Friday, March 12, 2010

Hypothetical Imperatives

The distinction between categorical and hypothetical imperatives is typically said to be that only the latter depend on our desires. This is potentially misleading. Certain categorical imperatives may also be conditional on our desires, after all. E.g. "if you want to torture children, you should seek psychiatric help." Of course, here the required course of action is not merely a means to fulfilling the antecedent desire. But we could imagine a (perhaps not very plausible) view -- somewhat akin to ethical egoism -- according to which we're categorically required to act so as to fulfill our present desires, whatever they may be. How are hypothetical imperatives different?

One formal difference to note is that categorical, but not hypothetical, imperatives allow for modus ponens, or detachment of the consequent obligation. Consider:

Argument 1:
If you want to torture children, you should seek psychiatric help.
You want to torture children.
Therefore, you should seek psychiatric help.

That seems perfectly valid. But suppose we replace the first premise with a merely hypothetical imperative:

Argument 2
If you want to torture children, you should volunteer as a babysitter.
You want to torture children.
# Therefore, you should volunteer as a babysitter.

When we affirm the first premise as a mere hypothetical imperative, we mean it in a sense that does not validate such an inference. We might add, "But of course you shouldn't want to torture children, and so you shouldn't take the means to this atrocious end either."

This suggests that we can't take the conditional at face value. One natural thought is that the hypothetical imperative is doing nothing but highlighting a means-end relationship. But as Sidgwick noted, this can't be quite right: there's some added element of normativity here, tricky though it may be to pin down. A more promising suggestion is to interpret the 'should' of the hypothetical imperative as really taking wide scope (i.e., "you should: either not want to torture children, or volunteer as a babysitter"). But does that imply that, if you fail to satisfy the first disjunct of the requirement, there's something to be said for the sadist's deciding to satisfy the second disjunct instead? One would hope not. Perhaps the best solution here is to interpret the antecedent of the hypothetical imperative as itself implicitly normative; something like: If you rightly want to torture children, you should volunteer as a babysitter.

Either way, the crucial point is that hypothetical imperatives don't necessarily have normative force even when the agent possesses the desire in question. Even if you have the desire in question, you may still disregard it on the grounds that it is misguided and lacking in normative authority. (This is a point Velleman makes nicely in his 'Brief Introduction to Kantain Ethics'.) Hypothetical imperatives are thus best understood as depending, not merely on your having certain desires, but rather on those desires having normative authority. The hypothetical imperative "if you desire X then you should Y" really means something like, "supposing that X is a worthy goal, you should Y."

Hypothetical imperatives thus reveal patterns of normative inheritance. But their highlighted 'means' can't inherit normative status unless the 'end' in question had prior normative worth. A view on which there are only hypothetical imperatives is effectively a form of normative nihilism -- no more productive than an irrigation system without any water to flow through it.

This observation puts advocates of Humean "desire-based views" in a bind. They had hoped to achieve a kind of unmysterious, uncontroversial normativity. But so long as they merely affirm hypothetical imperatives, they're stuck with a flimsily dressed up nihilism. Alternatively, if they say that we're categorically required to act on our present desires, then their view is absurd. ("You really mean that the self-controlled sadist is making a mistake when he disregards his urge to torture people!?") Once we think there are genuine reasons to do some things rather than others, we may as well pick a more plausible theory of what it is that we have reason to do.


  1. In "Relatives and Relativism" Jeske and Fumerton argue that a consequentialist who wants to defend special obligations to intimates is committed to "radical relativism" about value. This is not exactly egoist, since one may have genuinely other-directed values on the view they discuss.

    The conclusion about the sadist you discuss might be more plausibly taken at face value if the "should" in question is the "should" of rationality not ethics. Consequentialism plus radical relativism about value indeed seems to just collapse the ethical ought into the ought of maximizing expected subjective utility, typically a conception of rationality.

    I suspect Fumerton might think this collapse is a positive feature of the view. Consider the following question: "Should I do what I should-rationally do or what I should-ethically do?" To the extent that should-rationally and should-morally are distinct notions there's a deep problem about how to even understand the first "should". But if the shoulds all collapse the question is trivial and the answer is yes.

    If we're really bothered by what the sadist should do according to this view then maybe we can take solace in the fact that (given our values) we all should do everything we can to stop the sadist. Or we can give up the idea of special obligations to intimates... or give up consequentialism all together.

  2. Also, some sadists may wish to be rid of their sadistic desires because they value human well being. In which case they "should" see a psychiatrist. But the unrepentent sadist who does no value the well being of others "should" torture them, according to radically relativist consequentialism.

  3. Richard, I think Korsgaard makes similar criticisms of Humean "desire-based views" in her (1997) "The Normativity of Instrumental Reason".

    Don't quite see your reason for thinking that Argument 1 allows detachment of consequent obligation, while Argument 2 does not. The claim that you should NOT want to torture your children is consistent with both arguments, is it not? I wonder if you could explain why that claim implies the negation of the conclusion in Argument 2.

    The main reason for thinking that hypothetical imperatives have non-detaching oughts in modus ponens, it seems to me, is that we could then derive detached ought-statements that conflict with our better-considered judgments about what we ought to do. This leads us to suppose that the detached ought-statements are false, and thus that hypothetical imperatives have non-detaching oughts.

    But if that's the main reason for thinking that the initial premiss in Argument 2 has a non-detaching ought, then the same kind of reason seems to apply to the initial premiss in Argument 1. Suppose you promised never to seek psychiatric help. Then it seems you should not seek psychiatric help, so the conclusion of Argument 2 is false, which should be impossible if the argument is valid. Perhaps we can say that we can have conflicting obligations, each derived validly... but then why can't we also have conflicting obligations in the case of hypothetical imperatives?

    As for the nihilism of which you accuse Humean views, perhaps you are underestimating the resources available to such views, esp. to Hume's own view. Hume's own view makes room for stable desires that survive the test of critical and consistent reflection, and for the consistent use of sympathy that affords us an impartial point of view from which to evaluate states of affairs, actions, motives, etc.

  4. Hi Boram, I take it to be analytic that categorical but not hypothetical imperatives allow for detachment. That's just what we mean by the two kinds of conditionals. I don't know how to give a positive argument for analytic claims -- you should just be able to see it.

    To defuse your objection: note that if you convince me that really the sadist shouldn't seek help (because it'd violate an oath or something), then I'll retract my categorical claim that they ought to. The inference is perfectly valid, so if I don't accept the conclusion, I'm forced to give up the premise. There's nothing here to suggest that the inference is invalid. Valid arguments can have false conclusions, if one of the premises is false.

    The case of hypothetical imperatives is clearly different. There's no rational tension between affirming both premises together with the negation of the "conclusion". I don't need to retract my hypothetical imperative claim, because it's perfectly compatible with the claims that you want to torture kids and yet that you (categorically) ought not to become a babysitter. The claim that you want to torture kids, together with the hypothetical imperative that "if you want to torture kids you should become a babysitter", simply does not commit me to the conclusion that you should become a babysitter. (There's a sense of "if you want to torture kids you should become a babysitter" which would commit me to the validity of this inference -- that's what I'm calling the 'categorical imperative' interpretation of the conditional. But of course I'm not at all tempted to think that that version of the conditional is true, precisely because I wouldn't be willing to draw the conclusion from it.)

  5. Richard,

    Suppose you have to decide whether to fire Larry, Moe, or Curly. (You can't fire them all, but someone should (morally) be fired for doing what they did.) Suppose it's deontically best to fire Larry, second best to fire Moe, and third best to fire Curly. (Larry is the most responsible, say.) Suppose, however, that Larry is your favorite.

    It seems that we should say:
    (1) You should fire Larry.
    (2) If you don't fire Larry, you should fire Moe.

    I don't think you can detach (3) from (2) and (4):
    (3) You should fire Moe.
    (4) You won't fire Larry.

    I don't think (2) is a hypothetical imperative.

    (Also, the wide-scope account isn't very helpful here because while we want to say that (2) is true, we don't accept:
    (5) If you don't fire Larry, you should fire Curly.

    I don't think the limited resources of the wide-scoper can account for the fact that (2) is true but (5) isn't. (Although, "If you don't fire Larry or Moe, you should fire Curly" is true.))

  6. Richard, thanks for the reply, I see what you mean more clearly now. I'm raising some further questions because this is a topic (I mean on HI, not CI) I've been working on and find interesting.

    Now, if you retract the initial premiss of Argument 1, aren't you denying the conditional (but not means-to-end) connection between some categorical oughts and having sadistic desires? And as far as I can see the worry about conflicting obligations seems generalizable, so with your strategy we might end up denying any conditional connection between categorical oughts and having desires.

    Perhaps by retracting the initial premiss of Argument 1 you don't mean accepting its negation, but replacing it with another conditional, like: if I want to torture children and I have not made a promise to avoid psychiatrists, then I should see a psychiatrist. By allowing for exceptions in this way, you get to preserve the conditional connection you see between the antecedent and the consequent. But thoroughly following this strategy would likely involve a very long and perhaps open-ended ceteris paribus clause in the antecedent. And if this strategy is available in the case of the initial premiss in Argument 1, why not likewise for the initial premiss in Argument 2?

    You may be right that by definition hypothetical imperatives are non-detachable while categorical imperatives are not. I have not come across any philosopher who claims that hypothetical oughts have narrow scope and are detachable. But as you know there are philosophers like Schroeder who give the narrow scope reading to hypothetical reasons, and it doesn't take a big leap from that position to the position that hypothetical oughts also allow for a narrow-scope reading.

    I can do that in this way. First claim that "ought" is ambiguous between:
    (a) all-things-considered oughts, and
    (b) pro tanto oughts.
    Then read Argument 2 as involving pro tanto oughts throughout (so, for instance, don't read the initial premiss as having a pro tanto ought and the conclusion as having an all-things-considered ought... that's fallacy of equivocation).

    And if you want to preserve the conditional connection in the initial premiss of Argument 1, then I suggest doing the same as well: read it as involving pro tanto ought throughout, and the problems I mentioned earlier can be avoided. Or, if you find all this ridiculous, you can read both Argument 1 and Argument 2 as having wide scope.

  7. P.S. Ignore the 3rd paragraph (about building in exceptions in the antecedent) in my last comment, because it does not result in a valid argument when the exceptions apply....

  8. Clayton - good point, there are some 'second-best' ought conditionals that don't detach either (though they aren't thereby mere hypothetical imperatives). So non-detachability is not sufficient to qualify as a hypothetical imperative. That's compatible with my 'normative inheritance' account of hypothetical imperatives. The explanation of non-detachability in your case is going to be very different from the sort of 'inheritance' explanation I've given of hypothetical imperatives.

    Boram - we can also offer generalizations about pro tanto reasons (or, better yet for purposes of formulating broad generalizations, prima facie duties). That's a different claim from the particularized categorical imperative I mentioned, which is why retracting the conditional claim in one particular case does not amount to "denying the [general] conditional (but not means-to-end) connection between some categorical oughts and having sadistic desires".

    So, to clarify: there's an important interpretation of the claim "If you want to torture children, you should seek psychiatric help" which allows for modus ponens, and yields a conclusion about what you ought (conclusively) to do. (Note also that it can also very easily be true -- there's no universal quantifier in there -- the conditional concerns some particular person 'you'. It seems likely to be true for the overwhelming majority of possible 'you's.) This is the interpretation I have in mind when I talk about 'categorical imperatives' in this post.

    Moving on now to argument 2: re-interpreting it as a 'pro tanto' ought doesn't evade my worries. (Aside: it strikes me as bad English -- who says "you ought to phi" when they really mean "you have some reason to phi?" Oughts are conclusive; to make a merely pro tanto claim you should talk about reasons.) The most important problem here is that by affirming the hypothetical imperative, I'm not committed to thinking that there's any reason at all for the sadist to take the means to his vicious ends.

  9. Richard, I agree with you that that's the most important problem, but perhaps you will also agree that it's disputable that there's no reason at all for the sadist in your scenario (Schroeder, Williams and Hume for instance will likely dispute it).

    I want to return to your normative inheritance account of instrumental reasoning, and the charge that Humean desire-based theories leads to normative nihilism. I don't see why that has to be the case. Here is one way out. A Humean can maintain that if you desire that p, then it follows that you ought (ceteris paribus) to make it the case that p, given the world-to-mind direction of fit that desires have. Here 'ought...' has to be given a pro tanto rather than a conclusive reading, and if you don't like it then you can replace it with 'have a reason'. This question of the existence of a reason can be separated from the question of the weight of that reason, a la Schroeder. And I suggested in my first comment some Humean ways of assigning weight to reasons, and identifying what one has most reason to do (e.g., by using sympathy and consistent reflection).

    p.s. As for taking ought in "you ought to phi" in the pro tanto rather than conclusive sense.... Is it really bad English? Sure, if I just blurt out "you ought to phi", then I cannot reasonably expect you to take my advice in any specific pro tanto sense. But we usually utter ought-statements in some context or other.

    So, my position is that conversational background will determine whether the pro tanto or the all-things-considered reading of ought-statements will be the appropriate one. Alternatively, I could allow that an an ought-statement invariably expresses an all-things-considered judgment, so long as context suitably restricts the domain of quantification. Or, even if you firmly establish that, as a matter of linguistic fact, an ought-statement can only be used to express an all-things-considered judgment, this limitation in the expressive resources of the language can be overcome by enriching or reforming the language to adequately represent the facts that we are interested in. So I could introduce a pro tanto 'ought' in practical reasoning to reflect the existence of pro tanto reasons.

  10. Right, it's an open/disputed question whether his desire gives the sadist a reason to engage in torture. Personally, I think it doesn't. But the crucial point for my purposes here is that affirming the hypothetical imperative does not in itself commit me to thinking that any such reason exists. I haven't contradicted myself if I affirm the hypothetical imperative but then add "though there's nothing at all to be said for your taking the means to this rotten end." Hence the need to give some account of the meaning of hypothetical imperatives which doesn't analytically imply that there's any reason to follow them.

    Once we have such an account of hypothetical imperatives, we can ask the further question whether, as a matter of substantive normative fact, we always do have at least some ('categorical', normative) reason to fulfill any desire. You suggest that Humeans will make this further claim. I think this places them on the back foot, since I take it that the main motivation for desire-based views is not that they yield intuitively plausible results about what we have reason to do (they don't), but rather that other views are thought to be unacceptably "mysterious" in the kind of categorically binding normativity that they posit. If (as per the interpretation we're currently considering) the Humean is positing the same kind of high-falutin' normativity, and merely disagreeing about which acts possess this special status, then their view seems unmotivated. That's the worry, at least.

  11. [Andrew Sepielli writes:]

    One strategy is to take the should in the second inference as the should of rationality. Since the conditional is obviously no candidate for being a norm of rationality, things would have to be expanded a bit:

    1) If you want to torture children, you should torture children. (Narrow-scope rational norm)
    2) If you should torture children, you should volunteer as a babysitter.
    3) Therefore, if you want to torture children, you should volunteer as a babysitter.
    4) You want to torture children.
    c) You should volunteer as a babysitter.

    My guess is that you will want to reject 1), since it's a narrow-scope rational norm. I like narrow-scope norms, but I don't think desires have anything to do with rational action, so I don't like it either, but it's at least promising, I think.

    Now we're left with the nettlesome question of how to distinguish what it's rational to do from what we have objective reasons to do, given that both may depend on desires. My suggestion is that what it's rational to do is *explained fundamentally* by the intentional contents of our mental states, while what we have objective reason to do is not. So why rationally should I torture children? Because "to torture children" is the content of my desire. (I don't think that, but that's what defender of the view I'm sketching would say.) That is where the explanation bottoms out. Why objectively should I seek psychiatric help? Because of the content of my desire, sure. But there is a further level of explanation. Why should someone with a desire like this seek psychiatric help? Because such a person is dangerous/psychologically unhealthy/sociopathic/etc. I've been toying with this as a way to distinguish between rational and other norms generally, so I thought I'd throw it out there...

    [end quote]

  12. Richard, I would avoid the term "categorical" for the kind of normativity I have in mind, namely for the principle:

    (DO) S desires that p -> S ought to see to it that p.

    and interpreting it more specifically as the principle that S's desire gives S a reason (and a corresponding pro tanto ought) to see to it that p. The normativity in question here has its source in the desire, and is contingent on one's having that desire, so I wouldn't call it "categorically binding" in the usual sense. (I was making exactly the same claim as Andrew in his premiss 1, if that helps).

    So, in regard to your worry, no, I don't see Humeans as engaged in endorsing a mysterious, categorically binding kind of normativity. That doesn't mean they have to deny normative phenomena... they promise naturalistic explanations of such phenomena, perhaps even of the categorically binding kind. That promise is what draws me to Humeanism, and also to teleosemantic accounts of mental representation: the hope is that the psychological approach of Humean explanatory framework and the biological-evolutionary approach of teleosemantics can account for all the normative phenomena in the natural world.

    So, for instance, my proposal earlier was that the normativity attaching to a desire's direction of fit is what accounts for the truth of (DO). Teleosemantics (of the etiological kind) could explain that in terms of the proper function of the desire-producing mechanism, proper function itself being a normative feature the mechanism that can be explained in terms of ancestor mechanisms' contributions to the past fitness of ancestor organisms.

    That basic proposal needs to be refined in view of the fact that what we find psychologically salient or important can diverge from what has contributed to the fitness of our ancestors. So the refined proposal can be something along the lines of: it is the proper function of the desire-producing mechanism to produce mental representations that its consumer finds psychologically important to realize. Here past fitness explains why we have desires, or why we have mental representations that we find psychologically salient or important to realize in the actual world, and what I'm calling "psychological importance" (which is a matter of degree, including what is negligibly important at the lower bound) explains the 'ought' in DO.

    Sorry for going on for so long about my own proposal, but I was attempting an answer to your worry. Andrew's just above sounds interesting, and kind of similar to what I'm proposing, but it is too sketchy for me to tell.

  13. Interesting. The challenge here is to distinguish the normativity of your (DO) from that of categorical imperatives that are conditional on desires, e.g. "if you desire to torture people, you should seek mental help." A natural suggestion is that in the latter case, although the desire is a condition for the imperative's applicability, it isn't the source of the imperative's normative authority. (Most obviously, fulfilling the sadistic desire is not the aim or purpose of seeking mental help.)

    Even so, this suggests a more subtle challenge. Compare your (DO) to the following conditional:

    (SO) If the creature before you is sentient, you ought not to torture it.

    Just as the requirement in (DO) in some sense intrinsically stems from the antecedent desire, so it seems we could say that the requirement in (SO) intrinsically stems from the sentience of the creature in question. So why should the one be thought any more or less mysterious (or 'categorical') than the other?

    I guess this is where your story about 'directions of fit' and teleosemantics comes in. Although there's some sense of "grounding" by which we can say that the normative requirements DO and SO are each "grounded" in their respective antecedents, you will want to insist that there's a further (stronger, we might say 'meta-ethical' rather than merely 'normative-ethical') sense in which the Humean grounds normativity in desire.

    What further sense is this? Here my grasp of the proposal becomes a little slippery. One possibility is that the relation is meant to be reductive: perhaps for the Humean, to say that I ought to phi really means nothing over and above [insert some complex natural fact about my desires and/or evolutionary history here]. If that's the suggestion, then I'd worry that the proposal is better described as 'elimination' than 'reduction', at least given my understanding of the normative. (Compare Parfit's arguments that Williams lacked the concept of a normative reason.)

  14. I think the key difference between genuine hypothetical imperatives and mere material conditionals is that hypothetical imperatives have an implicit means component. It's not a hypothetical imperative unless the consequent is a good or the best means to accomplish the antecedent. Seeking psychiatric help is not a means to torturing children. Maybe hypothetical imperatives imply a material conditional, and they're explicitly stated as conditionals of some sort, but I think there's an implicit assumption of a means element to the claim, or you get something that seems like a category mistake. This also explains why "If you don't fire Larry, you should fire Moe" isn't a hypothetical imperative.

  15. Yeah, I think that's definitely part of the story. However, as noted in my intro paragraph, the means-connection doesn't seem sufficient to make a conditional 'ought' claim merely hypothetical, since we could imagine someone (very misguided) affirming that we categorically ought to take the means to whatever our desires may be. That is, they might affirm a conditional like "If you want to torture children, you should volunteer as a babysitter", and mean it in such a way as to allow for modus ponens.

    I think it's important to distinguish three possible interpretations of the quoted conditional:

    (1) Mere hypothetical imperative -- normativity is merely 'inherited', so doesn't allow for detachment by modus ponens.

    (2) Humean normativity -- where the normativity of the imperative somehow stems from the desire itself. Allows for detachment (at least on a 'pro tanto' interpretation of the 'ought'), but involves a somewhat deflationary or perhaps reductionistic interpretation of the normativity itself. (See my previous comment for some worries about making sense of this.)

    (3) Full-blown (e.g. 'non-naturalist realist') normativity, which just happens [as a matter of substantive content] to be requiring the agent to take the means to fulfilling their desires.

  16. Richard, thanks for the reference to Parfit, I will have to review his criticisms in some detail.

    Meanwhile, I'd like to make a first attempt (using ideas I already had) to respond to your worry, but the response turns out to be too long for the comment box (which has a word limit), and I'm not sure anyone would be willing to go through it all. So I will post it on my blog (linked from my profile name).


Visitors: check my comments policy first.
Non-Blogger users: If the comment form isn't working for you, email me your comment and I can post it on your behalf. (If your comment is too long, first try breaking it into two parts.)