A central intuition we have about singular causal relations is that they have an intrinsic character. I mean by this, roughly, that whether a causal relation obtains depends entirely on the local character of the events and the process that links them. If I drop a piece of sodium into a beaker of acid and that causes an explosion, this causal relation is an intrinsic feature of the cause-effect pair. So if there is another person waiting in the wings, ready to drop a piece of sodium into the beaker if I do not, that makes no difference to whether the causal relation holds between my dropping the sodium and the explosion. The presence of the alternative cause is neither here nor there to the causal relation that exists between the actual cause and effect. The causal relation does not depend on any other events occurring in the neighbourhood: the causal relation is intrinsic, in some sense, to its relata and the process connecting them. 1

This intuition lies at the heart of a number of recent philosophical theories of causation. Prominent among these theories are ones which identify the causal relation with some physically specifiable relation. David Fair (1979), for instance, proposes that the causal relation is the relation of transfer of energy-momentum from cause to effect. Fair observes that with many commonly recognisable causal relations there is a flow of energy or momentum from the cause to the effect: in fact, if the cause and effect belong to a closed system, the conservation laws enable us to identify the quantity of enery or momentum that is transferred. In a similar vein, John Bigelow and Robert Pargetter (1990) identify the causal relation between events with the aggregate of forces existing between them. They argue that the causal relation is best identified with an irreducible second-order relation consisting in an aggregate of forces and holding between complex structures comprising objects and first-order properties. These theories substantiate the intuition about intrinsicality in that they take causation to be a relation that depends on the local character of the cause and effect and the process connecting them.

To be sure, these theories admit a degree of extrinsicality in causation to the extent that they suppose that a causal relation must conform to the operation of a law. If it is conceptually necessary that a causal relation must fall under a law and a law is something that constrains the behaviour of widespread, diverse phenomena, then causation cannot be an entirely intrinsic phenomenon. However, other philosophers advancing a singularist conception of causation have denied that there is any conceptual necessity attaching to a causal relation’s being law-governed. Elizabeth Anscombe (1975) has been interpreted as advancing the view that we can perceive causal relations without any knowledge of whether they are law-governed. Michael Tooley (1990) explicitly argues for a singularist conception of causation according to which a singular causal relation can hold between events even though it does not fall under a causal law. David Armstrong (1997a; 1997b) has argued that causation is a conceptually primitive external relation and that there is no a priori reason for thinking that this relation must be law-governed, though he does admit that as a matter of a posteriori necessity each singular causal relation is identical with a particular instantiation of a law.

My aim in this paper is not to defend any one of these versions of singularism about causation. Nor is to decide which version of singularism—the weaker nomic version or the stronger anomic version—is correct. My aim is merely to defend the idea common to both versions of singularism, namely the idea that the truth-maker for singular causal claims is a relation that holds solely in virtue of the intrinsic character of the cause-effect pair. Many contemporary philosophers will view this idea with suspicion, no doubt motivated by the belief that Hume’s critique of the idea conclusively demonstrated its untenability. Hume’s critique was explicitly directed at the idea that causation consists in a necessary connexion, or a relation of power, force or energy between events. Generalising slightly, we can interpret the critique as an attempt to undermine any theory that says that singular causation consist in a local, intrinsic tie connecting events.

Hume’s critique has been enormously influential over the centuries, and this influence has continued into contemporary times. The lessons of Hume’s critique have been so thoroughly assimilated that few contemporary philosophers take the trouble to examine it closely. However, I want to argue in this paper that when we do examine it in detail, we can see that it completely misses its mark. I shall argue that Hume’s critique lacks cogency because it is based on questionable empirical assumptions about perception and even more dubious semantic assumptions about the meaningfulness of ideas. My attack on Hume’s critique will be two-pronged.

The first prong of my attack will be to cast doubt on the distinctively Humean claim that we do not have impressions of causation. I shall argue in section 2 that none of the arguments advanced by Hume shows that we do not perceive causal relations. In section 3 I examine the empirical question of what makes something observable and I rehearse the empirical evidence, taken from studies by cognitive psychologists, for the claim that in certain favourable circumstances we perceive causal relations.

The second prong of my attack will be to examine the central principle of Hume’s critique, the principle that every meaningful idea must have a source in an impression. In section 4 I shall claim that Hume’s version of this principle, along with more contemporary linguistic versions of it, are false and commonly known to be so. Indeed, it has become a familiar point in the philosophy of science that the empiricist requirement that a meaningful term be completely definable in observational terms is overly stringent in implying the meaningless of all terms for theoretical entities. When the empiricist requirements are weakened so as to apply plausibly to theoretical entities, there is good reason for supposing that terms denoting intrinsic causal links will meet the weakened requirements.

The upshot of this two-pronged attack will be that Hume’s critique is completely undermined. Contrary to Hume, there is good reason for thinking that we do receive impressions of causation; and even if we do not, there is every reason for thinking that any empiricist principle that makes terms denoting intrinsic causal links meaningless is absurdly strong.

2. HUME ON INTRINSIC CAUSATION

It is often thought that Hume showed conclusively that singular causation cannot consist in a local, intrinsic tie between events. Is this thought correct? In this section I shall try to answer this question by rehearsing Hume’s argument and testing its cogency.

In rehearsing Hume’s argument, I shall adopt a very traditional interpretation of it. I do not know whether this is the correct interpretation of Hume. My main purpose is not to find an interpretation of Hume that is historically faithful to his intentions and sensitive to all the subtleties of his position. It is rather to reach an understanding of how subsequent generations of philosophers, especially philosophers in this century, have been so persuaded by his critique of singular causation. For this purpose it is appropriate to focus on the interpretation contemporary philosophers have traditionally given the critique, however naive and historically inaccurate that may turn out to be.

I shall concentrate on Hume’s discussion of causation in the An Enquiry concerning Human Understanding (1748) rather than A Treatise of Human Nature (1739), as the later work is fuller and more explicit in its treatment of the issues. He begins his discussion of causation in the Enquiry (VII) with these words: “There are no ideas, which occur in metaphysics, more obscure and uncertain, than those of power, force, energy, or necessary connexion, of which it is every moment necessary for us to treat in all our disquisitions.” The ideas Hume mentions are very different in character; and one might think that a critique which targets one of them may not be effective against the others. In this connection, however, it is significant that all these ideas have something in common; and it is that all of them give causation a distinctive intrinsic character. The central target of Hume’s criticism is sometimes taken to be a standard rationalist view that singular causation involves a necessary connexion of a kind which would form the basis of a priori causal inferences; a connexion between events such that if we knew it, we would be able to infer from a cause that a certain effect must follow, or from an effect that it must have been produced by a certain cause. (See Mackie 1974.) This may well be the target that Hume had in mind in framing his argument. But it is also illuminating to read it, as I propose, as a critique of the conception of causation as an intrinsic relation consisting in a local tie between events, the kind of relation that can obtain in single instances, independently of what happens in neighbouring or distant spatiotemporal regions.

Hume’s critique takes the form of a “sceptical doubt” concerning this conception. By this, I mean that, by adopting the role of the sceptic who questions the legitimacy of the conception, he attempts to show that our confidence in our understanding of the concept of causation is misplaced: the concept is not as well understood or justified as we first thought. Central to his “sceptical doubt” is a certain principle in the theory of ideas that expresses his psychological version of empiricism. Put roughly, it is the principle that all ideas are derived from impressions of sense or inner feeling. Hume writes: “It seems a proposition, which will not admit of much dispute, that all our ideas are nothing but copies of our impressions, or, in other words, that it is impossible for us to think of anything, which we have not antecedently felt, either by our external or internal senses.” (VII, p. 62) In Hume's philosophy, ideas and impressions are distinctive kinds of “objects of the mind”: impressions are the more forceful or lively and also causally prior; ideas are caused by impressions and are “faint images” of them. Further, Hume classifies impressions into two categories according to their provenance: impressions which derive the senses he calls impressions of sensation and impressions which derive from awareness of one’s internal mental processes he calls impressions of reflections.

Despite the unqualified formulation above, Hume’s principle about ideas applies only to simple ideas that cannot be decomposed into other ideas: it is only for such ideas that there must be a corresponding simple impression that is its ultimate causal source. However, ideas that are not simple in this way can be defined, if they are meaningful, in terms of simple ideas that do have their ultimate causal source in impressions. Every meaningful idea must, therefore, be a copy of an impression, or be definable in terms of simple ideas which are copies of impressions.

Hume’s “sceptical doubt” starts with him seeking an experiential source for an impression of intrinsic causal links, links involving necessary connexions or relations of power, force, or energy. He considers whether we may derive such an impression from our observations of the causal operation of external bodies upon each other, from our experience of exercising control over our bodies, and finally from our experience of exercising control over minds. He sums up his discussion in the first part of section VII as follows:

We have sought in vain for an idea of power or necessary connexion in all the sources from which we could suppose it to be derived. It appears that, in single instances of the operation of bodies, we never can, by our utmost scrutiny, discover any thing but one event following another; without being able to comprehend any force or power by which the cause operates, or any connexion between it and its supposed effect.... So that, upon the whole, there appears not, throughout all nature, any one instance of connexion which is conceiveable by us. All events seem entirely loose and separate. One event follows another; but we can never observe any tie between them. They seem conjoined, but never connected. And as we can have no idea of any thing which never appeared to our outward sense or inward sentiment, the necessary conclusion seems to be that we have no idea of connexion or power at all, and that these words are absolutely without any meaning, when employed either in philosophical reasonings or common life. (VII, pp. 73-4).

It is important to point out the emphasis Hume gives to the word “seems” in this passage: he says that the idea of an intrinsic causal link appears to be meaningless. Of course, this is not Hume’s ultimate conclusion. In the second part of section VII, he goes on to provide a “sceptical solution” to his “sceptical doubt”, explaining how the the idea in question does after all have a source in an impression, but one that originates from an unobvious area of experience. He explains that the idea of a necessary connexion, in particular, derives from an impression of reflection we experience when the mind is determined “by habit or custom” to pass from the idea of a cause to the idea of its effect, where this “habit or custom” arises from observing constant conjunctions of similar causes and effects. This explanation vindicates the meaningfulness of the idea in terms of the principle about the meaningfulness of ideas, but is a “sceptical solution” because it implies that the originative impression of a necessary connexion does not have its source in any genuine feature of reality. According to the “sceptical solution”, there are in reality only relations of temporal priority, spatial contiguity, and constant conjunction. The idea of causation as an intrinsic causal link holding in reality is dismissed as a projection, a result of “the mind’s propensity to spread itself on the world”.

Let us set aside the issue of the plausibility of Hume’s “sceptical solution” and let us examine more carefully the argument of his “sceptical doubt”. As mentioned above, Hume criticises three hypotheses about ways we might receive impressions of causation. First, he argues against the view that we derive such impressions from our observations of the causal interactions of external objects. Secondly, he criticises the idea that we receive impressions of singular causation from our experience of exercising control over our bodies. Finally, he criticises the view that we receive impressions of singular causation from our experience of exercising control over our minds. 2

Against the first hypothesis that we can observe singular causation in the causal interactions among external objects, he gives an argument—in fact a master argument—that he uses, in slightly different formulations, against all the hypotheses about the observability of singular causation:

From the appearance of an object, we never can conjecture what effect will result from it. But were the power or energy of any cause discoverable by the mind, we could foresee the effect, even without experience; and might, at first, pronounce with certainty concerning it, by the mere dint of thought and reasoning. (VII, p. 63)

The form of this argument seems to be as follows. First premiss: if causation consists in an observable intrinsic link between events, then it should be possible for a person who has never previously experienced the causal relation to infer with certainty the effect from the cause. Second premiss: it is not possible to make such a priori causal inferences from cause to effect. Conclusion: causation cannot consist in such an observable intrinsic link. In other places, Hume supports the second premiss with an argument to the effect that we cannot make a priori causal inferences because cause and effect are distinct existences: any event can be conceived to follow from the occurrence of a given cause; and the only way in which a particular effect can be inferred from a cause is on the basis of experience, in particular by observation of a regularity holding between events of the same type.

Hume is undoubtedly correct in asserting the second premiss that inferences from causes to effects cannot be made on a purely a priori basis. It is simply not possible to predict what is going to result from a given cause without some experience of the relevant kind of causal relation. Causal inferences must, of course, be grounded in a posteriori knowledge. However, what is questionable is Hume’s assertion of the first premiss that if causation consists in an observable intrinsic link, it should support a priori causal inferences. This is extremely dubious. There are many ways of understanding intrinsic causal links—as necessary connexions, as relations of energy, power, or force between events. But under most of these conceptions it is not plausible to think of intrinsic causal links as transparent to a priori reflection in the way that Hume supposes they must be. Relations of energy, force, or power holding between events must be discovered a posteriori. It is true that if the intrinsic causal links consisted in traditional rationalist necessary connexions, they would be such as to support an a priori inference of effect from cause. But even here contemporary philosophers who have advanced the view that causal relations are necessary connexions usually reject the rationalist conception of them and see the necessary connexions as involving the kind of necessity that can only be discovered a posteriori. (See Armstrong 1983.)

So, the hypothesis that causal relations consist in observable intrinsic links does not necessarily imply that reliable causal inferences can be made without any experience whatever. Nonetheless, it does imply that one could, after a single observation of the relevant kind of causal relation, reliably predict the effect from the cause on another occasion. So, for example, having observed a single instance of an intrinsic causal tie between lighting a match and its catching light , or between choosing to raise one’s arm and raising one’s arm, or between deliberating about raising one’s arm and forming the intention to do so, one could reliably predict a similar effect as resulting from a similar cause on another occasion. The observation of the single instance of the relevant kind of causal relation would provide the a posteriori knowledge required to support the causal inference, but it would fall short of the regularity that Hume says is required. Hume effectively begs the question against this kind of position by assuming that the a posteriori knowledge must be knowledge of regularities. But there is no reason to think that single observations of causal relations cannot adequately provide the inferential support. Ultimately this is an empirical question: it must be determined by the psychological evidence whether we are able to observe causal relations and whether the single-case observations can support causal inferences. I shall address the psychological evidence in the next section. But the relevant point to make here is that Hume’s arguments against the a priori character of causal inferences do not carry any weight against the view that causal inferences can be given an a posteriori grounding in the single-case observations of intrinsic causal links.

Hume devotes the major part of his critique to the hypothesis that the idea of causation is derived from our experience of using our will to control our bodies and our thoughts. He repeats the master argument just discussed, but also invokes a number of independent arguments. Indeed, he adduces two sets of objections, one set directed against the view that the idea of causation is derived from the experience of exercising control over our bodies and the other set against the view that it is gained from the experience of exercising control over our thoughts. As these sets of objections are very similar, I shall only consider the objections to the first view in any detail.

In addition to the master argument that we cannot make causal inferences a priori but only on the basis of experience, he cites three further objections. The first objection is that it is a complete mystery how a spiritual substance like the will can exert an influence over the body:

Is there any principle in all nature more mysterious than the union of soul with body; by which a supposed spiritual substance acquires such an influence over a material one, that the most refined thought is able to actuate the grossest matter?... But if by consciousness we perceived any power or energy in the will, we must know this power; we must know its connexion with the effect; we must know the secret union of soul and body, and the nature of both these substances; by which the one is able to operate, in so many instances, upon the other. (VII, p. 65)

It is, indeed, mysterious how an immaterial substance can exert a causal influence over the body. However, the proponent of the view that we can perceive singular causation through the exercise of our will need not embrace a dualist conception of the will. Rather he could say that the whole process by which we exercise control over our bodies and our minds is a material process, capable of complete description in physical terms. This claim is compatible with the hypothesis that we can perceive the causal relations involved in the mechanisms of control. Indeed, I think that this is the most reasonable and viable version of the hypothesis in question. It is not a defect in this version of the hypothesis that ordinary subjects would not be aware of the physical mechanism whereby the will controls the body. For there are many phenomena which are perceivable, but nonetheless opaque to the understanding. Indeed, this seems to be of the very nature of many sensory experiences. This first objection, then, does not cut to the very heart of the hypothesis that we observe singular causation through the exercise of our wills, but touches only an inessential dualist rendering of it.

The second objection that Hume lodges is more to the point. It is that if we could perceive the causal power of the will over the body, we should be able to understand the scope and extent of the control of the will over the body. But evidently we are not able to do this.

We are not able to move all the organs of the body with a like authority; though we cannot assign any reason besides experience, for so remarkable a difference between one and the other. Why has the will an influence over the tongue and fingers, not over the heart or liver? This question would never embarrass us, were we conscious of a power in the former case, not in the latter. We should then perceive, independent of experience, why the authority of the will over the organs of the body is circumscribed within any particular limits. Being in that case fully acquainted with the power or force, by which it operates, we should also know, why its influence reaches precisely to such boundaries, and no further. (VII, p. 65)

On the contrary: that a range of experiences is sensorily available to us does not mean that the true scope or nature of the experiences is also disclosed to us. Sensory experience is not self-revelatory in the way that the objection assumes it should be. For example, we can perceive colours in a certain range of the electromagnetic spectrum but the reason why we can perceive colours only in that range is not revealed to us by our sensory experience. Again, the reason why human subjects can perceive sounds only within a certain frequency range cannot be retrieved by the untutored subject from his auditory experience. Why should we expect anything different in the case of our perception of the causal powers of the will?

Finally, the third objection that Hume raises is that if we could perceive the causal power of the will over the body, we should be able to perceive all the steps in the causal process connecting the volitional act and the bodily movement. But evidently we are not able to do this. Hume frames the argument in these terms:

We learn from anatomy, that the immediate object of power in voluntary motion, is not the member itself which is moved, but certain muscles, and nerves, and animal spirits, and, perhaps, something still more minute and more unknown, through which the motion is successively propagated, ere it reach the member itself whose motion is the immediate object of volition.... Here the mind wills a certain event: Immediately another event, unknown to ourselves, and totally different from the one intended, is produced: This event produces another, equally unknown: Till at last, through a long succession, the desired event is produced.... How indeed can we be conscious of a power to move our limbs, when we have no such power; but only that to move certain animal spirits, which, though they produce at last the motion of our limbs, yet operate in such a manner as is wholly beyond our comprehension. (VII, p. 66)

The questionable assumption in this objection is the first premiss. It is not obvious why, in being sensorily aware of the causal power of the will, we should be aware of the process by which the will exercises control over our bodily movements. Perhaps, in being aware of the causal connection between a volitional act and a bodily movement, we are simply aware that there is some or other process connecting the two events without being aware of the specific nature of the process. It is possible to be sensorily aware that two events are connected by a process without recognising all the stages of the process. One might see, for example, that the motion of one body imparts a certain momentum to another body after colliding with it and yet not see all the temporal stages of the process. This objection, like the second one, presupposes that the true nature of the objects of sensory experience should be completely transparent to sensory awareness. But, as remarked above, there is every reason to doubt this: the observability of causal relation is consistent with its having an essential nature which is hidden from our pretheoretic gaze. 3

To conclude this part of our discussion: we have seen that Hume’s objections to the various hypotheses about the observability of singular causation are less than conclusive. There are very plausible counters to these objections available to proponents of these hypotheses. This does not, however, establish the viability of any one of these hypotheses. To do this we need to consider whether there is any positive psychological evidence in favour of any of the hypotheses.

3. THE OBSERVABILITY OF INTRINSIC CAUSAL LINKS

Is any of the hypotheses about the observability of singular causation viable? Is there a source for an impression of causation? Can we observe causal relations? I want to take up these questions in this section of the chapter. However, before we can make any headway answering them, we must first consider what would make a causal relation observable. How exactly are we to understand the distinction between observable and unobservable states of affairs?

One commonsensical understanding of the distinction between observable and unobservable states of affairs is broadly functionalist in character. On this understanding, whether a state of affairs is observable or not is determined by the causal role that the state of affairs plays in the fixation of belief in normal human subjects. Thus, a state of affairs of given type is said to be observable, on this view, just in case a states of affair of that type typically causes the belief in normal human subjects that the state of affairs in question obtains and typically causes this by way of a distinctive causal path: the state of affairs in question must cause appropriate stimulation of the subjects’ sensory organs, which stimulation must directly cause the belief that the state of affairs obtains without the mediation of inference and without the aid of any additional collateral beliefs. Clearly, this characterisation presupposes that observation is a completely non-inferential and informationally encapsulated process: it involves no inference whatsoever and makes no appeal whatsoever to collateral belief. This understanding of observability, with its strong presuppositions about the character of observation, has appealed to some because it preserves the theory-neutrality of the observable world: no matter how much subjects differ in their inferential abilities and no matter how much they differ with respect to their theoretical commitments, observation will deliver up the same results to all of them so that the same states of affairs will count as observable to one and all.

For all its appealing simplicity, this characterisation of an observable state of affairs cannot be correct. For cognitive psychology tells us that the presupposition of this characterisation is almost certainly false; that is, it is almost certainly false that observation is a completely non-inferential and informationally encapsulated process. Contemporary cognitive psychology paints a quite different picture of perception, blurring the distinction between it and cognition. In the words of one authority, “perception involves a kind of problem-solving—a kind of intelligence” (Gregory 1970, p. 30). On the orthodox view of the matter, perception is the process wherein an organism assigns distal causes to the proximal stimulations it experiences. Since any given pattern of proximal stimulation is compatible with a great variety of distal causes, the organism has to engage in a non-demonstrative inference to work out the most probable distal cause. The kind of inference involved is not too different in kind from the kind of non-demonstrative inference that a scientist uses to infer the occurrence of a cause—the collision of a comet with the earth—from a pattern of effects—the extinction of the dinosaurs. What is more, in making the non-demonstrative inference, the perceiver has to call upon his collateral beliefs in order to determine a unique solution to a given perceptual problem. To eliminate sensory ambiguity, the perceptual solution to a given perceptual problem must be constrained not just by available sensory information, but also by the background beliefs the perceiver may bring to the task. What happens in perceptual processing, then, according to the orthodox view, is that sensory information is interpreted by reference to the perceiver’s background theories, the latter serving in effect to rule our certain causal etiologies as implausible causal histories for the given sensory array. So, it would seem that the lesson that contemporary cognitive psychology teaches us is that perceptual processing in its very essence is inferential and informationally unencapsulated in character.

How, then, are we to understand the distincton between observable and unobservable states of affairs if the commonsense assumptions about observation turn out to be mistaken? Well, though the commonsense assumptions are strictly speaking false, there is some grain of truth in them. As Fodor (1983; 1984) has emphasised, perceptual processes are not comprehensively penetrated by information: perceptual systems are modular in the sense that they can access only a restricted subset of the perceiver’s background beliefs. Fodor argues that this fact is brought home vividly by perceptual illusions such as the Mueller-Lyer illusion. This illusion is generated by two lines, placed one above the other, where one line has arrow heads at its ends pointing outwards and the other has arrowheads at its ends pointing inwards. Even though the lines are exactly the same length, perceivers perceive the line with the inward pointing arrowheads to be longer than the line with the outward pointing arrow heads. The textbook explanation of this illusion makes plenty of reference to the way in which perceptual analysis of the figures involves inference from background assumptions. But the important point—the point Fodor highlights—is that the perceptual analysis is not penetrated by all the background information available to the perceiver. For example, even when people learn, perhaps by measuring the lines for themselves, that they are the same length, they continue to see one line as being longer than the other. The belief that the lines are the same length does not affect the way the lines look to them. Indeed, what is true for this belief is true for many others too. As Fodor puts it, “how the world looks can be peculiarly unaffected by how one knows it to be. I pause to emphasise that the Mueller-Lyer illusion is by no means atypical in this respect. To the best of my knowledge, all the standard perceptual illusions exhibit this curiously refractory character: knowing that they are illusions doesn’t make them go away.” (1984, p. 242)

How does the process of perception look according to Fodor’s modularity hypothesis? Fodor suggests that we should conceptualise this process as involving three stages. In the first stage, sensory processes register the proximal stimulations that an organism encounters in its environment. These sensory processes are strictly non-inferential and insensitive to background information. In the second stage, perceptual mechanisms kick in to interpret the sensory states by assigning them probable distal causes. While thoroughly inferential in character, these perceptual processes are modular in the sense that they are responsive to only a select sort of information: their access to background information is sharply delimited by the character of the perceptual mechanisms. An important feature of the modular perceptual system is that it frames hypotheses about distal causes in a restricted vocabulary—the vocabulary of the accessible background theory. The terms of this background theory can be viewed as referring to observable objects, properties, and relations; and hypotheses framed in terms of this vocabulary can be seen as hypotheses about the observable appearances of things. The third stage in the perceptual process involves the comparison of these hypotheses with the rest of the perceiver’s background theory; and on the basis of this comparison, the perceiver fixes on a belief, a belief which, as cases of perceptual illusion illustrate, may differ from the hypothesis generated by the modular perceptual system. The relation between what one observes and what one believes is something like the relation between what one wants and what one wants on balance.

Fodor’s modularity hypothesis can help us frame a new understanding of the distinction between observable and unobservable states of affairs. The hypothesis suggests the following understanding of the distinction: a state of affairs of given type is observable (by human subjects) if and only if the appropriate modular perceptual system (of human subjects) delivers hypotheses that states of affairs of that type obtain. Therefore, if the obtaining of a causal relation is an observable state of affairs, then the relevant modular perceptual system must be capable of issuing the verdict that such a state of affairs holds. This would require, of course, that the vocabulary of the modular system—the language of the accessible background theory—would have to have terms denoting causal relations along with the all the other terms for the observable properties and relations.

Now we come to the main point. Are causal relations observable in this sense? Do the deliverances of the modular perceptual systems include information about causal relations? Is there a term denoting the causal relation in the language of the accesssible background theory of the modular systems? To be sure, these are empirical questions requiring detailed investigation of the workings of the human perceptual system. Very few philosophers—even among those who have maintained the observability of causal relations—have addressed these questions. Cognitive psychologists, on the other hand, have devoted a lot of energy to investigating them. To the extent that their work points in the same direction, it seems to point in the direction of an affirmative answer to these questions. It seems as though some causal relations are observable; a modular perceptual system issues in judgements about causal relations; and the language of this system includes terms for the causal relation. This is, at least, the way I read the general drift of the enormous psychological literature on the subject. In the following I summarise a small sample of the significant experimental results that support these conclusions.

Many of the psychological studies of the observability of causation have used the experimental framework developed by Albert Michotte. In his classic work (1963), Michotte argued that in certain experimental circumstances a causal relation between events can be immediately perceived on first exposure. A prime example of these circumstances is what Michotte called a launch event: in a launch event one object moves from left to right and collides with a second, stationary object; after the collision the first object becomes stationary and the second object moves away to the right. Michotte asked subjects what they saw in such launch events and classified their answers simply as causal or non-causal. Most subjects interpreted the launch event as causal. Michotte performed a large number of experiments designed to determine which stimulus conditions were required for the perception of causation. He found that the factors affecting the goodness of the causal impression all have to do with the unity of motion of the two objects. Foremost among these are spatial and temporal contiguity at the point of impact: if there were spatial or temporal gaps between the motions of the objects—between the first object’s collision with the second object and the second object’s moving away—subjects were less likely to interpret the event as causal. However, Michotte found that the basic conditions for the causal impression are not modifiable by experience: learning extra information about the event did not affect subjects’ reports of what they saw. Michotte himself interpreted the evidence as supporting the innatist hypothesis that a hardwired perceptual system, operating autonomously and independently of other systems, enables us to perceive causal relations in these circumstances.

We can also see Michotte’s work as supporting the conclusion that the perceptual system that delivers verdicts on launch events is informationally encapsulated to a large degree. This is brought out by the fact that subjects reported seeing a causal interaction between two objects in the launch event despite the fact that the stimuli used were just pencil marks on paper or coloured lights prorjected on a wall. The subjects knew full well how the displays were made and thus knew that there was no question of real objects really causing one another to move. But, as with the Mueller-Lyer illusion, this knowledge did not make the illusion go away. Despite the fact that the subjects knew the stimuli were not real objects, they still insisted that they looked as if they were causally interinteracting. The existence of this causal illusion and its incorrigibility in the face of what is known about the stimuli have been plausibly interpreted as showing that there is a modular perceptual system, operating in a fixed, automatic way and without access to all the information available to the subject, which is responsible for the perceptual verdicts about the launch events. This is the way the psychologist Alan Leslie sums up the situation. “I suggest that Michotte’s causal illusion results from the operation of an input module. The module applies a fixed algorithm and has access to only very restricted information, probably operating fairly early in the analysis of motion”. (1986, p. 414; see also his 1988)

Additional support for the idea that causal judgements about launch events are delivered by an informationally encapsulated perceptual system comes from experiments eliciting causal impressions from infants as young as six months old. (Leslie (1982); Leslie and Keeble (1987)) In these experiments the technique of habituation-dishabituation of looking was used to measure infant’s visual attention to various aspects of launch events. This technique allows the experimenter to measure the recovery of the infant’s interest following a change in a stimulus that has become familiar to the infant. From the patterns of renewed interest discernible over a number of experiments, it is possible to make inferences about the ways in which the infant is internally representing events. Alan Leslie and Stephanie Keeble (1987) report the results of experiments designed specifically to test whether infants perceive causal relations. In the experiments one group of six month old infants was habituated to a launch event involving a red object colliding with a green object, which was shown over and over again in a single direction. Meanwhile another group of infants was habituated to a sequence in which there was a delay between the collision of the red and green objects and the reaction of the green object. When a predetermined level of habituation was reached, the film the infants had been viewing was reversed so that the objects moved in the opposite direction. The reasoning behind the experiment was that reversing either event changes its spatiotemporal direction. But if the launch event is seen as causal, its reversal will change its causal direction as well. So if infants perceive causal direction only in the launch event but not in the delayed reaction setting, they will be differentially sensitive to its reversal. This is what the experimenters found in two experiments. Since spatiotemporal relations were controlled for, it would appear that that causal, as opposed to spatiotemporal, properties were involved in the infants’ differential reactions. Leslie and Keeble see the experimental data as providing evidence that infants, even at the age of six months, possess a visual mechanism which is responsible organising a causal percept. More particularly, they see it as supporting the idea that this visual mechanism is modular in character. They write: “The modularity of the device would enable it to operate independently of general knowledge and reasoning. If so, this would be ideal for a mechanism whose job might be to produce development and which may thus have to operate early in infancy when there is virtual absence of general knowledge and limited reasoning ability. Indeed, this might be the developmental significance of modular organisation.” (1987, p. 286)

Yet further support for the hypothesis of informational encapsulation comes from experiments performed by Anne Schlottman and David Shanks (1992). In their experiments launch events were embedded in event sequences where a second event—a colour change of the second object—was established as a competing predictor of the second object’s motion. The aim of the experiments was to see whether subjects’ learning of alternative predictive relationships would affect their causal impressions. Thus, in one experiment subjects were presented with launch events in which temporal gaps of varying lengths were interposed at the point of impact of the two objects so that the impact itself was not a reliable predictor of the second object’s motion. However, half of these trials contained a colour change of the second object which did reliably predict when it would move: the second object changed colour immediately before it moved off to the right. The results of the experiment showed that subjects’ causal judgements of causation were unaffected by the colour change: their causal impressions were determined solely by the degree of temporal contiguity at the point of impact.

Another experiment of Schlottman and Shanks was designed to test whether subjects’ perceptions of causation in launch events are influenced by the degree of correlation between events. It is known that in other contexts subjects’ causal judgements are sensitive to the degree of correlation between a cause and effect. For example, in one experiment subjects were asked to judge the extent to which pressing a key made a light come on; when the probability of the effect given the cause was held constant, judgements decreased as the probability of the effect in the absence of the cause was increased. So, one set of trials were arranged so that subjects were presented with launch events in which a correlation was established between the collision of the two objects and the second object’s moving; another set of trials were arranged so that there was no such correlation. The results of the experiment showed that the presence or absence of a correlation did not affect subjects’ judgements about causal structure of the launch event, though it did affect their judgements about whether the collision was necessary for the motion of the second object. The subjects drew a clear distinction in their judgements between facts about causation and facts about correlation in the launch events. Schlottman and Shanks conclude their discussion of the experiments with these remarks: “We take the results to argue for the plausibility of a distinct mechanism of causal perception....[The mechanism] would provide a robust intuitive understanding of the concept of cause that is not fraught with the ambiguities encountered when trying to define cause. This mechanism would allow us to recognise a deep distinction between merely correlated and causally connected events.” (p. 341)

To summarise this discussion: to the extent that there is a prevailing orthodoxy among cognitive psychologists, it is that there is a modular, informationally encapsulated system that is responsible for the perception of causal relations in special circumstances. This view is supported by evidence from a number of different sources. It is supported by the persistence and incorrigibility of causal illusions of the kind generated in experimental settings such as the launch event. It is supported by the experimental evidence that infants as young as six months old, possessing very little general knowledge of the world, are sensitive to the causal properties, as well as the spatiotemporal properties, of launch events. Finally, it is supported by the experimental evidence that the presence of alternative predictive relations does not affect people’s causal judgements about launch events. The significance of this view for our discussion is that it implies that some causal relations are perceivable. This consequence, backed by the weight of experimental evidence, contradicts a long empiricist orthodoxy that goes back to Hume.

Before I conclude this section, I want to mention and respond to a common objection to the position sketched above. It is often objected that cognitive psychologists’ notion of ‘perception’ is somewhat looser than, or at the very least very different to, the philosophers’ notion; and the fact that causal relations are sometimes perceivable in the psychologists’ sense does not establish that they are perceivable in the appropriate philosophers’ sense. (Tooley 1990; Sperber 1995) In particular, the psychologists’ sense of perception allows that one can perceive illusions, whereas the philosopher’s sense implies that one can perceive only what is real. This difference in usage becomes very evident in reactions to the causal illusions that are part of Michotte’s experimental set-ups. Psychologists may say that subjects perceive causal relations in these set-ups, but philosophers will see this as a violation of the minimal requirement on perception that it be veridical. So, it is pointless, according to this objection, to cite psychologists’ views about what subjects perceive in such situations, as they do not bear in the appropriate way on the proper philosophical sense of perceivability.

The force of this objection stems from the fact that in one usage ‘perceive’ is a success-verb: it is impossible to perceive a particular state of affairs if that state of affairs does not obtain. It is also true that psychologists’ use of the term brackets the success-connotation so that a person is said to perceive a particular state of affairs if the relevant perceptual module in the person delivers the verdict that it obtains, regardless of whether it really does. Nonetheless, I would argue, the question whether a state of affairs type (as distinct from token) is perceivable (as distinct from perceived)—for example whether the type of state of affairs that involves the causal relation is perceivable—is not concerned with the success-sense of perception at all. If it were a requirement on a type of state’s being perceivable that every perception of states of that type be veridical, then very few types of states would be perceivable. That is precisely the purport of the argument from illusion: since our perceptions of ordinary material objects are sometimes subject to illusion, we do not really perceive material objects at all, but only sense-data. But most contemporary philosophers regard the argument from illusion as mistaken; and mistaken precisely because it presupposes too stringent a requirement on perceivability. A more plausible requirement on the perceivability of a given type of state is that some perceptions of states of the type be veridical. But this implies that the kind of perception that can establish a given type of state of affairs as perceivable need not be veridical. This is precisely the kind of perception discussed by psychologists quoted above—the kind of perception delivered by fallible perceptual modules. Any special philosophical sense of perception involving guaranteed veridicality is quite beside the point here.

4. THE POSITIVIST CRITIQUE OF CAUSATION AS THEORETICAL ENTITY

As remarked above, the question whether singular causal relations are observable is an empirical issue; and the empirical evidence currently available strongly suggests that some causal relations are observable. But what if this evidence were controverted in the future? What if it were discovered that the cognitive mechanism responsible for delivering immediate reports of causation is not informationally encapsulated? Would Hume’s critique of the concept of causation as an intrinsic relation be vindicated?

I now want to argue that that would not be the case: even if we were to allow that causal relations are unobservable, we would not have to concede the force of the Humean critique. Let us suppose for the sake of the argument that the causal relation cannot be observed under any circumstances. Let us suppose, more particularly, that it is a kind of theoretical entity —like a quark, or a gene, or a species—whose character is given by the theory in which it appears—the folk theory of causation, as we may call it. What consequences would follow from Hume’s critique if this were true? Would we have to conclude that the concept of causation is meaningless, as Hume thinks?

We need to consider more closely the principle on which Hume relies for his affirmative answer. This is the principle that the only meaningful ideas are simple ideas that are copies of impressions, or complex ideas definable in terms of simple ideas that are copies of impressions. Expressed in this form, the principle will appear implausible to contemporary philosophers, presupposing as it does the outmoded theory of ideas. Contemporary philosophers will regard the theory of ideas as an inappropriate vehicle for expressing a criterion of meaningfulness because of its mistaken assumption that the primary bearer of meaning is an idea or mental image. Contemporary philosophy has rejected the whole ‘way of ideas’ and embraced the linguistic turn, according to which the primary bearers of meaning are linguistic entities such as words. Still, it is a simple matter to transpose Hume’s principle of meaningfulness into the linguistic key. This is what the early logical positivists did. In the role of the basic units of empirical significance, they substituted observational terms for Hume’s simple ideas that are copies of impressions; and so they proposed, in place of Hume’s principle that every meaningful idea is either a copy of an impression or is definable in terms of ideas that are copies of impressions, the principle that every empirically significant term is either an observational term or is definable in terms of observational (and logical) terms. (Carnap 1928; Schlick 1936) Apart from the substitution of observational terms for simple ideas, the two formulations are basically the same.

In terms of this formulation, Hume’s criticism of causation as an intrinsic relation—a necessary connexion, a relation of power, force, or energy—can be reconstructed as the criticism that the terms standing for such relations do not belong to the observational vocabulary, nor can they be defined in terms of observational (and logical) vocabulary. Hence, the concept of such a relation is not empirically significant.

Formulated thus, the criticism can be seen to be seriously flawed. As the later positivists came to see, the criterion of empirical signifcance it relies on is far too stringent, as it implies that the theoretical terms of scientific theories are not empirically significant. Carnap (1936-7) pointed out that if the logic of the acceptable definitions is restricted to first-order quantificational logic, definitions of theoretical terms standing for even the simplest dispositions are ruled out. For example, the obvious definition of ‘x is fragile’ as ‘For all times t, x is struck at t ? x breaks at t’ does not work because it unacceptably implies that any object which is never struck is fragile. In response to this difficulty, Carnap stipulated, as a condition of intelligibility, that each theoretical term should be linked to observational terms, not by logically necessary and sufficient conditions, but reduction sentences such as ‘For all times t, t if x is struck at t then x is fragile at t = x breaks at t’. Such reduction sentences state sufficient, but not necessary, conditions for the application of theoretical terms and so involve a weakening of the criterion of empirical significance. But this weakening did not go far enough. Positivists such as Hempel (1952) argued that it was a mistake to require that meaningful theoretical terms be individually definable by reduction sentences. Metrical theoretical concepts such as ‘mass’, ‘force’, ‘rigid body’, ‘electron’ etc are not, Hempel argued, introduced by reduction sentence, nor by any piecemeal process of assigning meaning to them individually; rather systems of such theoretical terms are introduced collectively by way of a set of theoretical postulates and this set has its meaning conferred on it, as a unit, by means of an empirical interpretation in terms of correspondence rules. These rules consisted of sentences stating analytic entailments between one or more theoretical terms and a number of observation terms. Although the correspondence rules did not provide definitions of the theoretical terms, they were said to provide a partial interpretation of them. The availability of a partial interpretation was thought to confer empirical significance on a theoretical term. On the basis of these considerations, the most sophisticated positivists simply required of an empirically significant theoretical term that it should belong to a system of theoretical terms that could be given a partial interpretation in terms of a set of correspondence rules. (See Hempel 1965; Scheffler 1963, Suppe 1977)

Is the Humean critique more successful when reformulated in terms of this weakened criterion? I would argue not. While the weakened criterion of empirical significance is more reasonable, the reformulated critique does not succeed in showing that the concept of causation is meaningless. The critique runs that the term for the causal relation, conceived of as a local tie between events, cannot be given a partial interpretation; in other words, it is not possible to formulate correspondence rules linking the term for the causal relation with appropriate observational terms. But what reason is there for supposing that this is not possible?

Suppose that the term for the causal relation denotes a particular kind of intrinsic relation, say a relation of necessary connexion which can hold in particular instances. This is a very contentious view which I do not wish in the end to embrace, but it is useful to fix on it for the purposes of illustration. It does not follow from the weakened criterion that the term for this relation is not empirically significant. For there may well be correspondence rules in the form of analytic implications linking this term, perhaps taken in conjunction with other theoretical terms, with certain observational terms. For example, it is reasonable to think that the existence of a necessary connection between cause and effect will analytically imply that, other things being equal, events of the same type as the cause will be followed by events of the same type as the effect. Of course, the converse implication does not hold, because it takes more than a constant conjunction to make up a necessary connexion. All the same, some such analytic implication is enough to satisfy the weakened criterion of empirical significance; so the term standing for the causal relation, construed as necessary connexion, need not be seen as meaningless. The same point could be made with other construals of the causal relation as a relation intrinsic to its pairs—the construals of it as a relation of energy, force, or power—since the holding of a causal relation, under all these construals, have consequences at the observational level.

As far as I know, neither Carnap nor Hempel—the two positivists who produced the most sophisticated criteria of empirical significance—actually applied the criteria to the term for causation. They were, nonetheless, very chary of the conception of causation as consisting in a local, intrinsic tie between events: in the few places they discuss this conception they reject it in favour of the Humean regularity conception of causation. Why did they reject it? Our preceding discussion shows that they were mistaken to have rejected it on the grounds of its not being empirically significant. To be sure, it is not significant according to the criterion requiring a term to be completely definable in terms of observational and logical vocabulary. But this criterion is implausibly strong, as it implies that any term standing for a theoretical entity is without empirical significance. When the criterion is weakened to the more reasonable requirement that a theoretical term should be partially interpretable in terms of correspondence rules, there is good reason for thinking that the term for the causal relation, conceived as an intrinsic relation, passes the criterion. In conclusion, the positivists should not have rejected the conception of causation as an intrinsic relation on the basis of a criterion of empirical significance.

5. CONCLUSION

The idea that causation consists in intrinsic relation of some kind is a very intuitive one. Despite its considerable influence, Hume’s critique of this idea is very far from being conclusive. I have argued that the critique is in fact mistaken on two counts. First, the critique mistakenly asserts that there is no impression, no source in empirical experience, corresponding to the idea of intrinsic causal links. Citing the experimental work of a number of cognitive psychologists, I have claimed that we do perceive intrinsic causal relations in certain favourable circumstances. Secondly, Hume’s critique mistakenly relies on a criterion of meaningfulness stating that the only meaningful ideas are simple ideas that are copies of impressions, or complex ideas definable in terms of such simple ideas. (A more up-to-date linguistic version of this criterion states that the only empirically significant terms are either observational terms or terms definable in terms of observational (and logical) terms. ) This criterion of meaningfulness is, I have argued, hopelessly strong. When it is weakened to attain some plausibility, there is good reason to think that ideas of intrinsic causal links might satisfy it. My overall conclusion is that, despite the enormous influence Hume’s “sceptical doubt” about intrinsic causal relations has had on contemporary philosophers, it is essentially flawed. This is one area of philosophy where Hume was hampered by the extreme verificationism of his psychological version of empiricism.

BIBLIOGRAPHY

Anscombe, G. E. (1975). ‘Causality and Determination’. In Causation and Conditionals (ed. E. Sosa). Oxford University Press, Oxford, 63-81.

Armstrong, D. (1968). A Materialist Theory of Mind. Routledge & Kegan Paul, London.

Armstrong, D. (1983). What is a law of Nature? Cambridge University Press, Cambridge.

Armstrong, D. (1997a). A World of States of Affairs. Cambridge University Press, Cambridge.

Armstrong, D. (1997b). ‘The Open Door: Counterfactual vs. Singularist Theories of Causation’. In Laws and Causation: Australasian Studies in History and Philosophy of Science, (ed. H. Sankey). Kluwer.

Berkeley, G. (1710). The Principles of Human Knowledge. In Berkeley’s Philosophical Writings. (1965, ed. D. Armstrong). Collier Books, London.

Bigelow, J. and Pargetter, R. (1990). Science and Necessity. Cambridge University Press, Cambridge.

Carnap, R. (1928). Der Logische Aufabau der Welt. Welkreis-Verlag, Berlin.

Carnap, R. (1936-7). ‘Testability and Meaning’. Philosophy of Science 3, 420-468; 4, 1-40.

Fair, D. (1979). ‘Causation and the Flow of Energy’. Erkenntnis 14, 219-50.

Fales, E. (1990). Causation and Universals. Routledge, London.

Fodor, J. (1983). The Modularity of Mind. MIT Press, Cambridge, Mass.

Fodor, J. (1984). ‘Observation Reconsidered’. Philosophy of Science 51, 23-43.

Gregory, R. (1970). The Intelligent Eye. McGraw-Hill Book Company, New York.

Hempel, C. (1952). The Fundamentals of Concept Formation in Empirical Science. University of Chicago Press, Chicago.

Hempel,. C. (1965). Aspects of Scientific Explanation and Other Essays in the Philosophy of Science. Free Press, New York.

Hume, D. (1739). A Treatise of Human Nature. (Second edition, 1978, ed. P. Nidditch). Clarendon Press, Oxford.

Hume, D. (1748). An Enquiry concerning Human Understanding. (Third edition, 1975, ed. P. Nidditch). Clarendon Press, Oxford.

James, W. (1912). Essays in Radical Empiricism.

Leslie, A. (1982). ‘The Perception of Causality in Infants’. Perception 11, 173-86.

Leslie, A. (1986). ‘Getting Modularity off the Ground: Modularity and Infant’s Perception of Causality’. In Theory Building in Developmental Psychology (ed. P. van Geert). Elsevier Science Publishers, Holland.

Leslie A. and Keeble, S. (1987). ‘Do Six-Month-Old Infants Perceive Causality?’. Cognition 25, 265-88.

Leslie, A. (1988) ‘The Necessity of Illusion: Perception and Thought in Infancy’. In Thought without Language (ed. L. Weiskrantz), pp. 185-210. Oxford Science Publications, Oxford.

Lewis, D. (1983). ‘New Work for a Theory of Universals’. Australasian Journal of Philosophy 61, 343-377.

Lewis, D. (1986). On the Plurality of Worlds. Blackwell, Oxford.

Locke, J. (1690). An Essay concerning Human Understanding. (1965, ed. A. D. Woozley). London.

Mackie, J. (1974). The Cement of the Universe. Clarendon Press, Oxford.

Menzies, P. (1997). ‘Intrinsic versus Extrinsic Conceptions of Causation’. In Laws and Causation: Australasian Studies in History and Philosophy of Science, (ed. H. Sankey). Kluwer.

Michotte, A. (1963). The Perception of Causality. Methuen, London.

Reid, T. (1785). Essays on the Intellectual Powers of Man. In Thomas Reid: Inquiry and Essays, (1983, ed. R. Beanblossom and K. Lehrer), Indianapolis.

Scheffler, I. (1963). The Anatomy of Inquiry. Knopf, New York.

Schlick, R. (1936). ‘Meaning and Verification’. Philosophical Review 45, 339-369.

Schlottman, A and Shanks, D. (1992). ‘Evidence for a Distinction between Judged and Perceived Causality’. Quarterly Journal of Experimental Psychology 44A, 321-342.

Searle, J. (1983). Intentionality. Cambridge University Press, Cambridge.

Sperber, D. (1995). ‘Introduction’. In Causal Cognition (ed. D. Sperber, D. Premack, A. Premack). Clarendon Press, Oxford.

Suppe, F. (1977). The Structure of Scientific Theories. Second edition. University of Illinois Press, Urbana, Illinois.

Tooley, M. (1990). ‘The Nature of Causation: A Singularist Account’. In Canadian Philosophers: Celebrating Twenty Years of the CJP (ed. D. Copp), Canadian Journal of Philosophy, suppl. 16, 271-322.