David Hume famously chides his contemporaries for sliding from claims about what is the case to claims about what ought to be the case. According to Hume, this slide could only be legitimate if the latter could be deduced from the former, but he says this deduction ‘seems altogether inconceivable’, for the subject matter of the former is ‘entirely different’ from that of the latter (Hume [1739–40]Reference Hume, Selby-Bigge and Nidditch1978: 469). This difference between the descriptive and the normative is marked in many corners of contemporary philosophy, including the distinction often made between explanatory reasons and normative reasons. If we generalize Hume's thought and think of all descriptive matters as ‘entirely different’ from normative matters, we may think that endorsing a given account of explanatory reasons puts no restrictions on the account of normative reasons one may simultaneously endorse. We may think, for example, that one can unproblematically hold a view that accepts some Humean account of explanatory reasons, which explains actions in terms of belief-desire pairs, but rejects Humean accounts of normative reasons, which characterize the goodness of an action in terms of the satisfaction of the acting agent's desires.Footnote 1
This two-part, half-Humean view is, I believe, widely held by contemporary philosophers. The view results from thinking that Hume's belief-desire account of action is basically correct but that the subjectivism pervading his moral account is unacceptable.Footnote 2 Regarding the former: one can hold a Humean account of action-explanation even if one agrees with J. David Velleman that the ‘standard story of action’, itself Humean, is incorrect (2009b: 123–24).Footnote 3 According to the standard story, all that one needs to mention in order to explain an action is a desire that specifies a goal and an instrumental belief that specifies a way of pursuing that goal. Velleman complains that this story ‘fails to cast the agent in his proper role’ (2009b: 123); to cast the agent in this role, Velleman thinks, a sub-agential motive for self-knowledge needs to be added to the story (2009a: 22–24). His explanatory strategy follows a model popularized by Harry Frankfurt, who claims that a theory of action must be capable of distinguishing akratic from non-akratic action, which the standard story cannot do. To make this distinction, Frankfurt thinks, a notion such as identification (1987a) or wholeheartedness (1987b) or volitional necessity (1999) must be added to the explanans.Footnote 4 Velleman and Frankfurt thus reject the standard Humean story, but they accept the underlying Humean commitment to the necessity of belief-desire attributions in action-explanations. Although they think that the basic belief-desire model can only produce a partial explanation of what Velleman calls ‘full-blooded action’ (2009b: 124), each holds that any explanation of an action requires mention of both a belief and a desire. For the purposes of this paper, it is commitment to this last claim that makes one's account of action-explanation Humean.
The subjectivism of the Humean normative account is perhaps best demonstrated by an example. If Raskolnikov wants to murder Alyona without getting caught, then, according to the account, he ought to take the necessary means to avoid getting caught, e.g., silencing anyone who happens to witness the murder. The Humean account denies that there is anything wrong with Raskolnikov's desire to murder Alyona, considered on its own. To ask whether Raskolnikov should have this desire, considered on its own, is to make a category mistake—it is not the sort of thing that, on its own, is apt for normative assessment. To many, this simply seems wrong. For example, the vast majority of those who write on the instrumental principle in practical rationality reject that one ought to take whatever means are necessary to satisfy a desire; sometimes, as in Raskolnikov's case, it is the initial desire that ought to go.Footnote 5 If one agrees with this, one may be inclined to endorse an account of practical normativity that is anti-Humean.
Again, if we accept the generalized version of Hume's thought about the descriptive/normative distinction, we may find no reason to worry about simultaneously endorsing a Humean account of action-explanation and an anti-Humean account of practical normativity. After all, they concern ‘entirely different’ matters, so it does not seem that there is any problem with combining them. The goal of this paper is to show that there is such a problem. I will argue that no viable anti-Humean account of practical normativity can be coherently combined with a Humean account of action-explanation.Footnote 6 If my argument is successful, it will demonstrate a way in which accounts of action-explanation can restrict accounts of practical normativity and vice versa. If I am right, the proponent of the Humean account of action-explanation must also accept a Humean account of practical normativity: if one is not a thoroughgoing Humean, one cannot be Humean at all.Footnote 7
My target here will be the view Michael Smith presents in The Moral Problem (1994) and several of the essays he has written in its wake (1997, 1998, 2003, 2010). Smith gives clear expression to the two-part, half-Humean view described above. Smith argues for the Humean explanatory account by pinpointing and defending one of its fundamental commitments, which I call the separability condition. This condition states that beliefs, on their own, cannot motivate action; for any belief that appears to be a counterexample, there exists a separable desire that, when separated from the belief, reveals the belief to be motivationally inert. The condition is my topic in section 1. The next section turns to Smith's anti-Humean normative account. My interests here are what, in general, it takes for a normative account to be anti-Humean and the specific features of Smith's view that satisfy these requirements. Smith explains practical goodness in terms of the advice that an agent's rationally ideal counterpart would give about that agent's desires. Now this might seem to be far too specific a view to warrant detailed discussion in an essay seeking to establish broad claims about Humeanism.Footnote 8 On the contrary, I believe and aim to demonstrate that a careful analysis of Smith's view reveals the central importance of representations of ideal rationality in accounts of action, Humean or otherwise. The key feature of Smith's view is the convergence thesis, which asserts that the hypothetical desires of fully rational agents, because they are fully rational, converge. As Smith himself notes and I explain at the end of section 2, anyone wishing to spell out an anti-Humean account of practical normativity in terms of a multiplicity of ideal agents must endorse this thesis.
One can be an anti-Humean about practical normativity, however, without articulating one's account in terms of a multiplicity of ideal agents. In section 3, I show how one can instead advance an anti-Humean account in terms of a generic conception of full rational agency. One can do this, I argue, by specifying the account in terms of the fully rational agent, which does not pick out any particular agent. In section 4 I argue that if an account is specified in this way, it cannot satisfy the separability condition, and thus it is incompatible with a Humean account of action-explanation. This by itself is a minimal result, but it lays the groundwork for section 5, where I argue for the general incompatibility of Humean accounts of action-explanation and anti-Humean accounts of practical normativity. There, I argue that Smith's convergence thesis is a mere means for specifying the justificatory foundation of his anti-Humean normative view. This foundation is determined by what the account takes to be the final grounds of acceptable justifications. A conception of these grounds, I argue, is a generic conception of full rational agency, so any account based on these grounds cannot satisfy the separability condition. Because the object of this generic conception is a proper explanandum of an account of action-explanation, accounts of practical normativity whose justificatory grounds include this conception—i.e., anti-Humean accounts, including Smith’s—cannot be coherently combined with a Humean account of action-explanation. If my arguments are successful, they demonstrate a way in which one's views in action theory can both limit and be limited by the ethical views one endorses; their subject matters are not, pace Hume's characterization, ‘entirely different’.
1. Separability and the Omnipresent Possibility of Akrasia
Let us open by investigating the place of the separability condition in Humean accounts of action-explanation. According to the Humean, we can never fully explain an agent's action simply by saying that she did what she believed was good to do. If such a description were explanatorily adequate, then going on to say that she desired to do what she did would be explanatorily superfluous. On the Humean view, citing a desire is never explanatorily superfluous; in cases where it appears that such an explanation will suffice, the Humean insists that a full explanation requires saying not only that the agent believed the action to be good but that, beyond this, the agent also in fact desired whatever she pursued in performing the action. In order for this to be true, it must always be possible to separate the desire that explains an action from any evaluative belief about the goodness—be it about the desirability, or the reasonability, or anything else—of that action. Put somewhat formally, we may express this condition as follows:
The separability condition: for any desire D belonging to any agent A, D is separable from any evaluative belief that A has about the desirability, reasonability, or other sort of goodness of the action or goal specified by the content of D.
To reject this is to claim that one can explain action without citing a psychological state that the Humean takes to be a necessary component of every action. To reject this, then, is to reject the conception of the mind that is fundamental to the Humean approach to moral psychology.Footnote 9
Following Smith, I think we should understand the separability at issue here metaphysically. There are at least two ways of formulating this metaphysical claim about separability: one as the rejection of a necessity, the other as the rejection of a possibility. The Humean rejects that there is any belief-desire pair to which the desire belongs as a necessary component. Consider again the case in which it seems like we can explain what a person does simply by saying she believes it good to do; the Humean insists that even if it seems such an explanation will suffice, the desire that motivates the action does not belong to the cited belief as a necessary component. If the Humean is wrong about this, then what he describes as a belief-desire pair is not, in fact, a pair; rather, it is a single psychological state to which the ‘desire’ belongs as a necessary, inseparable part. Such a psychological state is possible just in case it is possible that there is an agent and an evaluative belief that p such that, in every possible world in which she believes that p, she holds the desire D that corresponds with that belief. If such an agent and belief are possible, then D is not a psychological state separable from the belief that p; rather, D is a necessary part of the belief-desire ‘pair’, which, again, is properly conceived of as a single psychological state. Because the Humean rejects the necessity of any desire belonging to any belief, he also rejects the possibility of the agent just described.
Smith claims that the Humean ought not merely to assume these metaphysical views but should argue for them, and his own argument, he thinks, is a ‘knockdown’ against the view that there exist evaluative beliefs that are necessarily motivating (Reference Smith1994: 123).Footnote 10 Smith argues that whenever it seems we can explain an action simply by saying the agent found it worth doing, we can conceive of the agent evaluating the action as worth doing but lacking the desire to act in accordance with this evaluation. All that is needed to establish this, according to Smith, is to note that it is ‘merely a contingent fact that the agent in question does not suffer from weakness of will, for no mere mortal is necessarily virtuous . . . [which] entails that his beliefs about the desirability of his actions and his desires to act accordingly are distinct existences’ (Reference Smith, Bransen and Cuypers1998: 29–30, emphasis in original). If this is correct, then it is impossible that there be some agent who has a single cognitive state that includes a desiderative component as a necessary part. If such an agent is impossible, then the separability condition is satisfied. Because the argument turns on the claim that akratic failure is possible in every case that appears to threaten the separability condition, let us call it the omnipresent possibility of akrasia argument.Footnote 11
2. The Convergence Thesis
Although he endorses the Humean account of explaining individual actions, Smith denies the Humean claim that it is always rational to act so as maximally to satisfy one's desires, regardless of their content. On the Humean view of practical normativity, as Smith says, ‘[w]hat it is rational for an agent to do is . . . relative to what she most wants to do’ (Reference Smith1994: 130). Smith rejects this relativism, asserting instead that the standards of practical rationality are determined in a way that is not contingent upon a given agent's set of desires. According to Smith, what makes an action rational is not what an agent happens actually to desire; instead, ‘what it is desirable for us to do is what we would desire that we do if we were fully rational’ (Reference Smith1994: 151). In another formulation, this time couched in terms of goodness instead of desirability, Smith makes the point this way: ‘When a subject judges p's being the case in certain circumstances C to be good, what she is doing is making a judgment about what she would desire to do in the case of C if she had a psychology that eludes all forms of rational criticism’ (Reference Smith2003: 591). On Smith's view, an agent eludes all forms of rational criticism and so is ideally practically rational if she is maximally informed, if her behavior is not affected by physical or emotional disturbance, and if her set of desires is maximally unified and coherent. The desires of an agent with these characteristics define what is rationally desirable—and thus good—to do.Footnote 12
Smith has a great deal more than this to say about ideal practical rationality. The specific elements comprised by his conception of the ideal need not concern us, as they are not presently relevant. What is relevant, however, is his two-step articulation of the universality of this ideal. The first step of this articulation is to characterize ideal practical rationality in terms of a multiplicity of fully rational agents. Note the second formulation of his view presented above: there, Smith asserts that what is desirable for a given agent is what her fully rational self would desire, not what the fully rational agent would desire. Part of Smith's thought here is this: whereas what is desirable for me to do is determined by what I would desire myself to do were I fully rational, what is desirable for you to do is determined by what you would desire yourself to do were you fully rational. My fully rational counterpart is not the same as your fully rational counterpart, for my counterpart may rationally desire that I do certain things that your counterpart does not desire that you do. Nevertheless—and this is the second step, which secures the universality of the view—the desires of all fully rational agents, in Smith's words, ‘converge’. Smith tells us that ‘the convergence required is not at the level of desires about how each such agent is to organize her own life in her own world. . . . The convergence required is rather at the level of their hypothetical desires about what is to be done in the various circumstances in which they might find themselves’ (1994: 173). Although there are many fully rational agents, any would desire what any other does desire were the former in the latter's position. Call this last claim the convergence thesis.
Smith endorses the convergence thesis to distinguish his view of full practical rationality from its Humean alternative. As Smith notes in a discussion of Bernard Williams's view, an account of practical normativity can be Humean even if it explains the rationality of reasons in terms of the desires an agent would have were she more rational. The account will be Humean if it asserts (a) that the desires an agent would have were she fully rational are determined exclusively by a function on her actual desires and (b) that it thus follows that there may be fully rational agents whose desires do not converge in the manner just described. On such an account—which, as Smith sees things, is the best understanding of Williams's view—standards of practical rationality are, in the end, relative to subjective sets of desires. Smith rejects that the standards of practical rationality are so relative. Because he spells out his account of ideal practical rationality in terms of a multiplicity of fully rational agents, he articulates this rejection by endorsing the convergence thesis.Footnote 13
3. Representations of Ideal Practical Rationality
I turn now to showing why Smith's is not the only way of articulating an anti-Humean account of practical normativity. Here, I am interested in demonstrating how an anti-Humean account can be given in terms of a generic representation of ideal agency. As a first step toward this, note that there is nothing about the concept of the agent as it features in the anti-Humean view of ideal practical rationality that requires the account to be spelled out in terms of a multiplicity of individual agents. If an account happens to be spelled out in terms of a multiplicity of fully rational agents, the convergence thesis guarantees that all such agents agree on what is and is not desirable in a given context. Thus, in a given context, were a person to turn to a fully rational agent for advice—which is how Smith invites us to think of interacting with these agents (Reference Smith1994: 151)—it would not matter to which one she turned: each will give the same advice. This idea can just as easily be put, however, by speaking simply of the fully rational agent. If we reformulate the position in terms of the fully rational agent, nothing like the convergence thesis is needed to secure the account's anti-Humean aspect. There is only one fully rational agent giving out advice, and its views about what is desirable to do in a context trivially converge with themselves. This conception is anti-Humean, then, not because of the converging desires of many fully rational agents, but rather because the fully rational agent's desires bear uniformly on all rational agents.
Representations of ideal rationality articulated in terms of the fully rational agent can be nonindividual in a way that representations of a multiplicity of such agents cannot. The decision to articulate one's anti-Humean view in terms of a multiplicity of agents not only makes the convergence thesis a necessary part of one's account; it also commits one to representing fully rational agents as distinct individuals. While my fully rational counterpart and your fully rational counterpart always agree on what is desirable, mine is nevertheless different from yours, for there are facts true of mine—its desires, its situation—that are not true of yours, and vice versa. For these differences to be conceivable, these fully rational agents must be conceived of as distinct individuals. The differences here are irrelevant to the role played by these agents in the account of practical normativity; because their views on what is desirable always converge, these agents are functionally equivalent. This functional equivalence explains why, if you could actually speak with these agents, each would give you the same advice about what to do in a given situation, no matter which one you asked. In spite of this functional equivalence, the anti-Humean who spells out her account of ideal practical rationality in terms of multiple full rational agents must conceive of these agents as individuatable in some way even though the differences that individuate them are superfluous and irrelevant to their role in the account.
If, instead, the anti-Humean spells out her account in terms of the fully rational agent, she need not articulate and then ignore these superfluous differences, because her account need not discuss any particular agent or agents. If she presents her account in terms of the fully rational agent, she may deploy a generic conception of ideal practical rationality, which is not a representation of a particular agent. To understand the way in which these generic representations are nonparticular, consider Michael Thompson's recent discussion of the role these representations play in biological thinking (Reference Thompson2008: chapter 4). Thompson invites us to imagine the narrator of a nature documentary saying, ‘The bobcat has 2–4 cubs in the spring’. Suppose I hear and understand this claim. When I do, despite what the surface grammar might suggest, I do not thereby represent a particular bobcat. I do not represent a being that has its own history, is individuatable from others of its kind, and may be named to distinguish it from its fellows. This should be obvious: no particular bobcat can bear an indeterminate number of offspring. I am not confused about this, so when I understand the narrator's assertion, I do not have in mind some particular being capable of begetting an indeterminate number of offspring.Footnote 14
When I understand the narrator, I also do not represent some quantity of bobcats; his uttering ‘the bobcat’ does not cause me to represent all possible bobcats or all actual bobcats or all living bobcats or even most living bobcats. I would not represent any of these even if he had said, ‘Bobcats have 2–4 cubs in the spring’. Although the surface grammar of this claim differs from the one made in terms of ‘the bobcat’, both deploy generic representations to make assertions of what Sarah Jane Leslie calls characteristic dimensions of bobcats (Reference Leslie2008: 29–39). In Thompson's idiom, to make claims about the characteristic dimensions of biological kinds is to make natural-historical judgments (Reference Thompson2004: 47–57; Reference Thompson2008: 64–76). These judgments do not need to be true of the majority of that species’ members in order to be true of the species itself. Both Thompson and Leslie demonstrate this point via examples from Insecta. As Thompson notes, it may be true that the mayfly dies shortly after reproduction (or, equivalently, that mayflies die shortly after reproduction) even if, as a statistical matter, most mayflies never make it to reproductive age (Reference Thompson2008: 68). Leslie's example concerns mosquitoes: it may be true that mosquitoes carry the West Nile virus (or, equivalently, that the mosquito carries the West Nile virus) even if the majority of them do not carry the virus (Reference Leslie2008: 1; Reference Leslie2013: 14).Footnote 15 When one conceives of species via generic representations in order to make natural-historical judgments, then one does not represent a distinct species member, nor does one represent some quantity of members of the species.
Instead, when one makes a natural-historical judgment, one either conceives of a species as such or conceives of what it is to belong to a given species via a nonparticular representation of a species member. Regarding the former: suppose I judge, ‘Dinosaurs are extinct’. The predicate here applies only to the species as a whole, not to its members considered as individuals. It would be a category mistake to speak of an individual going extinct; only species as such can become extinct.Footnote 16 Manfred Krifka calls the subject of these kind-predications D-generics (Reference Krifka1987: 11–18).Footnote 17 Contrast this with the judgment, ‘The bobcat has 2–4 cubs in the spring’. Here, ‘the bobcat’ does not represent the species as such, for it is a species member, not the species itself, that begets cubs. Krifka calls the subject of this sort of predication an I-generic (Reference Krifka1987: 4–10).Footnote 18 Although ‘the bobcat’ does not here represent the species as such, neither does it represent any particular member of the species; again, no particular member can beget an indeterminate number of offspring. When I make this judgment, I represent what is characteristic of bobcats, what makes a bobcat a bobcat, without representing any particular bobcat or bobcats. There are plenty of questions that can be raised about thoughts involving I-generic representations; all I want to note presently is that we can think with this sort of representation, and when we do, we do not represent particular members of kinds.Footnote 19
This sort of representation can be employed not only to think about brutes such as bobcats, mayflies, and mosquitoes, but it also may be used to represent practical rationality, at least as the anti-Humean conceives of it. We have already noted that the anti-Humean does not need to speak of a multiplicity of fully rational agents to articulate her account of ideal practical rationality; she may, as we have noted, present her view by speaking instead of the fully rational agent. If she articulates her account in terms of the fully rational agent, making judgments such as, ‘The fully rational agent's desires define what is rational for anyone to desire’ or ‘The fully rational agent would advise me not to have another drink’, she need not thereby speak about any particular agent. If she proceeds in this way as she develops her position, her account will characterize what makes an agent fully rational without using a conception of a particular agent, a being with a history, individuatable from other agents, capable of bearing a name that distinguishes it from its fellows. Features of particular agents such as these do not belong to this generic representation of the fully rational agent any more than such features belong to the generic representations that are used in natural-historical judgments about brutes. Whereas representing full rational agency in terms of a multiplicity of agents requires conceiving of agents as having these features and then ignoring them when characterizing rational desirability, representing such agency via a generic conception of the fully rational agent neither requires nor has room for such superfluities.Footnote 20
4. Separability and Representations of Ideal Practical Rationality
Now, it may not be clear at this point why the possibility of a generic representation of the anti-Humean view of ideal practical rationality is of any concern to someone who endorses a Humean account of action-explanation. To see why this matters, let us review the separability condition—which, again, is a condition any Humean view of action-explanation must accept—and the argument for it. The condition asserts that there is no evaluative belief such that, if one has the belief, one necessarily has the desire that corresponds to the belief. The argument for this condition is the omnipresent possibility of akrasia argument, which asserts that in any case where an agent appears to perform an action directly on the basis of an evaluative belief, we can conceive of the agent not performing the action due to akratically lacking the corresponding desire. This argument will fail if there is a case in which a desire cannot be separated from its corresponding evaluative belief. If a view must admit such a case, it cannot satisfy the separability condition, and so it is a view with which the Humean account of action-explanation is not compatible.
This puts constraints on the conceptions of agency that are compatible with a Humean account of action-explanation. If the separability condition is to be satisfied, it must be true that whenever one thinks of a case that appears to thwart the condition, the conception of the agent in the case is of an individual that can be reidentified across possible worlds, one of which includes the agent with the relevant evaluative belief but without the corresponding desire. I will now argue that the generic conception of the fully rational agent sketched in the preceding section does not allow for the needed cross-world reidentification, and thus a view articulated in terms of this conception cannot satisfy the separability condition. To show why this is the case, it will be useful first to consider some of the ways in which generic representations can feature as the subjects of counterfactuals. Bringing these into view will help us see why these representations cannot feature as the subjects of the counterfactuals that articulate the cross-world identification needed for the omnipresent possibility of akrasia argument.
Let us divide generic counterfactuals into two sorts: those in which at least one of the characteristic dimensions of a kind is altered and those in which all are kept fixed. Examples of the first kind are expressed by judgments such as ‘Had the bobcat evolved a longer tail, its gait would differ’ and ‘If the fully rational agent did not have to nourish itself to survive, its desires would differ’. Focus specifically on the latter case. Should I make this judgment and should I go on to think about how these desires would differ, I would thereby consider how desirability would shift for all agents if they did not need to nourish themselves to survive. In doing this, I might be said to reidentify the fully rational agent across possible worlds. There are two things to note about this reidentification. First, in this sort of case it is the kind as a whole, not some member or members of the kind, that is reidentified. Second, this sort of reidentification amounts to a reconceptualization of what is normal: to think about the scenario is to conceive of a norm of practical rationality—specifically, that one ought to nourish oneself—as altering. With the second sort of generic counterfactual, there is no such alteration; all characteristic dimensions are held fixed, and one wonders what a generic representative of the kind would do or be like under specified circumstances. To demonstrate this sort of counterfactual, imagine thinking about what the bobcat would do were it to catch fire or what the fully rational agent would do were it to catch fire.Footnote 21 These are not thoughts about how members of a kind would differ were the kind's characteristic dimensions different; rather, they are thoughts about what a generic member of a kind would do or be like in a given circumstance, assuming that the kind (and therefore its generic member) retains its defining characteristics.
Neither sort of generic counterfactual allows for the cross-world reidentification that is necessary for satisfying the separability condition. With the first sort of generic counterfactual, no individual is reidentified across worlds. Again, it is the entire kind that is reidentified; again, to reidentify an entire kind is to conceive of the characteristic dimensions of the kind as altering. In the case of reidentifying the kind to which the fully rational agent belongs, this is to conceive of the norms of full rational agency as altering. These norms have to be held fixed, however, to formulate the counterfactuals involved in the omnipresent possibility of akrasia argument. These counterfactuals involve thinking of an agent who is fully rational in a given world as being less than fully rational in some different world. The norms of rationality are held fixed; only the psychological makeup of the agent alters across the worlds. The first sort of generic counterfactual does not hold norms fixed in the way needed, so it cannot be used to formulate the counterfactuals that are needed to establish separability.
The second sort of generic counterfactual is likewise incapable of articulating the needed reidentification. This sort of counterfactual holds the characteristic dimensions of the relevant kind fixed and describes what a generic member would do or be like in a specified set of hypothetical circumstances. These circumstances cannot include an alteration of the characteristic dimensions themselves; if these are altered, then the kind that is defined in terms of these dimensions is also altered. To imagine what a generic member of the kind that is characterized by full rational agency would do if that generic member were not fully rational just is to alter the characteristic dimensions of the generic member and then imagine how the resulting being would behave. To conceive of this is not to hold the norms of rationality fixed and then to alter the psychological makeup of an individual, as is needed for the omnipresent possibility of akrasia argument. As with the first sort of generic counterfactual, the psychological makeup of the being cannot be altered without thereby altering the norms that need to be held fixed. Thus, the second sort of generic counterfactual is also incapable of formulating the counterfactuals that are needed to establish separability.
If these two sorts of counterfactuals exhaust the ways in which one can think counterfactually about the objects of generic representations, then no account of full rationality that is articulated generically in terms of the fully rational agent can satisfy the separability condition. The sorts seem to be exhaustive. A generic counterfactual either concerns a kind as such, or it concerns a generic member of a kind. The two sorts of counterfactuals we have considered track this distinction. The former sort involves the reconception of the characteristic dimensions of a kind. These have to be held fixed for the omnipresent possibility of akrasia argument to succeed, so this sort of counterfactual cannot be used to argue for the separability condition. One can try to hold the relevant characteristic dimensions—i.e., the norms of practical rationality—fixed by using the second sort of generic counterfactual. This, however, also fails, for altering the relevant psychological facts of a fully rational agent across worlds just is to alter the norms that need to be held fixed; the agent's psychology defines the relevant norms. In neither case, then, can the object of the generic representation be reidentified as the separability condition requires. In no case, then, is an account of agency formulated generically in terms of the fully rational agent compatible with a Humean account of action-explanation.
5. Separability, Ideal Agents, and Justificatory Foundations
I have just arrived at a conclusion about accounts of agency that are formulated in terms of a generic conception of the fully rational agent. My aim, however, is broader; I intend to show that any anti-Humean account of practical normativity, whether or not it is explicitly formulated in terms of a generic conception of full rational agency, is incompatible with a Humean account of action-explanation. To this end, it will be useful here to show how the arguments of the preceding section bear on Smith's specific half-Humean view, for one might naturally wonder how these arguments pertain to a view that, like Smith’s, is not explicitly articulated in terms of a generic conception of full rational agency. The answer to this concern requires distinguishing between the particular means a theorist uses to articulate an account of practical normativity and what I will call the justificatory foundation of this account. We can explicate this notion by reflecting on the claims that, on a given view of practical normativity, count as final justifications. Consider, for example, the Humean view. It counts the maximal satisfaction of the most coherent set of desires derivable from a given individual's desires as the final justification for practical matters. If one cites this satisfaction to justify an action, nothing further need or can be said in favor of that action. A description of this satisfaction, then, is the proper end of any Humean evaluation or justification, which shows it in turn to be the justificatory foundation of the account.
Now consider Smith's version of the anti-Humean view. The advice of his ideal advisors is a mere means for specifying this view's justificatory foundation, for this advice is not, according to the view, the final grounds for any particular justification. To see this, imagine that Peggy is deliberating about what to do. Imagine Peggy thinks about her decision as the view recommends, i.e., by considering the advice her fully rational counterpart would give her. Suppose Peggy makes her choice on the basis of this imagined advice. After she performs the relevant action, her friend Stan might ask her, ‘Why did you choose to do that?’, and she may respond, ‘It is what my fully rational counterpart would have advised’. Stan may intelligibly respond, ‘Yeah, but why would your fully rational counterpart advise that?’ Continuing along Smith's lines, Peggy might say, ‘Because it is what any fully rational agent would desire in the situation’. The fact that the view recommends this answer shows that its grounds for justification are not the advice one's specific ideal counterpart might give, for this advice is itself justified by the fact that it states what any fully rational agent would desire. The justificatory foundation is not revealed until the chain of justification ends, and this chain does not end with what the individual advisors advise—the advice itself rests on further justificatory grounds.Footnote 22
When we articulate the chain of justification that underwrites Smith's view to its conclusion, we arrive at a generic, nonparticular conception of full rational agency. Note that Peggy's last answer abstracts away from any of the rationally irrelevant particularities of her fully rational counterpart; included among these particularities are the features of the counterpart that make the counterpart hers as opposed to Stan's (or anyone else’s). Those features are irrelevant to the justificatory question being asked. Once one has abstracted away from these features, the remaining conception of full rational agency is generic; its use does not require conceiving of any particular fully rational agent. It is this conception of full rational agency, with all irrelevant contingencies stripped away, that does the justificatory work in an anti-Humean account. Smith articulates this conception via the convergence thesis, but he need not do so—the grounds for Peggy's justification are what any fully rational agent would desire, and this can be articulated either by means of Peggy's fully rational counterpart or by skipping right to the justificatory end. Either way, what makes the view anti-Humean is its nonparticular, generic justificatory base. If we represent this base via the conception of an agent, the representation will be of the fully rational agent, generically conceived. For a normative view to be anti-Humean, then, it must be representable by a generic conception of full rational agency, a conception that, as we saw in the preceding section, is incompatible with a Humean view of action-explanation. Half-Humeanism would appear to be in trouble.
One might worry here that this generic, nonparticular conception of ideal agency is so abstract that it cannot offer determinate advice about what a person ought to do in a specific situation. Suppose Peggy is a farmer, and what Stan wants to know is why she has just destroyed a number of her tomato plants. The plants were infected by blight, she explains, and she says any good farmer would do the same under the circumstances. Any further appeal to ideal rationality as such appears uninformative and superfluous at this point; determinate advice comes from reflection on the ideal farmer, not from the ideal rational agent, considered as such. If this is right, then maybe there are facts about Peggy—e.g., that she is a farmer—that have to be included in a conception of ideal agency if the conception is to recommend substantive advice to her in specific circumstances. This in turn might seem to speak in favor of a view such as Smith’s: because this view is articulated in terms of ideal rational counterparts, it is equipped to supply these facts. This worry makes an important point, I think, but not one that counts against the present argument. Let us grant that in order for her to determine what to do with her blighted plants, Peggy needs to reflect on what a good farmer in her situation would do; let us likewise grant that reflection on ideal agency as such is too abstract to be of any use here. Her reasoning will nevertheless be grounded in a generic, nonparticular conception; the final justification of her action lies in what any good farmer would do, which, as per the argument above, is just what the ideal farmer would do. Justification thus grounds out with a generic conception, but it is a conception of the generic farmer, not of the generic fully rational agent. This point thus does not decide the argument in Smith's favor.
A defender of the half-Humean position might accept all this and yet still wonder how it counts as a challenge. Accounts of action-explanation aim to explain why humans act; the generic conception of the fully rational agent is not a conception of a human. Why should any account of action-explanation, Humean or otherwise, apply to a generic being who has no material existence and thus does not act? We will answer this question by considering what is necessary for an appeal to a fully rational agent to play its justificatory role. Return to Peggy's assertion in the original example that her choice is based on what any fully rational agent would desire. Note that what is desired is not a state (or at any rate, not only a state), but (also) the performance of an action. This must be so, for accounts of practical normativity do not merely concern the consequences of action; they concern actions themselves.Footnote 23 Now Stan might still ask Peggy, ‘But why would the fully rational agent desire that action?’ If Peggy is to explicate the rational desirability of the action, she must ultimately say, ‘Because it is what any fully rational agent would do in the circumstances’. What Peggy is ultimately justifying is the rationality of an action, and one cannot conceive of this without conceiving of—this may seem redundant, but it is just the point—an agent acting. For the fully rational agent to play its justificatory role, then, it must ultimately be conceivable as an agent acting.
Because the fully rational agent must be conceivable as an agent acting, it and its activities are proper explananda of an explanatory account of action. If an explanatory account of action does not apply to what the fully rational agent does, then what this agent does, according to the account, is not action. If it is not action, then any normative account whose justifications are grounded in conceptions of what this agent does is not a normative account of action; the subject matter of such an account can only be nonaction doings of the sort performed by this agent. The Humean account of action-explanation cannot characterize what the anti-Humean fully rational agent does as action. It cannot, because the agent that is the object of this conception cannot be conceived of as suffering from akrasia; its desires are not separable from their related evaluative beliefs. The anti-Humean account of practical normativity thus cannot be a normative account of what the Humean believes is action. This makes it impossible, then, to combine a Humean account of action-explanation coherently with an account of practical normativity whose justificatory grounds include an anti-Humean conception of full rational agency.
I can think of one last response the half-Humean might advance to try to save her view. What these arguments reveal, the response goes, is the trouble one gets into when one formulates an account of practical normativity in terms of ideal agents. These imaginary agents are dispensable: one can—and this response adds, should—give an anti-Humean account of rational desirability without mentioning any such imaginary agent or agents. My reply to this is to reiterate the point made two paragraphs above. An account of practical normativity is an account of what it is good to do. The object of evaluation is action, and to think about action, one must think about an agent acting. Even if an anti-Humean account is given without mentioning an ideal agent, a conception of ideal agency must be presupposed to make the account's claims intelligible. I have argued that the foundational justificatory conception for any anti-Humean normative account is not coherently combinable with a Humean account of action-explanation. If this is correct, then whether or not an anti-Humean explicates her account in terms of agents, ideal or otherwise, is irrelevant; it is the implicit justification-grounding conception of agency that matters.
6. Conclusion
If all of this is correct, then one cannot coherently maintain a half-Humean view of action; one must either hold a thoroughgoingly Humean view or one must reject both its explanatory and its normative components. I incline toward the latter option, in no small part because I am persuaded by Smith's own arguments against the Humean account of practical normativity (Reference Smith1994: 164–74). Defending these arguments, however, is a task for another occasion.
