2.1. The CTA's way of accounting for (Intention)

Let me begin with a brief discussion of the CTA and how it understands (Intention), i.e. the idea that, in acting intentionally, one's intention is the cause of what it represents. As I understand it, the CTA purports to give a reductive account of intentional action. This is an attempt to understand intentional action by stating necessary and jointly sufficient conditions for the existence of such action; it is reductive in the sense that the notions deployed in these conditions are taken to be more fundamental than and independently intelligible of intentional action itself. So, if we apply this sort of explanation to (Intention), we arrive at the following account of intentional action:

This is the core of the sort of account that I call ‘the CTA’.Footnote ³ To be sure, how both conditions are to be spelled out in any more detail is highly controversial. For instance, proponents of the CTA notoriously differ over what kind of mental state(s) must be part of a successful account of intentional action (e.g. whether or not intention can be reduced to some combination of belief and desire). They also differ as to which further condition(s) may be necessary to yield a sufficient account of intentional action. (I will discuss some proposals below.) But however significant the differences between particular versions of the CTA may be, they are unified by being precisely versions of the basic idea that, in acting intentionally, the agent's intention must both represent and cause her movement, where the relevant notions of causality and representation are taken to be more fundamental than and independent of intentional action. What's important for my argument in this paper is, if you like, the general structure of the account that the different versions of the CTA have in common, not the differences in content by which they are distinguished. So let me say a little more about the structure of this account.

First of all, the above statement of the CTA looks so close to (Intention) that one may wonder what the alleged difference between them really comes to. To see the difference, it is important to appreciate that the CTA applies a reductive account to (Intention), but that, just by itself, (Intention) doesn’t force such an account on us. To illustrate this, consider a more familiar point from epistemology: one may acknowledge that knowledge involves justification, truth, and belief, without thereby being committed to a reductive account of knowledge, where this means that justification, truth, and belief are more fundamental than the notion of knowledge itself. (An example is Williamson's [Reference Williamson2000] knowledge-first approach.) Similarly, one may acknowledge that intentional action involves causation (as (Intention) claims), without thereby being committed to the view that causation is the more fundamental notion here (as the CTA claims). I will say more about what motivates the CTA's reductive approach below; for now it's just important to note the difference between (Intention) and the CTA.

Furthermore, it is crucial to my argument that the CTA takes the notions of causality and representation it deploys in its account to be intrinsically independent of one another. By ‘intrinsically independent’ I mean that, just as such, the obtaining of one of these relations – the causal and the representational relation – doesn’t imply or require the obtaining of the other. That is to say, just by itself, the fact that a movement is caused by the intention doesn’t imply or require that the representational condition is satisfied as well. It may be that the movement caused by the intention is not the movement represented by the intention. And, conversely, it may be that the intention represents a movement without being its cause. After all, if in fact causality implied or required representation (or vice versa), both conditions wouldn’t be separately satisfiable. Consequently, causality and representation couldn’t obtain – or could obtain only in some necessarily defective form – outside their unity in intentional action. But then, it seems these notions would no longer be more fundamental than what they purport to explain, and thus, the account would no longer be reductive in the relevant sense. Hence, I take it that the CTA's reductive ambition commits its proponents to the intrinsic independence of causality and representation.

To be sure, this is not to say that the CTA must deny just any connection between causality and representation. On the contrary, as I shall discuss at great length in Sections 4 and 5, proponents of the CTA often require that, for there to be an intentional action, there must also be a modal connection between an intention's causality and its content, such that if the intention's content had been different, it would have caused a different movement. What I take the CTA to be committed to is just that the connection cannot be intrinsic in the sense explained. For, if it was, causality and representation would be of a special kind, the kind that had to be understood through their unity in intentional action. And this, I take it, is incompatible with the reductive aim of explaining intentional action in terms more fundamental than what they purport to explain.

Davidson, as we have noted, identified the mental cause of an intentional action with a belief-desire pair. On his view, we thus get the following version of the CTA:

The logical and the causal relation of which Davidson speaks correspond to the two core conditions of the CTA, the representational and the causal condition. That both relations are ‘very different’, as Davidson says, reflects the view that they are intrinsically independent. Furthermore, Davidson states that the representational relation has a normative significance: the agent's thoughts represent her movement as something desirable or valuable to do.Footnote ⁴ On Davidson's view, then, for a movement to be an intentional action it must stand in two ‘very different’ relations to the agent's thought: a representational (or rationalizing, or justificatory) and a causal relation.

2.2. Assumptions of the CTA

It is sometimes suggested or just assumed that (Intention) implies the truth of some version of the CTA: maybe not Davidson's belief-desire-model, but then some other more sophisticated version of the CTA. However, as I have already noted, I think the CTA is neither identical with nor entailed by (Intention), i.e. the bare idea that, in acting intentionally, one's intention is the cause of what it represents. We have to distinguish between (Intention) and the particular way in which the CTA attempts to account for it. However, one may wonder why the transition from the one to the other seems so natural and even inevitable to so many people. This becomes clearer if we consider what I take to be the two main motivating assumptions of the CTA. The first one concerns the notion of causality, the second concerns the notion of content. I will briefly discuss each one in turn.

The first assumption is sometimes explicitly stated by proponents of the CTA: ‘The [CTA's] perspective on agency is largely motivated by the worthy desire for a uniform understanding of causality’ (Bishop Reference Bishop1983, 63; my italics).Footnote ⁵ On a uniform understanding of causality, there is only one kind of causality. The causality that binds my action to my intention is of the same kind as the causality that governs the movements of non-human animals or even inanimate objects. When I explain my action by saying ‘I am doing A because…’ and cite my intention or my reason for doing A, the ‘because’ refers to a causal connection which as such is no different from the one I refer to in explaining: ‘The house is burning down because it was hit by lightening’. Thus, on this view, the causal role of my intention is specifiable using a notion of causality that also binds fire to lightening, or, notoriously, the movements of one billiard ball to the movements of another. Obviously, then, it can make no intrinsic difference to the causal relation in which my intention stands to what I am doing that the items thus connected also stand in a relation of representation (or rationalization). A prominent example of a uniform understanding of causality is Davidson's (Reference Davidson, Lawrence and Swanson1970) nomological view of causality, according to which it is part of the very idea of causality that causally connected events can be subsumed under a natural law (under some description). Thus, ultimately, any causal explanation is made true by the same fundamental kind of causality, the one described in the relevant natural law(s). The crucial point is that, on this view, the kind of causality is the same whether the items causally connected are an intention and an action or just two billiard balls.Footnote ⁶

The second assumption of the CTA is less often explicitly stated; nevertheless, I think it is as crucial as the first one. We can call this assumption a uniform understanding of content. On this view, there is only one kind of conceptual content that is uniform across theoretical and practical thought (e.g. belief and intention). Both kinds of thought are to be distinguished, not in terms of different kinds of content, but in terms of their different causal roles.Footnote ⁷ Very roughly, the basic idea is that, e.g. an intention is a practical thought in that it tends to cause what it represents, whereas a belief is a theoretical thought in that it is characteristically caused by what it represents – however mediated this causal dependence might be (see, e.g. Smith [Reference Smith1987, 55] for a clear statement of this view). According to this view, this distinction in terms of causal roles does not imply any difference in the respective kinds of content. So, given this view of content, it makes no intrinsic difference to the content of my practical thought that it is the content of a specifically practical thought, i.e. a thought that tends to cause what it represents.

Now, once such uniform notions of content and causality are in place, the CTA does indeed become the inevitable way of accounting for Davidson's insight. For, given these notions, we have to conceive of an intention's conceptual content and its causal role as two intrinsically independent components. But, I think we are now also in a position to better appreciate why the CTA is neither identical with nor entailed by Davidson's insight that, in acting intentionally, one's intention is the cause of what it represents. For, just as such, this latter idea doesn’t seem to involve or depend on any commitment to a uniform understanding of content and causality. Recall Davidson's argument for the claim that reasons – or, in our version, intentions – are causes: to exclude the possibility that one's movement conforms to one's intention by mere accident, one's intention must be the cause of the movement it represents, in the sense that it explains the existence of this movement. But this argument by itself doesn’t imply a uniform understanding of causality; that is, it doesn’t imply that the causal relation binding one's movement to one's intention is not intrinsically different from the causal relation that, say, binds fire to lightening. If this is right, one can accept Davidson's argument and the need to conceive of intentional action along the lines of (Intention) without thereby being committed to the CTA.Footnote ⁸

Now, what I will argue in the rest of this paper is that it is not only necessary to distinguish between (Intention) and the CTA's way of accounting for it, but that given the CTA's understanding of (Intention), it actually becomes quite unclear how we can account for (Non-Accidentality). This, then, will motivate the exploration of a different way of understanding (Intention), one that rejects the CTA's underlying assumptions.

3.1. A problem for the CTA's way of accounting for (Intention)

To bring out the difficulty facing the CTA, consider again that, on its view, in acting intentionally, an agent's intention must both represent and cause her movement, where both of these facts are intrinsically independent of one another. That is to say, considered as such, there is nothing in the notion of causality that requires the presence of representation; and, correspondingly, there is nothing in the notion of representation that requires the presence of causality. But, if that's so, it seems there is nothing that explains why it's not just an accident if representation and causality come together in the right way, namely so that what the intention causes is the movement it represents. In a case where the intention to do A causes one to do A, both conditions of the CTA are simultaneously satisfied: the intention both represents and causes one's doing A. But, given that content and causality are intrinsically independent of one another, how can this be anything but a mere fluke?

That this really is a problem can be illustrated by Davidson's (Reference Davidson and Ted1973, 79) famous example of the nervous climber. In this example, a climber wants to rid himself of the weight and danger of holding another man on a rope, and believes that loosening his hold on the rope will accomplish this. As a result, he forms the intention to loosen his hold, which, however, makes him so nervous that he loosens his hold unintentionally. In this case, the intention both represents and causes the climber's loosening his hold, yet no intentional action is taking place. Why not?

Arguably, this is so because it is a mere accident that the climber's intention causes the movement it represents (or, equivalently, that it represents what it causes). For, in this case, the intention's causality operates independently of the fact that it represents its effect. To make this more vivid, suppose the climber's intention had been different – say, he had intended, not to loosen his hold, but to cut the rope on which his partner dangles. Presumably, this would not have made a difference to the resulting behavior. Given the climber's nervousness, it is plausible to assume that both intentions would have had the same effect, namely the climbers loosening his hold. This means that, under the circumstances, it is not the content of the climber's intention that is relevant to his behavior. And in this sense it is just an accident that the behavior caused by the climber's intention accords with the intention's content. So, what the example helps to bring out is that, even though both conditions of the CTA are satisfied, it may still be an accident that the agent's intention causes what it represents – and in that case, no intentional action is taking place.

Now, one might think this is a marginal case, one that can be put aside in the CTA's pursuit of an account of the standard case. But this would miss the point of the example. For, given the CTA's assumption that an intention's content and its causality are independent of one another, the problem is that nothing seems to rule out the possibility that in any given case things are just like in Davidson's example, namely that the agent's intention causes what it represents purely by accident. Thus, what is called into question is precisely the CTA's entitlement to distinguish in this way between ‘marginal’ and ‘standard’ cases. From the perspective of the CTA, any case looks like one in which it is just mere luck that the intention's effect coincides with what the intention represents.

It thus starts to look quite puzzling how, given the CTA's way of understanding (Intention), an intention's causality is supposed to account for (Non-Accidentality). For, if it really is just an accident that one's intention causes the movement it represents – and, on the CTA's view, it is hard to see how this can be otherwise – then, even though one's intention is the cause of one's movement, this doesn’t ensure a non-accidental match between this movement and the content of one's intention. In Davidson's example, it is just an accident that the climber ends up doing what he intended to do – even though his movement is caused by his intention. But this means that, on the CTA's account of (Intention), the intention's causal role seems insufficient to account for the fact that, in acting intentionally, it is no accident that one is doing what one intends to do.

We may put this by saying that, strictly speaking, there are two sources of accidentality and, thus, two ways of violating the non-accidentality requirement on intentional action. First, this requirement is violated if the cause of the ‘matching’ behavior is something other than one's intention. But, second, this requirement is also violated if the causality of one's intention operates independently of its content. A proper account of the non-accidentality requirement needs to exclude both aspects of accidentality. However, this is not to suggest that they can be dealt with separately. For, as we have just seen, even if the cause of the ‘matching’ behavior is the intention itself, this doesn’t render the ‘match’ any less accidental as long as one hasn’t also eliminated the second source of accidentality. That is, a proper account of the non-accidental ‘match’ between movement and intention requires an account of the non-accidental connection between the intention's causality and its content.

So, I think what we can conclude is this. When one acts intentionally, it is no accident that one is doing what one intends to do. But for this to be no accident, it is not enough that one's intention both represents and causes what one is doing. Both, the fact that the intention is a representation of one's movement and the fact that it is the cause of that movement cannot be wholly independent of one another. Rather, there must be some sort of connection between these facts, such that it is no accident if one's intention causes what it represents. The challenge can be further clarified by relating it to the problem of causal deviance – of which Davidson's nervous climber is usually taken to be the prime example.

On the present interpretation, what's going wrong in the example is that it is just an accident that the intention causes what it represents. Thus, what the example illustrates is the need for an account of the non-accidental connection between the intention's causality and its content. On my view, then, the problem of causal deviance is just an example of a more general problem: the problem of how to account for the non-accidentality requirement on intentional action. Now, if this diagnosis of the problem is correct, then it seems that many attempts to solve it are misguided from the very beginning: namely those attempts that fail to address the question of how causality and content must be connected in order to meet the non-accidentality requirement. Consider, for example, the so-called causal immediacy strategy (see, e.g. Brand [Reference Brand1984, 20] and Mele [Reference Mele1992, 202]). The general idea behind this strategy is that solving the problem requires eliminating potentially ‘deviant’ events (like nervousness or excitement) from the causal chain linking behavior to intention. And it is claimed that this can be achieved by stipulating that the intention proximately causes the behavior, without any mediating causal links. Now, setting aside any question as to whether this strategy can be successful in its own right, it should be clear that it doesn’t help with our problem. For, even assuming that it succeeds in eliminating nervousness from the causal chain, this by itself does nothing to show that it is not just an accident that, in addition to being its cause, the intention also represents the behavior. What we are interested in is not how the causal chain is made up, but how it is related to the fact that the intention is a representation of the behavior.Footnote ⁹ In other words, what we need is not a specification of either the causal or the representational condition, but a specification of how they are connected, such that their co-presence is no accident.

3.2. Ways of solving the problem

Once we have a clear understanding of our problem, I think three ways of solving it suggest themselves, depending on how the non-accidental connection between causality and content is spelled out: the connection may be conceived of in modal, functional (or teleological), or constitutive terms. Corresponding to each of these accounts will be a different conception of accident. On the modal account, the relevant sort of accident is excluded if and because the agent is responsive to the content of her intention, not just in the actual situation, but also in a specified range of non-actual or counterfactual situations. Alternatively, it may be argued that it is no accident that the intention causes what it represents if and because doing so is its function, where this means that the intention's causal efficacy is in part explainable by the fact that it tends to bring about what it represents. Finally, the non-accidental connection between content and causality may be understood in constitutive terms, such that an intention is a kind of cause to which it is essential that it is a representation of its effect. Clearly, the constitutive account is incompatible with the CTA: it amounts to a rejection of the assumption that content and causality are intrinsically independent of one another. By contrast, I take the first two accounts to be compatible with the CTA. This is rather obvious for the modal account. After all, on this view, whether it is an accident that the intention causes what it represents is ultimately just a matter of the frequency with which it does so in a specified range of possible situations. It may be less obvious for the functional account. But here, everything depends on how the notion of function is itself understood, and I take it that a naturalistic understanding of this notion – one that appeals to natural selection – is indeed compatible with the CTA's commitments.

May there be other ways of spelling out the relevant non-accidental connection? Some have suggested that an intention must be causally efficacious in virtue of its content (see, e.g. Schlosser Reference Schlosser2007). However, I think that, as such, this is just another way of saying that there must be a non-accidental connection between causality and content. If the ‘in-virtue-of’ relation is not just a placeholder – much like the appeal to being caused ‘in the right way’ – it must somehow be fleshed out. And, here again, modal, functional, and constitutive considerations suggest themselves as the most obvious candidates. Thus, in the end, I think that an appeal to a grounding relation such as ‘in virtue of’ doesn’t get us anywhere we haven’t already been with the options we are going to consider.

So, I think that at the heart of the issue is the question of how to understand the relevant sort of accident. In what follows, I will argue for the constitutive account of accidentality by criticizing the modal as well as the functional strategy. Consequently, if I am right, there is good reason to think that the non-accidentality requirement cannot be accounted for within the framework of the CTA.Footnote ¹⁰

4.1. Modal condition I

The first version of the modal condition corresponds to what has been called the sensitivity strategy.Footnote ¹² It requires that the agent's behavior is responsive to the content of the intention that causes the behavior in the following sense: if the content of the agent's intention had been different, the resulting behavior would have differed correspondingly. If the agent satisfies this condition, it is no accident that the caused behavior conforms to the content of her intention. For instance, Davidson's climber does not meet this condition, because, as we have seen, it is not true that, if the content of his intention had been different, the resulting behavior would have been different too. However, as it stands, the condition is clearly too strong. For, in order to act intentionally, it is certainly not necessary that in just every counterfactual situation in which one had intended something else, one would have acted in conformity with this different intention. After all, innumerable things may happen and prevent one from responding successfully to one's intention. Consider a different version of Davidson's example. Suppose the climber isn’t nervous at all; instead, he goes through with his intention to loosen the hold on the rope in a very cold-blooded manner. Now, suppose further that, if his intention had been different – e.g. if he had intended to pull up the rope and save his companion – some external force (an evil demon, a falling rock, a sudden heart attack) would have prevented the climber from acting on this different intention. In a case like this, the relevant counterfactual is false: it is not true that, if the content of the climber's intention had been different, his behavior would have differed correspondingly. Nevertheless, this doesn’t show that, in the actual situation, the climber didn’t act intentionally. Hence, at least as it stands, the sensitivity condition doesn’t seem necessary for intentional action.Footnote ¹³ But, since the relevant sort of non-accidentality is necessary for intentional action, sensitivity seems inadequate to account for it.

Still, there seems something deeply right about the sensitivity condition. Normally, when someone acts intentionally, we do expect that, if she had intended something else, her behavior would have differed accordingly. It's just that we don’t expect this no matter what. We only expect this in circumstances where nothing interferes with the agent's ability to act on her intention. Thus, clearly, if the relevant counterfactual stands a chance of being true, its scope must be restricted to circumstances in which possible interveners are absent. Yet, the question is how to accomplish this restriction. It seems hopeless to produce a list that includes all possible interveners. Since there is no obvious limit to the things that might interfere with the agent's ability to carry out her intention, such a list would likely be endless. Certainly, the more natural move is to amend the counterfactual with a ceteris paribus clause. However, it is not clear that this move is compatible with the reductive ambition of the CTA. For, as is well known, the introduction of such a clause gives rise to the problem that it seems impossible to understand the clause in a non-circular way.

To see this, consider the events that would have to be excluded by the ceteris paribus clause. Among those events are falling rocks, heart attacks, and evil demons. Now, ask yourself, what do these events have in common with each other such that they are all covered by the ceteris paribus clause? It seems difficult to answer this question other than by saying something like: ‘each of these events prevents the agent from properly realizing her intention’. Thus, what justifies grouping these events together can be specified in only negative terms: they all interfere with the agent's ability to act intentionally. This is the respect in which events must resemble one another if they are to be covered by the ceteris paribus clause. But, if that's so, it is impossible to specify the relevant respect of similarity without making reference to the notion of intentional action itself. An understanding of the ceteris paribus clause would thus contain and depend upon an understanding of acting intentionally. This means that, if we render the relevant counterfactual true by introducing a ceteris paribus clause, we thereby render it unavailable to a reductive account of intentional action. As I said, I don’t want to deny that a counterfactual of the relevant sort is true when one acts intentionally. What I claim is just that a proper understanding of the counterfactual is not more fundamental than the notion of intentional action itself.

Arguably, there are other, perhaps more sophisticated versions of the sensitivity strategy. Recently, for example, it has become popular to claim that, in acting intentionally, an intention must not just trigger but continue to guide a movement through the course of its occurrence.Footnote ¹⁴ It is tempting to understand this guidance-condition as an attempt to account for the non-accidentality requirement. Thus, for it to be no accident that one is doing what one intends to do, one's intention must cause one's movement in such a way that it is guided through the course of its occurrence. But what does ‘guidance’ mean here? A natural interpretation appeals to a certain sort of counterfactual: one's movement was guided in the relevant sense just in case, had the successful completion of one's movement been in danger, there would have been a corresponding correction of one's behavior.Footnote ¹⁵ Suppose, for instance, in letting go of the rope, Davidson's climber was in fact guided by his intention to do so. Then, a counterfactual of the following sort is true: had he encountered an obstacle – e.g. had the rope been stuck – he would have corrected or modified his behavior accordingly, e.g. he would have released the rope.

The problem with this account is the same as the one we just discussed. As it stands, the relevant counterfactual is too strong. Imagine that, in the actual situation, the climber didn’t encounter any obstacles, but if he had, some external force would have prevented him from correcting his behavior. In this case, the counterfactual is false, but this does not imply that, in the actual situation, he didn’t act intentionally. So, to save the account from counterexamples, one has to amend the counterfactual with a ceteris paribus clause. But then, again, the account threatens to become circular. For, it is difficult to see how the set of events covered by the ceteris paribus clause can be specified other than in negative terms: as those events that interfere with an agent's intentional agency.

4.2. Modal condition II

Now, let's turn to the second version of the modal condition we distinguished above – the one where the agent has the same intention in the relevant non-actual situations as in the actual situation. In spelling out this condition, we can draw on some recently influential work in epistemology, according to which the notion of an accident is understood in terms of possible worlds. In the words of Duncan Pritchard (Reference Pritchard2005, 130–31), the basic idea is this: ‘when we describe an event as an accident we thereby typically imply that in a wide class of the relevant nearby possible worlds it doesn’t occur’. This is a general claim about what constitutes an accident. If we adopt this general view to our case, we arrive at the following characterization: it is an accident that the agent's intention to do A causes what it represents if, and only if, in most nearby possible worlds where the agent intends to do A, the intention does not cause her to do A and, thus, she is not doing A. So, if we want to exclude this sort of accident, we will have to amend the CTA with the following modal condition: S is doing A intentionally only if, in most nearby possible worlds where S intends to do A, S is doing A.Footnote ¹⁶

Just like with the previous proposal, I think that, as it stands, the condition is too strong, and that an improvement on the condition is likely to render it unavailable to a reductive account of intentional action. To begin, consider what is meant by ‘nearby’ in ‘nearby possible worlds’. The qualification is intended to restrict the set of possible worlds relevant to the assessment of whether what is happening in the actual world is an accident. Thus, importantly, to determine whether what's going on in the actual world is an accident, we don’t need to look to just any possible world. We only need to look to possible worlds that are sufficiently similar to the actual world (or, as it is sometimes put, to possible worlds that could have easily been the case). However, I think the task of specifying the right degree and respect of similarity between worlds gives rise to difficulties very much like those we encountered in specifying the content of the ceteris paribus clause. In other words, it strikes me as quite unlikely that we can specify how similar a possible world must be to qualify as ‘nearby’ in the relevant sense, without making reference to the notion of intentional action.

To bring this out, consider the following case. John intends to kill a tyrant by poisoning the supply of drinking water to the tyrant's palace. However, unbeknownst to John, the tyrant is well aware of his plans and orders Jones to prevent John from realizing his intention. Now, suppose further that, due to some quite unlikely event, Jones himself is prevented from interfering with John's plans – imagine, for example, that he is run over by a car on his way to where John intends to poison the water supply system. In that case, John will realize his intention without any interference by Jones. But there are many nearby possible worlds in which John is not doing what he intends to do, i.e. worlds in which Jones is not run over by a car and prevents John from poisoning the water supply. Thus, the relevant modal condition is false: in many nearby possible worlds where John intends to poison the water supply, he is not doing so. However, this doesn’t seem to imply that, in the actual case, John is not acting intentionally when he is poisoning the water supply. Hence, as it stands, the relevant modal condition doesn’t seem necessary for intentional action and, thus, it is inadequate to account for the non-accidentality requirement.Footnote ¹⁷

In response, one might want to argue that, although worlds in which Jones is not run over by a car and prevents John from poisoning the water supply are indeed possible and nearby in some sense, they are not sufficiently similar in the relevant sense. To determine whether, in the actual case, John is acting intentionally, we should consider only those worlds in which Jones is run over by a car. If we do, the relevant modal condition may turn out to be true after all. Perhaps, this is the right way to preserve what seems intuitively right about the modal condition. But if that's the way to save the condition from counterexamples, it is hard to see how the relevant respect of similarity can be specified without reference to the notion of intentional action. It certainly seems that, when we exclude worlds in which Jones is not run over by a car from the set of worlds that qualify as relevantly similar, we rely on a previous decision to treat what John actually did as an intentional action. Thus, in ascertaining whether the similarity to a nearby possibility is relevant or great enough, we do seem to be guided by our grasp of what it is to do something intentionally.Footnote ¹⁸ Moreover, examples such as the one we just considered can be easily multiplied. But if that's so, then things start to look quite the opposite from what the modal account suggests: it is not that we determine whether someone is acting intentionally by considering if the relevant modal facts obtain, rather we determine whether the relevant modal facts obtain by considering if someone is acting intentionally. Once again, I don’t want to deny that, when someone is acting intentionally, a modal condition of the relevant sort is true. It's just that I don’t think the condition is intelligible independently from the idea of intentional agency itself, which it would have to be if it is to be employed in a reductive account of this idea.

So, to sum up, if these considerations are correct, the attempt to properly understand both versions of the modal condition returns us to the notion of intentional action. This is to say, even though it is true that, in acting intentionally, an agent is modally responsive to her intention, it is not because she is modally responsive that she is acting intentionally. Rather, the connection between intentional action and modal responsiveness runs in the other direction: it is because an agent is acting intentionally that the relevant modal condition is true. Now, what matters for our purposes is that this suggests that there should be a non-modal account of the non-accidentality requirement. For, if a proper understanding of the modal condition is dependent on the notion of intentional action, then by the same token it is dependent on a notion of non-accidentality, the one that is implied by the notion of intentional action. Hence, if we don’t want to rest content with a brute notion of non-accidentality, we should be looking for a non-modal account of the non-accidental connection between an intention's causality and its content. Recall that earlier I distinguished between three different ways of accounting for this connection: in modal, functional, or constitutive terms. In the next section I will briefly discuss and dismiss the functional account. If this is right, the need for a non-modal account suggests that we should conceive of the relevant non-accidental connection in constitutive terms – something I will turn to in Section 6.