Hume's path, and see whether the view to which it leads is one we want to accept.
Hume was a polymath: philosopher, historian, man of letters . . . and I think it is appropriate to call him a psychologist as well. In the mid-1700s the term ‘psychology’ was not yet widespread, and a century was to pass between Hume's death and the attempt to establish psychology on a scientific basis. But enquiries into how we actually acquire certain concepts, enquiries we would now think of as the legitimate concern of psychology, were often pursued under the name of ‘mental philosophy’. Hume was an empiricist: he thought that our knowledge of the world was founded on experience. More precisely, he held that simple ideas (concepts which cannot be broken down into constituent concepts) had to have been acquired from a previous impression, that is, an experience (whether sensory or otherwise). Complex ideas do not need precisely corresponding impressions, but can be built up from simple ideas.
So what of the idea of causation? Is this a complex or simple idea? Putting that question on one side for a moment, let's consider the kinds of experience which lead us to form causal judgements: perhaps the origin of the idea of causation can be found there. The case of the ball striking the window and smashing it is an obvious case: if this isn't causal, what is? But what do we actually directly observe here from which we might form an idea of causation? Two things, it seems:
(i) Contiguity in time and space: the contact of the ball against the window and the window's smashing are right next door to each other, spatially and temporally;
(ii) Temporal priority: the contact happens before the smashing.
Something needs to be added to this list of ingredients if it is to lead to a sense of causation. What is it? The necessary connection between the first event and the second, is the natural answer. This isn't just a random pairing. The first event necessitated the second event. Given that this was a normal football, moving at some speed, and this was a normal window, not a super-reinforced one, this kind of contact was just bound to result in a smashing, wasn't it? Granted that we have this idea of causal necessity, from what impression, what experience, did we acquire the idea? As long as we look just at the scene itself, we search in vain for the origin of the idea of causal necessity. For the scene itself only gives us the events, and the temporal and spatial relationships between those events. That's it. The mysterious relationship that binds the two events together, the ‘secret connexion’, as Hume calls, it, is invisible, inaudible, intangible, and so elusive.
There is, admittedly, an idea of necessity, of a certain kind, which we do not get from (sensory) experience, but it won't help us here, thinks Hume. It is the idea of logical necessity, the relation between ideas. The idea of twelve objects is necessarily connected to the idea of two, three, four and six objects, in that twelve objects can be grouped into two groups of six, or three groups of four, and so on. Where there are twelve objects, there are, as a matter of necessity, also six objects (though not just six objects). But that kind of necessary connection is not the kind of necessity to be found between cause and effect. For, as Hume points out, there is no contradiction in supposing the event we think of as the cause occurring, but the event we think of as the effect, not. We can imagine one billiard ball hitting another, and the second not moving. Surprising, perhaps, but not logically impossible. So the (idea of the) cause does not logically necessitate the (idea of the) effect. We will have to look elsewhere for the origin of the idea of a necessary connection between cause and effect.
The Account: Repetition and Projection
There is another aspect of those scenes we think of as causal which is open to observation: we may notice similarities with other scenes we have witnessed. Perhaps this is the first time this particular ball has come into contact with this particular window, but we have seen this kind of event pairing (contact of a reasonably dense object with a fragile object, followed by a smashing of the latter) many times. So we can add a third item in our list of observables:
(iii) Repetition, or what Hume calls constant conjunction: an event of the first kind is invariably followed by an event of the second kind.
So, what is there out in the world which could possibly give rise to an idea of causation? We have identified three aspects: contiguity, priority in time, and constant conjunction. And since, according to empiricists, our ideas of the world cannot completely transcend our experience, it seems that the only idea of causation to which we can legitimately lay claim is one which is defined by those three components. And Hume does in fact present as a definition of causation one which talks simply of the constant pairing of an event of one kind with an event of the other kind. The only thing that distinguishes cause from effect is priority. The event which came first we name ‘cause’; the event which came after we name ‘effect’.
Now, this view of causation has a rather surprising consequence: that whether or not two particular events are related as cause and effect depends on whether or not they are part of a pattern. It is not just the relationships between those events that matters, but also whether similar events also stand in those relationships. Suppose we designate individual events by lower-case letters (a, b), and the kinds of events they are instances of by upper-case letters (A, B). Then whether ‘a caused b’ is true depends on whether A-type events in general are followed by B-type events. To frame this in a slogan: there is no singular causation. For to say that causation is singular is to say that whether or not a and b were causally connected is to do just with the relations between those particular events, and nothing else. But the only (relevant) observable relations between two particular events are their spatial and temporal ones. There is nothing here we can identify as a necessary connection between cause and effect.
Are we then to abandon the search for the idea of necessary connection? Not quite. There may be no causal necessity out there in the world, but we need to be able to explain why we are inclined to attribute something other than temporal and spatial relationships to cause–effect pairs. If causal necessity is an illusion, where does the illusion come from?
Hume suggests that the constant conjunction which defines causation also explains why we are inclined to attribute a closer connection between cause and effect than is suggested by mere contiguity and temporal priority. For, having been exposed to the constant conjunction of As and Bs, whenever we see an instance of A, we immediately think of B. We see the ball heading towards the window, and we just know what to expect. Psychologists now call this phenomenon ‘classical conditioning’. The most famous illustration of this was the pioneering study by the Russian physiologist Ivan Pavlov in the later nineteenth century. Pavlov was studying the salivary reflex in dogs when presented with food. He found that, if a bell was repeatedly rung before the presentation of food, dogs would eventually salivate when the bell was rung, even before the food appeared. This was termed the ‘conditioned response’. So, as we might put it, in admittedly anachronistic terms, Hume sees classical conditioning as the origin of such ideas we have of causal necessity. Although the expectation of the effect when presented with the cause is, of course, a purely mental phenomenon, we project this psychological necessity onto the external objects, and imagine that there is something in the cause which compels the effect to occur. This is an example of the tendency of the mind, in Hume's memorable phrase, to ‘spread itself’ on the world.
What Hume has given us is a pleasingly unified theory of causation and of our experience of it. Causation is a matter of constant conjunction of certain kinds of events, and it is exposure to this constant conjunction which creates an expectation in our minds, when we see an event of one kind, that it will shortly be followed by an event of some other kind. This is the origin of the confused idea of causal necessity; there is no counterpart in the real world of such an idea. The account is, perhaps, a surprising one, and Hume concedes that we might feel some resistance to it. Compare the events of the world to a mosaic of tiles. It is natural to think that causation is something that binds individual events, like the cement between the tiles. (In fact, Hume does in one his discussions of causation talk of it as ‘the cement of the universe’, but here he is thinking of it in psychological terms as something that ties our thoughts together.) Instead, it turns out to correspond to the pattern of the mosaic. That, at any rate, is the conclusion empiricism seems to take us towards. But does the account work?
A word of warning before we proceed. What I have just described is the account of causation, and of causal necessity, traditionally ascribed to Hume. However, some scholars within the last forty years or so have preferred a quite different interpretation: that, for all he appears to say to the contrary, Hume does after all believe in the real existence of the ‘secret connexion’, that causation is something that links individual events, and is not just a matter of patterns in nature. I will not enter that scholarly debate, but merely note it. The interested reader is invited to consult the suggestions for reading at the end. My current concern, which I invite you to share, is whether the account traditionally ascribed to Hume is a workable account. I suggest not, for reasons I shall now give. But for simplicity, whenever I talk of ‘Hume’ in what follows, I am to be understood as meaning ‘Hume, as traditionally interpreted’, and by ‘Hume's account’, ‘the account traditionally ascribed to Hume’.
The Tension
There are two standard objections to Hume's account, neither of which will detain us long here, but they should be mentioned. The first is that the constant conjunction condition is too strong, and the second that it is too weak. To elaborate on the first: we might reasonably suppose that there are cases of As genuinely causing Bs, where A's aren't invariably followed by Bs. The application of lit matches to flammable materials is often followed by a conflagration; but not always. So there can be causation without constant conjunction. And to elaborate on the second objection: the constant conjunction account fails to honour the distinctions we would ordinarily want to make between different kinds of regularities. Let us suppose that A-type events are invariably followed by B-type events. There are three possibilities, we are apt to suppose: (i) this is just coincidence; (ii) the As are causing the Bs; (ii) the As are not causing the B's, but both are the result of a common cause. So there can be constant conjunction of As and Bs without a direct causal relationship between As and Bs.
In response to the first objection, some so-called ‘regularity’ accounts of causation agree with Hume (as traditionally interpreted, recall) that causation is a matter of regularity, but relax the constant conjunction condition. They might say, for instance, that there is a very high statistical likelihood that As will be followed by Bs. In response to the second, they might say that some regularities are more fundamental than others. These are the regularities (perhaps at the molecular or even sub-atomic level) which explain observed regularities. If it is objected that, even at this supposedly more fundamental level, there can be merely coincidental regularities, or regularities due to a common cause, regularity theorists can dig their heels in. If causation is a matter of regularity at the most fundamental level, then there is no distinction to be made, at this level, at any rate, between ‘mere’ regularities and ‘genuinely causal’ regularities.
The key point about both the basic constant conjunction account, and its more sophisticated successor regularity theories, is that they deny singular causation. It is this that is the sticking point. Even if each part of Hume's account is independently defensible, the two parts together do not add up to a workable theory, for they are in tension with each other. Let me explain.
Hume's account of how we come to have a (possibly confused) idea of causal necessity is itself a causal account: it is as a result of our experience of the constant conjunction of As being followed by Bs that we are led to think of Bs whenever we perceive As. Now, that it is a causal account isn't itself a problem. Hume is not denying that there is such a thing as causation, despite the fact that he is, or was, sometimes called a ‘causal sceptic’. The problem is that if causation is simply constant conjunction, then the ‘classical conditioning’ story does not get off the ground. As Hume notes, the first time an event of a specific kind is observed, there is no expectation that an event of any specific kind will follow it. It is only after repeated pairings that a mental association is formed between two kinds of events. But it is not as if this suddenly happens after a definite number of pairings. Rather, the effect of each pairing must be cumulative. It is the steady drip-drip-drip of experiences of As being followed by Bs that gradually forms and strengthens the association. But for that to happen, there must be singular causation: each individual pairing is making a causal contribution. There is simply no plausible way to recast the causal consequence of pairings in constant conjunction terms (e.g. every instance of forty AB pairings is followed by an instance of a mental association between A and B). Hume's account of the origin of the idea of causal necessity may be correct, but only if it is based on an assumption of singular causation. Alternatively, Hume's constant conjunction account of causation, or some more sophisticated regularity account, may be correct; but then his explanation of the idea of causal necessity has to be abandoned. He cannot have both parts of his account.
Perhaps this tension between the two parts of the account, the metaphysical and psychological parts, is one reason to rethink the traditional interpretation of Hume. Perhaps, after all, he does believe in the existence of a ‘secret connexion’, and this is what he is (implicitly) appealing to when expounding his psychology of causation. I will make no further comment on that. My present concern is what the truth is about causation and our perception of it.
The Truth (?)
Time and causation are intimately linked, it seems. We suppose that effects come after their causes, not before. And the more immediate the effect, the closer it is to its cause in time. But what explains this relationship between time and causation? Is it perhaps because one is more fundamental than the other? Hume's (official) account of causation and of our perception of causation prioritizes time. Causation itself is defined in temporal terms: constant conjunction is a temporal notion (repetition of togetherness in time), and what marks out the difference between the cause and the effect is order in time.
But what if it is causation that is the more fundamental of the two? Let us see what kind of account might emerge if we take causation to be basic. Let us begin, then, by assuming that there is such a thing as singular causation. Hume's ‘secret connexion’ really does link individual events. This particular ball-kicking caused this particular window to break, whether or not there have been or will be any other cases of balls kicked towards windows, to the detriment of the latter. Moreover, it is because the kicking of the ball caused the breaking of the window that the breaking happened after the kicking, not before. The direction from cause to effect is defined by the direction of causation. ‘But’, someone will object, ‘surely there are cases of causally unconnected events which nevertheless are part of a temporal sequence. The shooting star appeared just before the kettle boiled, but there is no causal connection there at all – it's just the way things turned out.’ Well, that may be true. We can accommodate that kind of objection, and still define the direction of time in causal terms, as follows: ‘x is earlier than y if and only if x is simultaneous with a cause of y.’ That accommodates the shooting star and kettle case. The appearance of the shooting star may not have caused the kettle to boil, but it was simultaneous with something that did.
Defining time's direction in terms of causal direction explains why we never perceive (as opposed to merely expect, or imagine) the future. To perceive something is to be causally affected by it. The explosion occurs; light and sound from it travel to our sense organs, and strike our retinas and ear drums, respectively; we then, as a result, see and hear the explosion. So, if causal order defines temporal order, it follows that what we perceive – the explosion – must always be earlier, and never later than, our perception of it. But what we call ‘the future’ is always after our current experience, and so cannot cause that experience. If instead we defined causation in terms of time, we would have to reverse the order of explanation: it is because what we perceive comes before our perception that we call what we perceive the ‘cause’ of our perception. But what then would explain why what we perceive comes before our perception of it?
As far as reality is concerned, then, causal order may be more fundamental than temporal order. But what about our experience of reality? Here, it is much more plausible to say that we perceive causal connections by virtue of perceiving temporal connections, as the following example suggests. You are looking at two discs on a screen. Slowly, one is moving towards the other, stationary one. Just after they ‘touch’, the stationary one starts to move in the same direction, while the first has come to a standstill. You cannot help seeing this in causal terms: the one disc hit the other, making it move. Of course, what is really going on is more complex. All you actually saw, as Hume no doubt would point out, is one movement followed by another. Had the gap between the two discs meeting and the second moving off been much longer, or if the second disc had started to move before the first disc had reached it, you would probably not have read the scene in causal terms: the first disc did not ‘make’ the second move.
If all we perceive in these events is their temporal relationships, why do we read causation into them? Why do we have this vivid sense that we are seeing causal connections? This is really a version of Hume's original challenge. The answer may be that we are hard-wired to interpret some kinds of temporal sequence as causal, as a result of natural selection. Creatures who ‘see’ causality, and act accordingly, are at least some of the time, perhaps most of the time, tracking genuine causal connections in their environment, and so stand a better chance of survival than those who fail to track causal connections.
Perhaps perception of temporal sequence is more fundamental than perception of causal sequence. In that, Hume may well be right. But he seems to take the perception of temporal sequence as unproblematic. Here, though, there is another puzzle. Perceiving one event as following another - for example, the playing of three notes on the piano - isn't simply a matter of perceiving one, and then perceiving the other. Suppose the notes to be separated by a significant interval. Then, although there is a succession of perceptions (first of an E, say, and then an F#, and then a G), there is no perception of succession, and so the notes are not experienced as a musical phrase. When the notes are close together, hearing the later notes is coloured by the very recent memory of hearing the earlier notes. That is, there is a causal connection between the earlier perceptions and the later ones. So, although the perception of time order may be psychologically more fundamental than the perception of causal order, still the perception of time order requires a causal order among the perceptions. Here, again, we need to appeal to singular causation. It is this particular hearing of the E that affects this particular hearing of the F#. A constant conjunction account of causation, in contrast, is powerless to explain the experience of temporal succession. For on that account, ‘The perception of E causally affects the perception of F#’ is equivalent to ‘Perceptions of E are always followed by perceptions of F#’, which is not only demonstrably false, but even if true would not distinguish between the perception of F# as following E on the one hand, and the perception of F# just following the perception of E.
All this does not amount to proof, but it does show how much more can be explained if we take causation to be fundamental, both to the explanation of time order, and to the explanation of our perception of time order. We are better off assuming that what Hume, perhaps ironically, referred to as ‘the secret connexion’ really exists. Causation really is a most singular affair.
Suggestions for Reading
Hume's first discussion of causation is in his Treatise of Human Nature, Book I, Part III. See especially Sections II, III, XIV and XV. His later discussion can be found in An Enquiry Concerning Human Understanding, Section VII (in Part II of which he provides no less than three definitions of causation). There are some differences of detail between the Treatise and the Enquiry, but the key elements of the constant conjunction account, and the psychological account of necessity, appear in both. Hume summarises his account of causation in an Abstract of the Treatise, published one year later, in 1740. It is at the end of the Abstract that he describes causation (along with resemblance and contiguity) as ‘the cement of the universe’.
The traditional interpretation of Hume, and a defence of a more sophisticated regularity theory, is given in J. L. Mackie's The Cement of the Universe: A Study of Causation (Oxford: Clarendon Press, 1974). Galen Strawson offers a reinterpretation of Hume, in which he argues that Hume does, after all, believe in a much more robust kind of causation than mere regularity, but thinks our knowledge of it is limited, in The Secret Connexion: Causation, Realism and David Hume (Oxford: Oxford University Press, 1989). For a thorough assessment of the debate between ‘Old Hume’ and ‘New Hume’ interpretations, see Peter Millican, ‘Hume, Causal Realism and Causal Science’, Mind 118 (2009): 647–712.
For a defence of singular causation, and a critique of regularity accounts, see Michael Tooley, Time, Tense and Causation (Oxford: Oxford University Press, 1997). See also D. H. Mellor, The Facts of Causation (London: Routledge, 1995), for another defence of a robust notion of causation, and an account of causation's role in explaining both order and the perception of time order.
Psychological experiments concerning the perception of causation, involving moving discs, are reported in Albert Michotte, The Perception of Causality (New York: Basic Books, 1963). These explore the ways in which our causal judgements, on the basis of what we perceive, are highly sensitive to the temporal relations between perceived events. See also David Danks, ‘The Psychology of Causal Perception and Reasoning’, in Helen Beebee, Christopher Hitchcock and Peter Menzies (eds.) The Oxford Handbook of Causation (Oxford: Oxford University Press, 2009), 447–70.
