Instrumental Reasons1 “What reason have I to do that?” someone asks. The answer (if there is one) is often: “Doing that is a means to something else that you have reason to do.”2 When is this answer correct? What does the existence, or weight, of such an “instrumental reason” depend on? The focus of these questions—instrumental reasons—should be distinguished from what might be called “instrumental rationality.” Instrumental rationality is a matter of the coherence, or unification, of the deliberating agent’s state of mind. Insofar as one wills an end and believes that something is a necessary means to it, one is, in refusing to will those means, at odds with oneself.3 By contrast, instrumental reasons concern not the coherence of one’s deliberation, viewed as a psychological episode, but instead the structure of the subject matter about which one deliberates. If there is reason for one to pursue an end4 (whether or not one pursues it), and if an action is in fact (or relative to the better information of an advisor or onlooker5) a means to that end (whether or not one knows it), then that is itself a reason for one to adopt the means— quite apart from any considerations about one’s psychological coherence.6 This is reflected in familiar patterns of advice. If I know that there is reason for you to lower your blood pressure, and if I know that your taking this medicine will do that, then I am liable to tell you, if I can, that that is a reason for you to take this medicine—whether or not you intend to lower your blood pressure, or know that this medicine will do that. Furthermore, at least on the assumption that what one ought to do depends on the reason there is for one to do it, a sort of transmission of ‘ought’s from ends to means might be expected. If I know that you ought to lower your blood pressure, I may well conclude that, because of this, you ought to take the medicine. A burgeoning literature offers many proposals about how this instrumental transmission of reasons and ‘ought’s works. Common proposals are: Ought Necessity: If one ought to E, and M-ing is a necessary means to E-ing, then, because of that, one ought to M7 and: Strong Necessity: If there is reason for one to E, and M-ing is a necessary means to E-ing, then that is a reason, at least as weighty, for one to M,8 which in turn entails: Weak Necessity: If there is reason for one to E, and M-ing is a necessary means to E-ing, then that is a reason, of some weight, for one to M. However, Ought Sufficiency, Strong Sufficiency, and Weak Sufficiency, which substitute “sufficient means” for “necessary means,” have also been suggested.9 I begin by considering some objections to Strong Necessity, Ought Necessity, and the Sufficiency Principles: namely, they do not meet two basic desiderata, which are identified in sections 1 and 4. The problem with Strong Necessity, and perhaps also Ought Necessity, is that necessary means may do little to “probabilize” the end, as I explain in sections 2–3. The problem with Sufficiency principles is that sufficient means may be “superfluous” with respect to the end, as I explain in section 5.10 Section 6 puts forward a positive proposal, titled “General Transmission.” Indebted to the work of Raz (2005a and b), it aims to capture, admittedly more pedantically, the ideas animating his “Facilitative Principle” (although I leave explicit discussion of the connections to notes). One might wonder, however, why we should care what the correct account of instrumental reasons is. So long as we grant that there is such a phenomenon as instrumental transmission, why bother to describe it precisely? Perhaps it is enough to cite the aim, which fuels so much of philosophy, of making explicit the conceptual structures that implicitly underwrite our thought and practice. But there are two more specific reasons for caring what the principles of instrumental transmission say. First, such principles have been pressed into service by various treatments of the “normativity” of instrumental rationality: treatments of the demand, or seeming demand, to make our psychology means-end coherent. In particular, some have suggested that there are reasons to be instrumentally rational: for example, that one has “wide-scope” reason (either not to intend the end of, say, starting World War III, or to intend the means of launching a nuclear missile). Others have objected that, according to some principles of instrumental transmission, it would follow, absurdly, that one had “narrow-scope” reason to intend to launch a nuclear missile. While I think that this line of objection is fundamentally correct, its proponents have relied on questionable principles of instrumental transmission, and, accordingly, they have been met with plausible replies. Section 7 shows how General Transmission might put this objection on a surer footing. Second, once we get clear on what the principles of instrumental transmission would have to look like, we may begin to doubt whether there really is such a phenomenon as instrumental transmission. Section 8 explores this doubt. There I resist a line of thought that suggests, to a first approximation, that General Transmission is just a minor theorem of a certain kind of decision theory and, going further, that the very idea that reasons for means are explained by reasons for ends is a kind of myth. 1. First desideratum: means must probabilize the end In this section, I discuss the first of two basic desiderata for an account of instrumental transmission. Intuitively, reasons for an end, one’s E-ing, should transmit to a means, one’s Ming, to the extent that one’s M-ing “probabilizes” one’s E-ing, or “makes it likely.” If the end for which I have reason is my putting out the fire, and using a fire hose makes that more probable than using a garden hose, which in turn makes it more probable than blowing weakly on the raging flames, then, intuitively, I have more reason to use a fire hose than a garden hose, and more reason to use a garden hose than to blow. To a first approximation, we might say that if there is reason for one to E, and if there is positive probability, conditional on one’s M-ing, that one E’s, then that is a reason for one to M whose strength depends on the reason to E and the probability.11 But this won’t do. Suppose a boxer’s end is landing a punch. In every possible outcome in which he decides to throw a punch, he “telegraphs” his intention by conspicuously gritting his teeth. In spite of inadvertently warning his opponent, he nevertheless sometimes connects. So the probability, conditional on telegraphing, of connecting is positive. However, intuitively, no reason transmits to telegraphing. This is because at no possible outcome does it “help to bring it about” or “help to make it the case.” Instead, telegraphing is merely a by-product of something (deciding to throw the punch) that does help to bring it about. So I formulate the desideratum as: Means Probabilize: If there is reason to E and there is positive probability, conditional on one’s M-ing, that one’s M-ing, or some part of one’s M-ing, helps to bring it about that one E’s, then that is a reason to M, whose strength depends on the reason to E and on the probability. “Helping to bring about” need not mean only causing. It can also mean constituting, satisfying preconditions of, or preventing things that would prevent,12 as well as helping to cause, constitute, satisfy preconditions of, or prevent things that would prevent. This means that one’s refraining from doing something that would prevent one from E-ing may help to bring about one’s E-ing, insofar as such refraining, in effect, prevents a preventer. We can leave open whether there are still other ways of helping to bring about an end.13 Why “or some part of one’s M-ing”? We don’t even need this clause, if we assume that an action helps to bring about at an outcome whatever any part of it helps to bring about at that outcome. But suppose we do not assume this. Consider a case in which (i) giving the patient Drug A brings it about that we save his life at every outcome whether or not we also give Drug B at that outcome, (ii) that (giving Drug A and giving Drug B) does not help to bring it about that we save his life at any outcome, and (iii) that (giving Drug A and giving Drug B) brings it about that we improve his complexion at every outcome. Intuitively, we have at least as much reason (to give Drug A and to give Drug B) as we have to give Drug A. But, without the phrase “or some part of one’s M-ing,” why should this be? There would be very strong reason to give Drug A, because it is instrumental to saving the patient’s life, but only much weaker reason (to give Drug A and to give Drug B), which is instrumental only to improving his complexion. If we add “or some part of one’s M-ing,” by contrast, then (giving Drug A and giving Drug B) inherits the instrumental reason of its part, giving Drug A. The relevant notion of “probability” can be understood in different ways. I favor an epistemic framework, where the possible outcomes (perhaps sets of worlds or underspecified worlds) are ways things might be relative to a relevant body of information, and where the probabilities assigned to those outcomes are likewise fixed by that body of information.14 I stress that the relevant body of information need not be that of the agent at the time of acting. It may be instead the information, for example, of an advisor, or an onlooker. More generally, it may be contextually specified.15 However, there are other frameworks, which are compatible with much of what I go on to say. For example, on a historical framework, the possible outcomes are ways things might unfold, compatibly with how they have actually unfolded up to the relevant time (and possibly also constrained by certain physical and psychological laws), and the probabilities (here not epistemic) attached to those outcomes are determined by how things actually are (and the laws in play) at the relevant time.16 When I say that something is “necessarily” the case, that is equivalent to saying that it is the case at all possible outcomes, however these are understood. I assume that some positive probability attaches to every possible outcome (although, in principle, this may fail when there are infinitely many possible outcomes). 2. Why Strong Necessity has trouble with the probabilization desideratum If something along the lines of Means Probabilize is correct, it suggests an objection to Weak Necessity. “Weak Necessity may seem plausible, because if one fails to take a necessary means, then there is no chance of achieving the end. However, it is compatible with that truth that even if one does take the necessary means, one has no chance of achieving the end. If so, then those necessary means do not probabilize the end at all. So, if something along the lines of Means Probabilize is correct, then no reason would transmit to such necessary means.”17 However, there are plausible replies to this objection. First, if there is no chance of one’s E-ing even if one takes the necessary means, then there is no chance of one’s E-ing at all. And if there is no chance of one’s E-ing, then it is not clear that one has reason to E in the first place. That would follow, at least, from the conjunction of the claims: Reasons Availability: One has reason to X only if X is available to one. and: Weak Availability: X is available to one only if there is some possible outcome at which one X’s. Second, if there is no chance of E-ing, then there is a danger that everything, trivially, becomes a necessary means to E-ing. If no matter what one won’t E, then can’t one say of anything that, unless one does that, one won’t E? Taking these replies on board, I work with the following: Definition of Necessary Means: One’s M-ing is a necessary means to one’s E-ing iff: (i) in every possible outcome at which one E’s (one’s M-ing helps to bring about one’s E-ing and if one had not M-ed, one would not have E-ed), and (ii) there is some possible outcome at which one E’s. This definition ensures that necessary means probabilize the end to at least some degree. So, Means Probabilize gives us no grounds to doubt that some reason transmits to necessary means, at least on this definition. So Weak Necessity seems secure. Yet why should we expect Strong Necessity to hold? Granted, since necessary means probabilize the end to some extent, they inherit some of the reason. But necessary means might probabilize the end only to some minute degree. Why then should the necessary means inherit, as it were, all of the weight of the reason for the end? Consider Lucky and Unlucky, who occupy parallel universes. Each has an antique sitting on his front porch, which the rain threatens to ruin. A necessary means to saving the antique is taking a taxi back home. There is reason to refrain from18 taking the taxi; it costs money, say $20. But this cost is outweighed by the value of the antique, say $100. The only difference in their situations is that in Lucky’s universe, the rain will be slow in coming, and so he is very likely to get him home in time, if he takes the taxi: say that the probability is .9. In Unlucky’s universe, by contrast, he is extremely unlikely to get him home in time, even if he takes it: say the probability is .1. Intuitively—and as Means Probabilize implies—Unlucky has weaker reason to taxi than has Lucky, given the low probability that the taxi will get there in time. Therefore, if Lucky and Unlucky have the reason of the same strength to save the antique, then we have a counterexample to Strong Necessity. For Strong Necessity would then imply, incorrectly, that Unlucky has reason of the same strength as has Lucky to take the taxi. Lucky and Unlucky would seem to have reason of the same strength to save the antique. If Unlucky saves the antique, things will be, in relevant respects, just as they are if Lucky saves it. Granted, if Unlucky tries to save the antique (say, by taking the taxi), then things are not likely to be as they are if Lucky tries to save the antique. But that seems relevant not to the question of whether there is reason to save the antique, but instead to the question of whether there is reason to try to save it (say, by taking the taxi). If this is right, then Lucky and Unlucky have the same reason for the end. Means Probabilize implies that Lucky has more reason for his necessary means, since his means have a higher probability of achieving the end. This contradicts Strong Necessity, which says that there is as much reason for the necessary means as there is for the end.19 3. Why Ought Necessity may have trouble with the probabilization desideratum Is Ought Necessity likewise falsified? Can we say that, while Unlucky ought to save the antique, it is not the case that he ought to taxi? The argument that Unlucky ought to save the antique is as follows. Lucky ought to save the antique. As far as saving the antique is concerned, Unlucky’s situation is the same. If Unlucky saves the antique, things will be exactly as they are if Lucky saves it. The argument that it is not the case that Unlucky ought to taxi rests on the point just made: namely, that Unlucky’s reason for taxi-ing weakens as the probability lowers.20 By contrast, his reason to refrain from taxi-ing—namely, to avoid its cost—remains constant. So, as the probability lowers, a point is reached at which there is more reason to refrain from taxi-ing than to taxi. Assuming: Refrain Reasons-Ought: One ought to X only if there is not more reason to refrain from X-ing than to X, it would follow that it is not the case that Unlucky ought to taxi. This would also follow from: Relevant Reasons-Ought: One ought to X iff X-ing is, from among the relevant, available options, what one has most reason to do (or, in the case of ties, the disjunction of the options that one has most reason to do). However, one might worry that Relevant Reasons-Ought proves too much. If one ought to E, and the set of relevant alternatives to M-ing includes E-ing, then it will be the case that one ought to M only if M-ing is not only necessary, but also sufficient for E-ing. But this would mean that it isn’t the case that Lucky ought to taxi. More generally, if we accept Relevant Reasons-Ought, then Ought Necessity seems underspecified. With respect to which set or sets of relevant alternatives does Ought Necessity say that if one ought to E, then one ought to M? One answer is: Coeval Default: By default, the relevant alternatives to Z-ing are those alternatives that one might choose to do at the time at which one would Z: i.e., every available X such that there is some available Y such that it under one’s control knowingly to Y at the time that one would Z and necessarily, if one Y’s, then one X’s. Coeval Default represents the natural thought that the relevant alternatives to an action that one might choose to perform at a time are the alternatives that are open to one to choose at that time. It implies that Lucky ought to taxi, but that it is not the case that Unlucky ought to taxi, even though both ought to save the antique. If Ought Necessity is understood as claiming that when one ought to E, relative to its default alternatives, then it is the case that one ought to M, relative to its default alternatives, then, on this understanding of the default alternatives, it is false. I now address three main lines of reply to this argument against Ought Necessity. (The reader may skip these and turn to section 4 without any loss of continuity.) Reply 2 and, especially, Reply 3 parallel, in large part, “possibilist” criticisms of the “actualism” of Jackson (1985) and Jackson and Pargetter (1986)—or, more accurately, what Goble (1996) calls “probabilism.” Much of my response, then, is, in effect, a defense of probabilism. I doubt that many possibilists will be convinced; the gulf between basic intuitions may be too wide for that. But it is worth saying something about the debate for purposes of illustration. Reply 1: The first reply takes aim at the argument that Unlucky ought to save the antique, on the grounds of: Refrain Strong Necessity: If there is reason to refrain from M-ing (say, because M-ing would incur some cost), and if one M-ing is a necessary means to one’s E-ing, then that is a reason at least as strong to refrain from E-ing.21 Since Unlucky has stronger reason than Lucky to refrain from taking the taxi, he has stronger reason than Lucky to refrain from saving the antique. So, it is compatible with its being the case that Lucky ought to save the antique that it is not the case that Unlucky ought to save it. The first problem with Reply 1 is that Unlucky’s reason to refrain from taking the taxi is that doing so would avoid the cost of the taxi. But Lucky would seem to have a reason of precisely the same strength to refrain from taking the taxi: namely, that it would avoid exactly the same cost. So it is not the case that Unlucky has stronger reason than Lucky to refrain from taking the taxi. (Granted, Unlucky has weaker reason than Lucky to take the taxi: namely, its lower probability of saving the antique. But, again, that doesn’t imply less reason to save the antique, since the probability of Unlucky’s saving the antique saving the antique is the same as the probability of Lucky’s saving the antique saving the antique: namely, one.) Refrain Strong Necessity would seem, therefore, to transmit the same reason to refrain from saving in both cases. And in both cases there is just as much reason to save. But if the balance of reasons, for saving and refraining, is exactly the same for both Lucky and Unlucky, how then can it be that Lucky ought to save, but not Unlucky? The second problem with Reply 1 is that Refrain Strong Necessity seems questionable. Are we to accept it because it is a special case of: Strong Necessary Condition: If there is reason to X, and necessarily, one X’s only if one Y’s, then that is a reason at least as strong to Y? The arguments of the last section against Strong Necessity (as well as the points in note 7) already tell against this principle. The truth in the vicinity, it seems to me, is: Cost Avoidance: If there is a positive probability, conditional on one’s X-ing, that one’s X-ing helps to bring it about that one avoids some cost C, then there is reason to X, whose strength depends on the cost and the probability.22 In other words, the reason that Unlucky has to refrain from saving is that it is a means of avoiding the cost of taxi-ing. And how strong this reason is depends on the extent to which refraining from saving helps to bring it about that that cost is avoided. But Cost Avoidance doesn’t imply Refrain Strong Necessity. This is because avoiding the end is not always a good means to avoiding the cost of the necessary means. Indeed, this fact allows us to construct a different kind of counterexample to Ought Necessity. In Cost-Unlucky, the taxi is guaranteed to get Cost-Unlucky home in time. Here Cost-Unlucky’s misfortune is that the cost of the taxi is now $120: more than the value of the antique. Clearly, there is stronger reason to refrain from taxi-ing than there is to taxi, and so (according to Refrain Reasons-Ought) it is not the case that Cost Unlucky ought to taxi. But is there likewise stronger reason to refrain from saving than there is to save? According to Cost Avoidance, it turns on how probable it is, conditional on his refraining, that his refraining helps to bring it about that he avoids the cost of taxi-ing. And this turns on how Cost-Unlucky is likely to refrain from saving, if he does refrain. If—as we would usually expect—Cost-Unlucky is likely to refrain from saving, if he refrains, by refraining from taking the taxi, then, yes, refraining from saving is very likely to avoid the cost of the taxi. But if—to consider a more unusual case—Cost-Unlucky is likely to refrain from saving, if he refrains, by akratically taking the taxi and then dawdling on the front porch, gossiping with his neighbor, until the antique is ruined, then, no, refraining from saving isn’t very likely to avoid the cost of taxi-ing. In this more unusual case, Cost-Unlucky might have very weak reason to refrain from saving, since he is so likely incur the cost of the taxi anyway. By manipulating the probabilities, we can make his reason to refrain from saving weaker than his reasons to save. Thus, it would be consistent with Refrain Reasons-Ought, and entailed by Relevant Reasons-Ought (at least assuming Coeval Default), that he ought to save. As we might think: “It is not the case that Cost-Unlucky ought to take the taxi in the first place, since it costs more than the antique. But given that he is so likely (foolishly) to take the taxi whether or not he saves the antique, he ought at least to save the antique and cut his losses.”23 Reply 2: A second line of reply is to deny that Unlucky ought to save, because saving is not available to him. This reply would assume: Ought Availability: One ought to X only if X-ing is available to one, and something stronger than Weak Availability, such as: Strong Availability: X is available to one only if there is some 1, 2, … N such that it is necessary that knowingly 1-ing is fully under one’s control and if one knowingly 1’s, it will be necessary that there is some 2 such that knowingly 2-ing will be fully under one’s control and if one knowingly 2’s, … it will be necessary that there is some N such that knowingly N-ing will be fully under one’s control and if one knowingly N’s, then it will be necessary that one X’s. Saving the antique is not Strongly Available to Unlucky, so long as the probability that if he takes the taxi, he will get there in time, is less than one, which the counterexample, as we have been understanding it, requires. The first problem for Reply 2 is that Strong Availability is very restrictive. For instance, it would mean that saving the antique is not available even to Lucky, and so that it is not the case that Lucky ought to save it. Worse, in conjunction with Reasons-Availability (which was used to defend Weak Necessity), it would follow that Lucky does not even have reason to save the antique. Worse still, that would in turn threaten the judgment that Lucky even has reason to take the taxi. After all, any such reason would be instrumental, in the service of some end. But of what end? We cannot say that the reason is transmitted from the end of saving the antique, since we are denying that Lucky has reason for that end. It only pushes the problem back to say that it is transmitted from reason to try to save the antique. That reason is also instrumental, in the service of what such trying might bring about.24 More generally, Strong Availability tends to crowd out much of the phenomenon of instrumental transmission. Very often the ends for which we are said to have reason are not strongly available; something outside of our control might keep us from achieving them.25 The second problem for Reply 2 is that there are counterexamples to Ought Necessity where the end is strongly available. One such counterexample is Cost-Unlucky, for whom saving the antique was strongly available, albeit a bad bargain. Another such example is Procrastinate, described by Jackson (1985) and Jackson and Pargetter (1986). Procrastinate has strong reason to review a book. A necessary means to this is accepting the commission to review it. Accepting has one cost: namely, that of contacting the difficult-to-reach review editor in time to accept. Procrastinate is extremely unlikely to write the review, if he accepts. Reviewing the book is Strongly Available. If he accepts the commission, then it is necessary that writing the review will be, at a later time, fully under his control. But otherwise the structure of the case is relevantly similar to Unlucky. Intuitively, Procrastinate has weaker reason to accept than to refrain, and so Refrain Reasons-Ought and Relevant Reasons-Ought imply that it is not the case that he ought to accept. And on Relevant Reasons-Ought, Procrastinate ought to review. One can rule out these counterexamples by ratcheting up the requirements on availability to: Super-strong Availability: X-ing is available to one only if there is some Y such that necessarily, knowingly Y-ing is fully under one’s control, and if one knowingly Y’s, then it is necessary that one X’s.26