Instrumental Transmission 1. For many philosophers, it is a truism that if there is reason for the end, then there is, because of

Transcription

1 Instrumental Transmission 1 For many philosophers, it is a truism that if there is reason for the end, then there is, because of that, reason to take the means. 2 Whether reasons for ends derive from our desiring those ends, or whether they derive from their relation to things of independent value, these reasons transmit or transfer to their instruments. 3 Assuming that there is such a phenomenon as instrumental transmission, how is it best understood? I begin by criticizing some existing proposals, in particular that reason for the end transmits in its full force to necessary means. I then venture my own proposal: roughly, that reason for the end transmits to the means to the extent that the probability of achieving the end is higher if one takes the means. Among the implications of this principle, which I label General Transmission, are the following. It suggests that the difference between practical and theoretical reason is more pronounced than is often thought. It lends support to the charge that instrumental rationality, as it is commonly understood, is myth. It validates two rules for the detachment of wide-scope reasons. 4 It suggests, furthermore, that instrumental transmission may be a special case of a broader phenomenon, 1 Thanks to Ryan Milsap, Mark Schroeder, Kieran Setiya, and Jonathan Way. 2 See, for example, Bratman ms, Darwall , and 2006, Hubin 1999, Raz 2005a and 2005b, Kolodny 2007 and 2008b, Price , , Schroeder 2009, Setiya 2007, and Way forthcoming. Passages in Scanlon 2004 and Wallace 2006 assume or grant its correctness. Korsgaard 1997 criticizes the idea as it stands, but endorses something in its vicinity. 3 Some desire-based theorists may deny, against Hubin 1999, that their theory is a combination of the principle of instrumental transmission and the principle that reasons for ends are provided by desires. Instead, they may say, there is just one principle, a principle of, if you will, instrumental transmutation: if one desires the end, then one has reason to take the means. However, much of what the paper says about instrumental transmission applies, mutatis mutandis, to instrumental transmutation. This is particularly clear, if, as seems plausible, the principle of instrumental transmutation is extensionally equivalent to, even if not identical to, the combination of the principle of instrumental transmission and some principle that reasons for ends are provided by desires. 4 I discuss further implications of General Transmission elsewhere. Kolodny 2008a suggests that it lends support to the charge that practical consistency is a myth. Kolodny forthcoming suggests that it helps to explain how intending an end can sometimes give one reason to intend the means, most notably in the sort of tie-breaking cases discussed by Scanlon

2 which permits, first, transmission to actions that are not means to an end, but conditions that more broadly help to explain its occurrence, and, second, transmission of reason that we do not actually have, but would have under other conditions. Finally, it raises the possibility that instrumental transmission is itself something of a myth: that while reasons for ends are correlated with reasons for means, they do not explain them. 1. Necessary means? Many philosophers have been drawn to the claim I will call it Strong Necessity that reasons transmit their full force, weight, or strength to necessary means. Reasons for me to make something my end are, owing to the hypothetical imperative, equally reasons for me to take the necessary means to it (Darwall 1983, 16). If one has conclusive reason to believe that one will E only if one Fs, then one has reason to F that is at least as strong as one s reason to E (Kolodny 2007, 251). If X has an objective reason to do A and to do A X must do B, then X has an objective reason to do B of equal weight to X s objective reason to do A (Schroeder 2009, 245). If R is a practical reason in favor of X, X is attainable by the agent, and M is a necessary means to or necessary constitutive element of X, then R is a practical reason in favor of M (Bratman ms., 11). If you have a reason to do A and doing B is a necessary means to doing A, you have a reason to do B which is at least as strong as your reason to do A (Way forthcoming, ms. 21). Strong Necessity speaks only of pro tanto reason for. It says nothing about pro tanto reason against, or about whether, all things considered, one has conclusive, or even sufficient, reason to M. However, some have argued for Ought Necessity : that if one ought (or has conclusive reason), to take the end, then one ought to take the necessary means. If you should do E, all things considered, and doing M is a necessary means to doing E, you should do M, all things considered (Setiya 2007, 660). If X objectively ought to do A, and to do A X must do B, it follows that X objectively ought to do B (Schroeder 2009, 239). 2

3 I propose regimenting the two ideas as: Ought Necessity: If one ought to E, and it is necessary that (one E s one M s and one s M-ing is a means to one s E-ing), then one ought to M, and: Strong Necessity: If there is reason for one to E, and it is necessary that (one E s one M s and one s M-ing is a means to one s E-ing), then there is at least as much reason for one to M as there is for one to E. Some of the details of these formulations call for clarification. First, in what sense is the parenthetical condition supposed to be necessary? The abstract answer is: in whatever sense bears on the truth of judgments about what one has reason, or ought, to do. Without meaning to rule out other possibilities, I focus on two more specific candidates: P is historically necessary if and only if P is true at all worlds identical to the actual world at all times no later than the relevant moment, 5 and P is epistemically necessary if and only if P is true at all candidates for the actual world not ruled out the by the relevant body of information. Personally, I favor the second, epistemic interpretation, at least insofar as transmission principles are supposed to apply to the deliberative or decision-guiding senses of reason and ought. Claims about what there is reason for someone to do in this sense have their home in the firstperson deliberations of an agent deciding what to do going forward and, by extension, in secondperson advice that aims to help the agent reach a decision. And what matters in deliberation and advice, it seems, is what is known (or is epistemically likely) to be the case. For example, if everyone knows that the agent won t fly to the conference (that is it epistemically impossible), then the fact that he can fly to the conference (that it is historically possible) seems neither here 5 For an alternative in terms of histories rather than worlds, which is more hospitable to indeterminacy, see Belnap, et al

4 nor there for his deliberation, or for our advice, about whether he should book a ride to the airport. Of course, that fact may be relevant for our evaluation of him or his choices: for our blaming him for arriving a day late, or for judging that he is irrationally fearful. That being said, my discussion will, except for signaled exceptions, apply to both interpretations of necessary. Second, Ought and Strong Necessity do not apply to all necessary conditions of E-ing. Such necessary conditions would include: (in the famous paradox of Ross 1941) posting the letter or burning it, where E-ing is posting the letter, and (in an example of Price 2008) suffering a hangover, as it was known one would, where E-ing is drinking too much. The Necessity principles expressly apply only to necessary conditions that are necessary as means. Posting the letter or burning it is not a means to posting the letter, and a hangover tomorrow morning is not a means to tonight s bender. I assume that there is a relation, being a means to, that can hold between M-ing and E-ing at a possible world (although it may depend on what is true at other worlds). Some special cases of this relation are: causing or helping to cause, constituting or helping to constitute, and satisfying, or helping to satisfy, preconditions of. But this isn t meant as an analysis of means, let alone an exhaustive analysis. It will suffice to let intuitions about the correct use of the term be our guide. Finally, the truth or falsity of normative claims may be relative to certain features of context. For example, if one accepts the epistemic view of necessity that I favor, then their truthvalue will be relative to a contextually specified body of information. 6 Hence, the Necessity principles, and all of the other transmission principles that we will consider, should be 6 I favor the view of Kolodny and McFarlane ms. a and b: that the truth of occurrences of an ought -sentence depends on the information relevant at a context of assessment. As discussed in section 7, Cariani 2009 suggests that the truth of occurrences of an ought -sentence is additionally relative to a contextually specified set of alternatives. For a more general discussion of the context dependence of claims about reasons, see Price

5 understood as applying only within a fixed context. It is no argument against these principles that a use of There is reason for one to E is true relative to one context whereas a use of There is not reason for one to M is false relative to a radically altered context. Objectionable Means: Ought Necessity might seem to be refuted by cases in which there is strong reason against taking the necessary means. Suppose the end is slightly improving college policy, and a necessary means to that, since a unanimous vote is required, is lacing, with a lethal poison, the tea of the kindly old don who insists, without ever considering the merits, that nothing must change. Clearly, it is not the case that one ought to kill the don. However, it should be no less clear that it is not the case that one ought to improve college policy either. Of course, we can freely admit that, if circumstances were otherwise, it would be the case that one ought to improve college policy. But, the circumstances being what they are, one ought not. The explanation of this lies in something like: Strong Necessary Cost: If there is reason against one s M-ing, and one s E-ing requires one to M, then there is reason at least as strong against one s E-ing. Since the fact that that improving the policy requires killing the don is a vastly stronger reason against improving the policy than the fact that improving the policy would improve the policy is a reason for it, one ought not kill the don. Not only is not the case that one ought to take the necessary means, therefore, but also it is not the case that one ought achieve the end either. So this is not a counterexample to Ought Necessity after all. also to: Some may think, however, that this is a counterexample not only to Strong Necessity, but Weak Necessity: If there is reason for one to E, and it is necessary that (one E s one M s and one s M-ing is a means to one s E-ing), then there is some reason for one to M. 5

6 For, first, it is certainly odd to say that there is reason, namely improving college policy, to kill the don. Second, a fully virtuous agent would not treat the improvement to college policy as a reason for killing the don. He would give no weight to it, or even consider it, in his deliberations; it would be, for him, silenced. These observations seem to suggest that, in such cases, there is no reason to kill the don at all: no reason to take necessary means to the end of improving college policy. 7 I am reluctant to place much weight on this sort of silencing argument. First, I doubt that we should conclude from the observations that it is odd to say that there is reason to kill the don, and not fully virtuous to entertain in deliberation the thought that there is reason to kill the don, that it is false that there is such reason. As Schroeder 2004, 2005, and 2007 points out, it is a violation of the Maxim of Relation of Grice 1989 to say that there is a reason, when, as one knows, it is vastly outweighed. Thus, it can be odd without being false. Similarly, to entertain the thought, at least in the context of a live deliberation, does indeed indicate a vice. 8 (Doesn t it go without saying, or thinking, that the agent shouldn t kill the don, in which case entertaining thoughts about the reason for doing so is idle? So why is he entertaining thoughts about them? Is the verdict somehow not obvious to him? Or is he somehow tempted to defy it?) But this is compatible with the thought s being true. It similarly indicates a vice of one kind to entertain lascivious but true thoughts, during wedding vows, that sex with the officiant would be 7 See McDowell 1998a, b, and c; Price n. 9, 184 n. 72; and Setiya Thanks to Kieran Setiya for alerting me to this line of objection. 8 It is worth noting that McDowell sometimes characterizes silencing in this way: in terms of the virtuous agent s not treating the consideration as a reason instead of in terms of its not being a reason. Of course, on some views (and perhaps McDowell s own), there is little or no difference between the two; a reason just is a consideration that a virtuous agent would treat in a certain way. 6

7 pleasurable, and a vice of another kind to entertain distracting but true thoughts, during oral argument, that it is time to treat oneself to a new judicial robe. Second, if, despite this, we accept that there is no reason to take the necessary means, then we seem pressed to accept that there is also no reason for the end as we were pressed earlier to accept that one ought not achieve the end. Again, we can freely accept that, if circumstances were otherwise, there would be reason to improve college policy. But we seem pressed to accept that, the circumstances being what they are, in which improving college policy requires killing the don, there is no reason to improve college policy. After all, it seems similarly odd to say, and less than fully virtuous to think, such things as: There is still reason to go ahead and improve college policy here and now, even though, of course, this requires killing the don. And if there is not only no reason for the means, but also no reason for the end, then such cases are not counterexamples to Weak or Strong Necessity. The crucial point that we seem similarly pressed to accept that there is no reason for the end is also supported by the following, more analytical line of thought. First, the most plausible explanation of why there is no reason for the means (assuming that there is no reason) is that the reason against the means vastly outweighs the reason that there would be for the means if there were not so much reason against the means. The cost of denying this, it seems, is a kind of absolutism: that the fact that the means is a killing suffices, no matter what else is so, to make it the case that there is no reason for the means. But such absolutism seems implausible. If the killing was necessary for something of vital importance (e.g., it is the only way to stop the don from releasing a microbe that will destroy the human race you get the idea) it might well ring true to say that there was reason (even if outweighed) for it. Second, this suggests that, if the reason against the end similarly outweighs the would-be reason for the end, then there is, 7

8 similarly, no reason for the end. It would be odd to say that would-be reason for the means is silenced when outweighed, whereas would-be reason for the end is not silenced even when outweighed to the same degree. Third, in the present dialectical context, we cannot deny that Strong Necessity applies even in situations in which there is not strong reason against the means. At this point, no reason has been given for this denial. So we seem pressed to accept: Qualified Strong Necessity: If there is reason for one to E, and it is necessary that (one E s one M s and one s M-ing is a means to one s E-ing), then there is reason at least as strong for one to M, unless there is sufficiently greater reason against M-ing. Hence, we cannot say that the reason (or would-be reason) for the end is stronger than the would-be reason for the means. The remaining question is whether the reason against the end is somehow weaker than the reason against the means. Given Strong Necessary Cost, the fact that the end requires killing the don is surely no weaker a reason against the end than the fact that the means would be or bring about killing the don is a reason against the means. So, if we accept that there is no reason for the means, it is hard to see how we can claim that there is, as things are, reason for the end. 9 In sum, I doubt that in such objectionable means cases it is true that one has no reason to take the means, as opposed to having reason that is dramatically outweighed. And if one accepts that there is no reason to take the means, then similar considerations seem to press one to accept that there is no reason to take the end either. Either way, cases of objectionable means do not, I think, present a compelling argument against any of the Necessity Principles. 9 The foregoing is compatible, for the most part, with the particularist idea that a consideration that is a reason in one context may not be a reason in another. For example, that something would bring one pleasure is often a reason to do it, but not when the something is causing an innocent to suffer. All that I have been arguing against is the specific idea that, within a fixed context, a consideration can fail to be a reason for the necessary means while remaining a reason for the end that requires those means. 8

9 Ineffective Means: I find a different line of argument more convincing: that depending on the circumstances, taking necessary means can be more or less effective at realizing the end, and that this affects how much reason transmits to the means. Take Professor Procrastinate, a character from Jackson 1985 and Jackson and Pargetter He and Professor Dispatch have equally strong reason to review a book. A necessary means to this is accepting the commission to review it. There are no side-benefits to accepting; the only reason to accept, if there is any, transmits from the reason to review. Dispatch is extremely likely to write the review, if he accepts. Procrastinate is extremely likely not to write the review, if he accepts. (Here likely corresponds to whatever sense of necessary we have assumed is relevant to judgments about reason, where being necessary is being maximally likely. If we have assumed historical necessity, for example, then we might say that P is likely relative to a moment if and only if P is true in most of worlds identical to the actual world up until that moment; if epistemic necessity, then P is likely relative to a body of information if and only if P is true at most of the worlds not ruled out by the body of information.) According to Strong Necessity, Dispatch and Procrastinate have the same reason to accept: namely, as much reason as they have to write the review. But surely Dispatch has more reason to accept than Procrastinate has. 10 We would unhesitatingly advise Dispatch to accept, while being very reluctant to advise Procrastinate to do the same Compare Price This is a misdescription, one might object. While Procrastinate has just as much reason to accept, he has more reason against accepting. For example, his accepting might prevent someone else from accepting who would almost surely write the review, making it very likely that the book will not be reviewed by anyone. Hence, we are disinclined, for pragmatic reasons, to say that he has as much reason to accept just as we are disinclined, for pragmatic reasons, to say that one has reason to kill the don. (Thanks to Mark Schroeder and Ryan Milsap for suggesting this objection.) Yet it seems possible to imagine cases in which Professors Procrastinate and Dispatch have the same reason against accepting (e.g., say merely the cost of 9

10 Similar considerations argue against Ought Necessity as well. Suppose Dispatch s and Procrastinate s reasons to write the review are the same, and their reasons against accepting are the same, but Dispatch s reason to accept is very strong, whereas Procrastinate s reason to accept is very weak. In such a case, it may be that Procrastinate, like Dispatch, ought to write, but, unlike Dispatch, ought not to accept. 12 It might seem that a strengthened version of the Procrastinate case would refute Weak Necessity as well. Suppose that there is no chance that Procrastinate will write the review. In such a case, there seems no reason to accept. In the circumstances, whether or not Procrastinate accepts makes no difference to whether he reviews. Thus, as far as bringing it about that he writes the review is concerned, it would seem that Procrastinate has no more reason to accept than to decline. And surely, as far as bringing it about that he writes the review is concerned, Procrastinate has no reason to decline. So how, as far as bringing it about that he writes the review is concerned, can Procrastinate have reason to accept? 13 There are, however, two fair objections to this strengthened argument. First, if it is necessary that Procrastinate doesn t write the review, then it is no longer clear that he has any reason to write it in the first place. Indeed, in section 10, I will argue that one does not have reason for actions that are not available to one, and actions are available to one only if they are epistemically possible. (However, I also argue in section 10 that it is a mistake to limit transmission principles to ends that we actually have reason for. Reason for means is often transmitted from reason for the end that we would have, if the end were available. And if we replying to the review editor, who has written, You need only reply if you d like to review the book ). Suppose each is the only qualified reviewer, for example, so that no one else will review the book if he doesn t. It still seems that Procrastinate has less reason to accept, even though the pragmatic explanation no longer applies. 12 Way forthcoming 22 n. 32 reports a similar objection to Ought Necessity from John Broome. 13 Compare Price

11 were to extend Weak Necessity in this way, to include reason for the end that one would have reason for if the end were available, then the argument could be reinstated.) Second, it might be replied that Weak Necessity misrepresents what its supposed proponents really have in mind. Of course, they may say, there is no reason for a necessary means that doesn t actually make the end any more likely. 14 What they really mean to defend is: Weak Effective Necessity: If there is reason for one to E, it is necessary that (one E s one M s and one s M-ing is a means to one s E-ing), and there is nonzero probability (one E s one M s), 15 then there is some reason for one to M. This principle is not touched by the strengthened Procrastinate case. If it is necessary that Procrastinate does not write the review, then the second, effectiveness condition is not satisfied. No wonder that, in such a case, Procrastinate has no reason to accept. Even if Weak (Effective) Necessity survives, however, Strong and Ought Necessity do not. This raises the question of why they were ever thought true. Darwall suggests that Strong Necessity resembles a plausible principle of deductive transmission: [T]he force of reasons is, as it were, transferred back and forth along the line connecting an end and its necessary means in the same way that the rational force of a deductive argument is transferred between premises and conclusion (Darwall 1983, 47 48). And Ought Necessity may perhaps draw support from the broader principles by certain semantics for ought. I will try to explain why these sources of support are only apparent in sections 6 and 7, after General Transmission has been introduced. 3. Sufficient means? 14 Compare Price The previous condition that at every world at which one E s, one s M-ing is a means to this ensures that this probability is not merely evidential. 11

12 Writers who are attracted to Necessity principles tend to reject analogues about sufficient means. 16 Indeed, Ought Sufficiency: If one ought to E, and it is necessary that (one M s one E s and one s M-ing is means to one s E-ing), then one ought to M is routinely falsified. 17 I ought to get hold of some flowers, on behalf of the kids, for my wife on Mother s Day, but it s not the case that I ought to steal the ones my brother is bringing our own mother, although doing that certainly would be sufficient for getting hold of some for my wife. and But what about: Strong Sufficiency: If there is reason for one to E, and it is necessary that (one M s one E s and one s M-ing is means to one s E-ing), then there is at least as much reason for one to M Weak Sufficiency: If there is reason for one to E, and it is necessary that (one M s one E s and one s M-ing is means to one s E-ing), then there is some reason for one to M? 18 Broome 2005 offers an objectionable means argument against Weak (and in turn Strong) Sufficiency. Suppose one has reason to relieve one s hunger. Surely, he suggests, one has no reason to kill oneself, even those this will suffice to relieve one s hunger. My response corresponds to my first worry about the objectionable means argument against Weak and Strong Necessity. I am skeptical, with Raz 2005b and Price n. 9, of the claim that one has no reason to kill oneself. One has some reason, it seems to me. This much can be said for killing oneself: it would release one from these pangs of hunger. But it s laughably weak in comparison to the reasons against, and so it is both silly to report and alarmingly reckless to give serious 16 Bratman ms. is explicit about this. 17 The logic of satisfactoriness of Kenny 1966, however, validates an analogue for fiats, which are something like expressions of intention (and so verdictive or all things considered in the way that ought -judgments are). 18 A principle of this kind is endorsed by Way forthcoming

13 consideration. It is worth noting, however, that no response is available that would correspond to my second worry about the objectionable means argument against Weak and Strong Necessity. If one accepts that there is no reason to M, then one needn t feel similar pressure to accept that there is also no reason to E. If there are other means that are sufficient, then E-ing does not require M-ing. So we cannot appeal to Strong Necessary Cost to conclude that there is as much reason against E-ing as against M-ing. Superfluous Means: A more convincing argument against Strong Sufficiency, I think, is that sufficient means can be more or less superfluous. Suppose that there is another sufficient means, M*, besides M. It is very likely that Mr. Twoways will M* anyway, whereas it is very unlikely that Mr. Oneway will M*. It seems that Oneway has more reason to M than Twoways has. Moreover, a strengthened version refutes Weak Sufficiency. Suppose that it is necessary that one will M*. Then M-ing makes no difference to whether one E s. So no reason to E-ing, it seems, transmits to M-ing. E-ing is possible, so it cannot be said, as it could be against the strengthened Procrastinate case, that there is no reason to E. And the effectiveness condition that we added to Weak Necessity won t help Weak Sufficiency. There is nonzero probability that (one E s one M s). Weak Sufficiency seems to need a non-superfluity condition: not that there is nonzero probability that (one E s one M s), but instead that there is nonzero probability that (one E s and it is not the case that one would have E-ed even if one had not M-ed one M s). Suppose, then, we replace Weak Sufficiency with: Weak Non-superfluous Sufficiency: If there is reason for one to E, and it is necessary that (one M s one E s and one s M-ing is a means to one s E-ing) and there is nonzero probability that (one E s and it is not the case that one would have E-ed even if one had not M-ed one M s), then there is some reason for one to M. And suppose that we make the following 13

14 Assumption about Counterfactuals: In evaluating the counterfactual, It is not the case that one would have E-ed even if one had not M-ed, at a possible world, W, the closest worlds to W in which one does not M are themselves possible worlds. Then we have a principle invulnerable to the strengthened version of the argument. If it is necessary that one M* s, then, at every possible world, if one had not M-ed, then one would still have M*-ed, in which case one would still have E-ed. So the probability is zero, in which case the fact that there is no reason for one to M does not falsify the principle. 4. Raz s Facilitative Principle In what is perhaps the most developed discussion of instrumental transmission, Raz 2005a strikingly avoids endorsement of any of the Necessity principles, which are otherwise so popular. 19 Instead, he puts forward the: Facilitative Principle: When we have an undefeated reason to take an action, we have reason to perform any one (but only one) of the possible (for us) alternative plans that facilitate its performance (Raz 2005a, 6), and what we might call the: Steps Principle: Steps within a plan we have only conditional reason to take, the condition being that we have adopted and are pursuing the plan (and that it is still reasonably likely to facilitate what it is meant to facilitate) (2005a 6). The Facilitative and Steps Principles reflect a more or less explicit concern with the problems that we have discussed: objectionable means, less effective means, and superfluous means. First, the undefeated reason restriction rules out reason for objectionable necessary means, at least when they are so objectionable that there is not sufficient reason for the end (as in the don case). However, it does not rule out reason for objectionable nonnecessary means, which are compatible with there being sufficient reason for the end (as in the hunger case). (As I have said, though, I am skeptical that we need to rule out reasons of either kind.) 19 The closest he comes is, in Raz 2005b 9, to allow that the fact that means are necessary may affect the stringency of the reasons transmitted to them. 14

15 Second, Raz requires that M-ing be an at least somewhat effective means to E-ing: We have reason to adopt and pursue a plan only if, and for as long as, it affords reasonable chances of successfully facilitating realization of the action it is designed to facilitate (2005b 6). The concern with effectiveness also seems to lie behind the Steps Principle s requirement that we have adopted and are pursuing the plan. If we have not adopted or are not pursuing the plan, then our taking a particular step within the plan is likely to be akin to Procrastinate s accepting the review. It is unlikely that we will take other steps without which the present step has little or no chance of realizing the end. Finally, the but only one restriction seems to reflect a concern with superfluous means. I assume that the restriction to be understood in the following way. 20 Suppose that it is necessary that one takes some facilitating plan. Perhaps one already has taken it, so that it is historically necessary that one takes it, or perhaps it is known that one will take it, so that it is epistemically necessary that one takes it. Then there is not reason (at least as far as the end is concerned) to take another facilitating plan in addition, presumably because taking another facilitating plan would be superfluous. Yet a second plan need not be superfluous. For example, the second plan might be a failsafe, in case the first plan doesn t come to fruition. Isn t there then at least some reason to take the second plan? Now, if by facilitating plan Raz means sufficient means, then this can t happen. If it is necessary that one takes a sufficient means, then there s no possibility that it doesn t come to fruition. So additional facilitating plans would be superfluous. But Raz writes that the Facilitative Principle is only roughly speaking about sufficient means or what we may crudely and inaccurately describe as means sufficient for its realization (or ones which increase 20 Thanks here to Mark Schroeder. 15

16 its probability) (2005b 9). Perhaps although this is clearly speculation Raz intends his principle to apply more broadly to unimprovable means, where: one s M-ing is an unimprovable means to one s E-ing if and only if (i) there is a nonzero probability that (one E s and one s M-ing is a means to one s E-ing one M s) and (ii) if it is necessary that one M s, there is no distinct M* such that it is possible that (one E s and one s M*-ing is a means to one s E-ing and one would not have E-ed if one had not M*-ed one M* s). Unimprovable means need not be sufficient; they may fail to achieve the end. But they fail to achieve the end only when this couldn t have been helped: only when taking additional means would not have achieved it either. This would explain why the Facilitative Principle applies to plans, but not to steps within a plan, such as turning on one s computer, as a step within a plan to book a flight. Such a step is not unimprovable, since there are further steps (such as surfing to the website and selecting a flight) such that it is (to say the least) possible that if one did not additionally take these further steps, one would not book the flight. Condition (i) merely reflects the requirement, noted earlier, that the means be effective. It is condition (ii) that makes the but only one restriction reasonable. If it is necessary that one takes some unimprovable means, then taking another is indeed superfluous. The suggestion, then, is that we rewrite the Facilitative Principle as: Facilitative Principle First Revision: If there is (undefeated) reason for one to E, and one s M-ing is a unimprovable means to one s E-ing, and it is not necessary that one takes some distinct unimprovable means to one s E-ing, then there is some reason for one to M. The last condition would replace the but only one restriction. The problem is that one s M-ing might be superfluous even if it meets these conditions. It might be necessary that one takes some M* that is not unimprovable, but that nevertheless makes one s M-ing superfluous. Suppose that it is necessary that one gives the patient Drug A. 16

17 Drug A will cure the patient if the patient has Disease 1, as seems likely. However, administering Drug A is not unimprovable. If one additionally gives the patient Drug B, this will not interfere with the effect of Drug A and will cure the patient on the off chance that the patient has Disease 2. If one additionally gives the patient Drug C, this will not interfere with the effect of Drug A and will itself have the same effect, so that now it is causally overdetermined that if the patient has Disease 1, the patient will be cured. However, giving Drug C is unimprovable, let us suppose, since it precludes Drug B from having any effect. According to Facilitative Principle First Revision, one has reason to give Drug C even if it is necessary that one gives Drug A. If we concerned about superfluous means in general, then it seems that we should rule this out, with a further condition: where: Facilitative Principle Second Revision: If there is (undefeated) reason for one to E, and one s M-ing is a non-superfluous unimprovable means to one s E-ing, and it is not necessary that one takes some distinct unimprovable means to one s E-ing, then there is some reason for one to M, one s M-ing is a non-superfluous unimprovable means to one s E-ing if and only if (i) there is a nonzero probability that (one E s and one s M-ing is a means to one s E-ing and it is not the case that one would have E-ed even if one had not M-ed one M s) 21 and (ii) if it is necessary that one M s, there is no distinct M* such that it is possible that (one E s and one s M*-ing is a means to one s E-ing and one would not have E-ed if one had not M*-ed one M* s). Notice that this obviates the need for the second condition of Faciliative Principle Second Revision. If it is necessary that one takes some distinct unimprovable means, then M-ing is superfluous. We might therefore rewrite it as: 21 The effectiveness and nonsuperfluity conditions have to be, as it were, jointly satisfied. It is no good if M is effective only at worlds where it is superfluous, and nonsuperfluous only at world where it is ineffective. 17

18 Facilitative Principle Third Revision: If there is (undefeated) reason for one to E, and one s M-ing is a non-superfluous unimprovable means to one s E-ing, then there is some reason for one to M. 5. General Transmission We now have a list of plausible principles: Weak Effective Necessity, Weak Non-superfluous Sufficiency, Facilitative Principle Third Revision, and the Steps Principle. But we are still short of a complete account of instrumental transmission. First, reasons for ends transmit to means that are neither necessary, nor sufficient, nor unimprovable, nor steps in adopted, unimprovable plans. For example, although an aimless teenager has not adopted any plan that contains taking the SAT as a step, a parent may know that an aimless teenager has reason to take the SAT exam as a means to a college education (perhaps partly because the parent knows that, in time, the teenager will begin to take his future seriously and undertake other steps, such as filling out application forms). Second, none of these four principles, unlike Strong Necessity and Strong Sufficiency, tells us how much reason is transmitted. Nor do we have a unified account of instrumental transmission, if such an account is in the offing. There remains the question whether anything more basic explains this list of transmission principles. Notice that Facilitative Principle Third Revision is of the form: If there is reason for one to E, and if there is nonzero probability that (one E s and one s M-ing is a means to one s E-ing and it is not the case that if one had not M-ed, one would have E-ed one M s), and one s M-ing meets some other conditions, then there is reason for one to M. Weak Effective Necessity can also be put in this form. If it is necessary that (one E s one M s and one s M-ing is a means to one s E-ing), then, first, at every world in which one E s, one s M-ing is means to one s E-ing. Second, given our Assumption for Counterfactuals, at every world at which one E s it is false that one would have E-ed if one had not M-ed. So if there is a nonzero probability that (one E s one M s), then there is a nonzero probability that (one E s and 18

19 one s M-ing is a means to one s E-ing and one would not have E-ed if one had not M-ed one M s). Weak Non-Superfluous Sufficiency is also of this form. If it is necessary that (one M s one E s and one s M-ing is a means to one s E-ing), then at every world at which one M s, one s M-ing is a means to one s E-ing. So if there is a nonzero probability that (one E s and one would not have E-ed if one had not M-ed one M s), then there is a nonzero probability that (one E s and one s M-ing is a means to one s E-ing and one would not have E-ed if one had not M-ed one M s). This suggests a possible unification: namely, that each of these principles is merely a special case of a more general principle: If there is reason for one to E, and if there is nonzero probability that (one E s and one s M-ing is a means to one E-ing and it is not the case that if one had not M-ed, one would have E-ed one M s), then there is reason for one to M. This principle, in turn, promises a complete account of instrumental transmission. At very least, it covers the SAT case. Since it is probable that the teenager will eventually take other steps jointly sufficient for going to college if he takes the SAT, and since there is no chance of a college education unless the teenager takes the SAT, there is a nonzero probability that (the teenager goes to college and his taking the SAT is a means to this and it is not the case that if he had not taken the SAT, he would have gone to college he takes the SAT). Moreover, once this principle is in view, it suggests a natural answer to the question of how much reason is transmitted: more reason the higher the probability. This would explain why Dispatch more reason than Procrastinate, and why Oneway has more reason than Twoways. The suggestion, then, is that the phenomenon of instrumental transmission is captured by: General Transmission: If there is reason for one to E, and there is nonzero probability that (one E s and one s M-ing is a means to one s E-ing and it is not the case that if one had not M-ed, one would have E-ed one M s), there is reason to M and more reason the 19

20 higher this probability and at least as much reason as there is to E if the probability is one. 22 Dropping the means condition? It might be thought that and one s M-ing is a means to one s E-ing is redundant. If one M s and one E s and it is not the case that if one had not M-ed, one would have E-ed, doesn t follow that one s M-ing is a means to one s E-ing? Not obviously. Suppose a boxer invariably telegraphs that he is about to throw a punch by setting his back foot. It is true, at every world at which he lands the blow, that he would not have landed the blow if he had not telegraphed it (because if he had not telegraphed it, he would not have thrown it at all). But it seems wrong to say that telegraphing the punch is a means to landing it. Instead, it is a necessary consequence of something that is a means to it, such as the decision to throw the punch. Even if and one s M-ing is a means to one s E-ing is not redundant, it may be too restrictive. That is, if we replace this condition with the more inclusive and one s M-ing helps to explain why it takes place that one E s, 23 General Transmission may be able to explain a further phenomenon: how a cost provides reason against doing what incurs that cost. Suppose we grant, as seems plausible, that reason against X-ing just is reason to refrain from X-ing. Then Strong Necessary Cost, from section 1, becomes: If there is reason for one to refrain from M-ing, and it is necessary that (one E s one M s), then there is reason at least as strong for one to refrain from E-ing. 22 Compare a proportional transmission principle suggested by Schroeder : If X has reason to do A and X s doing B would facilitate her doing A, then X has reason to do B of weight at least proportional to X s reason to do A, and to how well her doing B would facilitate her doing A. 23 Notice that the simpler helps to cause would not be more inclusive than is a means to, since something can be a means not in virtue of causing the end, but instead in virtue of (say) helping to constitute it. 20

21 For now familiar reasons, this cannot be accepted as it stands. The reason for one to refrain from E-ing varies with the probability that one will refrain from M-ing given that one refrains from E- ing. If it is necessary that one will M no matter what, then it is not clear that there is (this) reason to refrain from E-ing at all. It seems that we need to add an effectiveness condition: Weak Effective Necessary Cost: If there is reason for one to refrain from M-ing, and it is necessary that (one E s one M s), and there is nonzero probability (one refrains from M-ing one refrains from E-ing), then there is some reason for one to refrain from E-ing. Note that if it is necessary that (one E s one M s) and there is nonzero probability that (one refrains from M-ing one refrains from E-ing), then there is nonzero probability that (one refrains from M-ing and it is not the case that one would have refrained from M-ing if one had not refrained from E-ing one refrains from E-ing). So we needn t worry that refraining from E- ing might be superfluous: that one might have refrained from M-ing anyway. Still, the principle still isn t quite right. Suppose a boss, similar to our telegraphing boxer, invariably mutters to himself after he yells at his subordinates and invariably refrains from muttering whenever he refrains from yelling. There is nonzero probability that (the boss refrains from yelling and it is not the case that he would have refrained from yelling if he had not refrained from muttering the boss refrains from muttering). We don t want to say that because the boss has reason to refrain from yelling, he has reason to refrain from muttering. To rule this out, we need something like the condition that there is a possibility that refraining from E-ing is a means to refraining from M-ing. But it seems odd to say that refraining from the end is a means to refraining from the necessary means. We need something broader: that refraining from the end helps to explain why it takes place that one refrains from the means. One refrained from the means because one refrained from the end. By contrast, the boss s refraining from muttering does not help to explain why it takes place that the boss refrains from yelling (although it may be 21

22 evidence that it has taken place). He did not refrain from yelling because he refrained from muttering. If we add this condition, then Weak Effective Necessary Cost is of the form: If there is reason for one to refrain from M-ing, and there is nonzero probability (one refrains from M-ing and one s refraining from E-ing helps to explain why it takes place that one refrains from M-ing and it is not the case that one would have refrained from M- ing if one had not refrained from E-ing one refrains from E-ing), and M-ing meets some other condition, then there is some reason for one to refrain from E-ing. Hence, it would be a special case of General Transmission (where M s in General Transmission is filled in by refrains from E-ing and E s is filled in by refrains from M- ing ), if we replaced is a means to with the broader helps to explain why it takes place that. 24 Is there any reason against this broadening? One might worry that General Transmission would then validate the following principles: Disjunction: If there is reason to E and it is possible that one E s, then there is reason to (E or F). Conjunction: If there is reason to E and it is possible that (one E s and F s and if one had not both E-ed and F-ed, one would not have E-ed), then there is reason to (E and F). Disjunction would imply, counterintuitively, that our boxer has reason (to throw the punch or to dance the hokey-pokey). And Conjunction would imply, counterintuitively, that our invariably 24 To illustrate these points, consider an example of illegitimate practical reasoning from Price , modified in a way to which I expect he would not take exception: I have reason to get drunk. Whoever gets drunk is bound to have a hangover. So, I have reason to have a hangover. Price notes that the conclusion can be blocked if we accept transmission only to means, not to consequences generally. As I see things, this is overkill, in two ways. First, it is enough to block the conclusion to accept transmission only to what helps to explain why the end takes place. Second, if we accept transmission only to means, rather than to what helps to explain why the end takes place more generally, then we risk depriving ourselves of an explanation of the following legitimate reasoning: I have reason not to have a hangover. Whoever does not get drunk does not have a hangover. So, I have reason to not to get drunk. 22

23 telegraphing boxer has reason (to throw the punch and to telegraph it). Neither (E-ing or F-ing), nor (E-ing and F-ing) is naturally thought of as a means to one s E-ing. But the worry goes they might be said to help to explain why it takes place that one E s. It is not terribly counterintuitive to say that one s (E-ing and F-ing) helps to explain why it takes place that one E s. By the same token, however, there seems to be something to be said for one s (E-ing and F-ing): namely, that it realizes or brings about one s E-ing. Perhaps then Conjunction is not so troubling. Note that Conjunction does not imply that one ought to (E and F) follows from one ought to E, which is clearly unacceptable. 25 Disjunction is more troubling, however, because it licenses a kind of reasonlaundering. 26 Suppose that (i) there is reason to E, (ii) it is possible that one E s, and (iii) it is possible that (one F s and if one had not F-ed, one would not have (E-ed or F-ed)). Then, from Disjunction and (i) and (ii), one has reason to (E or F). Then, from General Transmission, (iii), and the plausible assumption that whenever one F s, this helps to explain why it takes place that one (E s or F s), it follows that there is reason to F. This is troubling enough, since F was almost arbitrary. Worse, there is reason to F transmitted from reason to E. Yet our suppositions are compatible with there being zero probability that one E s given that one F s. In that case, there would be E-given reason to do something that ensures that E does not occur! 25 With reference to Anscombe , Price 2008, 22 writes: So we cannot permit as pieces of practical reasoning inferences from I will phi either to I will phi or chi or to I will phi and chi, where or chi adds an arbitrary disjunct and and chi an arbitrary conjunct. Nothing here commits me to disagreeing with this. For all I have said, it might be the case that one ought to phi, but also the case, as one knows, that one ought not to do the disjunction or conjunction. In that case the first intention would be correct, but the latter ones not. Whether Anscombe or Price would reject similar inferences from There is reason for me to phi to There is reason for me to phi or chi and There is reason for me to phi and chi, however, is less clear to me. 26 [Ryan Milsap unpublished?] 23