What is a Categorical Imperative?

I’ve been struggling with the question, What feature must a requirement exhibit in order to qualify as a categorical imperative?

The first thought is that a categorical imperative is one that binds unconditionally. Thus, a requirement that p that binds only on the condition that q, is not a categorical imperative.

One problem with this is that whenever this is the case, we can readily produce a categorical imperative by introducing the condition into the requirement’s content. So, if the problem with the requirement that p is that it binds only on condition that q, then the requirement that q à p avoids that problem. Of course, it might be that the requirement that p binds also only on the condition that r, but we get around this by formulating the requirement that (q & r) à p. And so on.

This may lead to the suggestion that a requirement is categorical just in case it has non-conditional content. But such requirements may bind only conditionally.

So the natural thought is that a categorical imperative is a requirement with non-conditional content that binds non-conditionally.

There are both technical and non-technical problems with this. Technically, conditionals are equivalent to non-conditionals and vice versa. Thus, p à q is equivalent to ~p V q, and p is equivalent to (q V ~q) à p. Technical problems tend to have technical solutions that smarter people can figure out, however, so let me move to a non-technical problem. 

Take Kant’s own (and only) categorical imperative, in its second formulation: “always treat rational beings as ends rather than mere means.” On a literal reading, this requirement is too strong, as it requires that one seek rational beings so that one could treat them as ends, which is silly if not contradictory. Probably the spirit of the Kant’s categorical imperative is better captured by: “if and when dealing with rational beings, treat them as ends rather than mere means.” But that’s a conditional. It’s similar to another example sometimes given of a conditional categorical imperative: “if you’re a parent, treat your children well.” All conditionals. Yet we want to say these are categorical imperatives. So the suggestion under consideration returns the wrong results.

So maybe it’s OK for a categorical imperative to be conditional (either in content or in bind) – as long as it’s not conditional on one’s desires. This captures the intuitive idea that a categorical imperative is “one that each person has reason to follow no matter what her desires” (as Jamie Dreier puts it in his paper in the Cullity & Gaut book). Being a parent, or dealing with a rational being, is not a desire, but a state of affairs. So a requirement’s being conditional on their obtaining would not disqualify it from being a categorical imperative, on the present proposal. Thus the proposal returns the right results.

One problem is how to handle trivial or necessary desires, such as the desire to do what’s desirable, or perhaps the de dicto desire to do the right thing. In a way, every categorical imperative is conditional on one’s desire to do what’s desirable. But this shouldn’t disqualify it from being a categorical imperative.

The best way to handle necessary desires is to ignore them. We may simply modify the suggestion to read: a requirement is a categorical imperative just in case it’s not conditional on one’s contingent desires.

That’s my thought right now for how to construe categorical imperatives. I suspect there are more considerations that are relevant to this issue. There is, for starters, the whole business of desires for means v. desires for ultimate ends that needs to somehow feed into it. Difficult.

62 Replies to “What is a Categorical Imperative?

  1. Hi Uriah,
    I think the formulation you end up with is close to Kant’s own formulation. Given that for him the distinction is between hypothetical and categorical imperatives, and given that hypothetical imperatives are conditional upon something contingent about the agent (more on that in a sec), the contrast means that categorical imperatives’ bindingness is not conditional on anything contingent about the agent’s volitional makeup.
    On what the contingent thing is, desire is not an unreasonable way of thinking about it. Another useful way might be via Kant’s concept of “subjective ends.” That is, categorical imperatives are imperatives that bind independently of whatever subjective ends you have. There might, of course, be questions about this idea, but it seems to be consistent with the incorporation of “necessary desires,” if those can be construed as desires to promote/respect/etc. objective ends (e.g., for Kant, the value of humanity).
    Note that this doesn’t mean that CI’s can’t be conditional in another sense, e.g., if you have made a promise, keep it.

  2. Good questions, but I thought that “hypothetical” and “categorical” refer not to the circumstances under which the imperative holds, but the intrinsic bindingness of the imperative itself. A hypothetical imperative is only binding insofar as it is necessary for some independently binding end. A categorical imperative is binding in itself (for all rational beings). How does that sound?

  3. I’ve forgotten where I heard this example, but consider the following two “imperatives” (scare quotes because they’re actually assertions rather than imperatives):
    1. If you want to have sex with children, you should not work at a day care center.
    2. If you want to have sex with children, you should work at a day care center.
    (1) is a categorical, moral imperative. (2) is a hypothetical imperative. There’s some reading on which each of those is true–(1) is true in the sense that it would be morally bad for you to work in a day care center if the antecedent holds; (2) is true in the sense that working in a day care center would facilitate your hypothesized aim.
    What this suggests is that the difference between “hypothetical imperatives” and “categorical imperatives” is not in their logical form; it is in the intended sense of “should”. As a first pass, it seems that the categorical imperative uses a moral “should”, whereas the hypothetical imperative is non-normative and non-evaluative.

  4. Josh and Justin:
    Thanks for that – I definitely need to think about the means/ultimate-ends distinction as key to the hypothetical/categorical imperative one. One thought is that when requirements of means are conditional on the end for which these are means, whereas requirements of ultimate ends are not. This thought sends us back to the conditional/unconditional way of drawing the distinction, though.
    Mike:
    That’s a cool example, which reinforces the point of the first half of my post. On the face of it, though, you’d think there’d be hypothetical imperatives using the moral “should,” as in, say, “one should be polite to one’s neighbors.”

  5. I think Michael is right. The initial example of this sort was given by Hare in “Wanting: Some Pitfalls,” and I gave a version in Impartial Reason, something like: “If you want to kill someone in the most grisly possible way, you should use a cleaver,” vs. “If you want to kill someone in the most grisly possible way, you should see a psychiatrist.”
    The distinction is between a consistency constraint on ends and beliefs (a Broome-ish “normative constraint” or a Dancy-ish “wide-scope ought”), which does not detach, and an ought or normative reasons claim that does detach (see Impartial Reason, Broome’s “Normative Requirements” and Pat Greenspan’s 1975 JPhil paper, “Conditional Oughts and Hypothetical Imperatives”). I discuss many of these issues in my “Because I Want It.”

  6. This is interesting. Michael Huemer thinks that the comparison and contrast between (1) and (2) suggests that the difference between hypothetical and categorical imperatives is not to be found in their logical form. Steve Darwall thinks that Michael is right. Broome apparently agrees with Steve. But Broome thinks that the difference is to be found in the logical form.
    Maybe we can put it this way. Apparently (1) and (2) can both be true, even in the same context. But, the Law of Conditional Non-contradiction says that sentences like (1) and (2) contradict each other. The resolution must be either (a) that one of the two sentences does not really have the logical form that its English surface suggests, or else (b) that the word “should” is ambiguous. Broome (and I thought Darwall) say it’s (a). Huemer says (b).
    By the way, Conditional Non-contradiction is questionable, but I don’t think it could be the culprit here.
    My intuitions (and my broad philosophical commitments) lean toward (b), with Huemer. I’m against the Broome (and Darwall?) view that the standard hypothetical imperatives examples have misleading grammatical form. I hope to explain why in Bowling Green, and I hope to have a draft available within a couple of days.

  7. Sorry to have kicked up dust by responding too quickly. When I said that I agreed with Michael, I meant to be agreeing that the difference between hypothetical and categorical imperatives is not between conditional and unconditional imperatives, or between imperatives that are conditional on the agent’s ends, versus those that are conditional, if on anything, on things other than the agent’s ends. I didn’t mean to be saying that (1) and (2) have the same logical form. Thanks to Jamie for forcing me to clarify. As Jamie suspected, I side with Broome that (1) and (2) differ in their logical form. Let p=you want to have sex with children and q=you work at a day care center. Then, roughly:
    (1) If p, then Oq
    (2) O(If p, then q)
    I don’t agree with Michael that 2 is a nonnormative claim about what will facilitate one’s end.
    Thanks again to Jamie and I eagerly await his Bowling Green paper.

  8. “As Jamie suspected, I side with Broome that (1) and (2) differ in their logical form. Let p=you want to have sex with children and q=you work at a day care center. Then, roughly:”
    (1) If p, then Oq
    (2) O(If p, then ~q)
    This seems like a natural way to go, since the operator in (2) is clearly not conditional in the usual O(-/p) form. But I don’t think that (2) can be very close to what you want to say (maybe this is why you say “roughly”).
    Let r = you prey on children. Since N((p & r & ~q) -> (p -> ~q)), the obligation in (2) is fulfilled by all who prey on children outside of daycare. But, needless to say, that hardly fulfills any moral obligation. So I’m less sure that (2) captures the categorical imperative in question.
    On the other hand, I admit that insisting that someone solve every deontic paradox in a post or two is setting the bar a bit high (on yet another hand, Castaneda once claimed to have done that in one fell swoop!).

  9. There is an unoccupied alternative here that seems not unattractive, viz., that both of Michael’s (1) and (2) share the logical form of (2) in Steve’s post, but that Jamie is nevertheless right that ‘should’ is ambiguous. Both oughts might not detach, that is.

  10. Ummmmm.
    Steve seems to have forgotten a negation sign. Michael Almeida has graciously supplied one, but I am sure the result is not what Steve intended.
    I think Steve’s view is that the Huemer conditionals have the following forms:
    (1) p ⊃ O~q
    (2) O(p ⊃ q)

  11. I know that I’ll be eternally in a very small majority about this (probably including only me), but my strange intuition is that (1) and (2) share the same logical form (If p, then ought to q (or not-q)) and ought is used unambiquously in the robust sense. It’s just that (1) is true, and (2) is false. I guess this ultimately comes down to saying that there are no mere hypothetical imperatives – unless they are derived from categorical imperatives that carry the normative oomph. I’m still working on the arguments (the book will be out 2015).

  12. Jamie,
    Now I’m a little lost. According to Mike H. (1) is the categorical or moral ought and (2) is the hypothetical ought. He says,
    “1. If you want to have sex with children, you should not work at a day care center.
    2. If you want to have sex with children, you should work at a day care center. (1) is a categorical, moral imperative. (2) is a hypothetical imperative”
    You say that these should be represented in (1) and (2) below (acc. to Steven D.)
    (1) p ⊃ O~q
    (2) O(p ⊃ q)
    But that’s very strange. The categorical ought is in (1)? It couldn’t look much more hypothetical. And the hypothetical ought is the one in (2) that is not governed by factual detachment? I can’t see the motivation for these representations under these names.

  13. I think people use the notion of categorical imperative in different ways, perhaps thinking they come to the same thing. People sometimes say that an imperative iff it is not contingent on the agent’s subjective motivational set and other times they say that an imperative is categorical iff it authoritatively applies to all rational agents as such. But neo-Aristotelians have it that it is an agent’s species or life-form that is central to her reasons–they thus think there are imperatives that are not dependent on an agent’s desires but also do not apply to all rational agents as such. Are these categorical imperatives? In one of the above two senses yes and in the other no. I believe Kant uses both of these formulation, thus we might not be able to defer to him on this question.
    I myself think we should stick with the “authoritatively applies to all rational agents as such” version, but I suppose I mainly think we need to note the difference between the two versions.

  14. Jussi, that does seem like an odd view. I guess you disagree with Kai and Sabine about what to do if you want to get to Sugar Hill in Harlem.
    Mike A., the Darwall/Broome/Dancy view does imply that ‘hypothetical imperatives’ are misnamed, I guess.
    Dave, I never realized your surname was so hard to spell.

  15. Here’s another proposal.
    Let there be, as is common in deontic logics, a dyadic operator,
    O(p/q)
    to be read ‘it ought to be that p, given that q’, such that O(p/q) is true iff all the best q-worlds are p-worlds. We may then define a monadic operator O(p) by
    O(p) < -> O(p/T)
    where T is a tautology. So O(p) is true iff all the best worlds are p-worlds.
    Now, we might say that a hypothetical imperative is one given by a sentence of the form O(p/q) for some contingent q, and a categorical imperative is one given by a sentence of the form O(p).
    On this proposal, Uriah’s strategy for converting hypothetical imperatives into categorical ones won’t work, because O(p/q) is not equivalent to O(q⊃p). To see this, let q = ‘you are given poison’ and p = ‘you are given the antidote’. Plausibly, you ought not to be poisoned, and so we have O(~q). But this implies O(q⊃~p). But it does not follow that O(~p/q), i.e. you ought not to be given the antidote, given that you are poisoned.

  16. Here’s how Kant himself defines hypothetical and categorical imperatives in the Groundwork:
    ‘Now, all imperatives command either hypothetically or categorically. The former represent the practical necessity of a possible action as a means to achieving something else that one wills (or that it is at least possible for one to will). The categorical imperative would be that which represented an action as objectively necessary of itself, without reference to another end’.
    I think Mike A. is right – neither (1) or (2) would qualify as categorical imperatives. Rather the relevant CI would be ‘Do not put yourself in a position where you might be tempted to have sex with children’.
    If I don’t remember wrong, Kant actually discusses the Aristotelian position in this respect. I think he thought that the Aristotelian ‘quasi-categorical’ reasons would actually be hypothetical – conditional on having the end of human flourishing. That I think he thought to be a natural, unavoidable end for all humans, so in the sense these hypothetical reasons were commands of reason for all humans. But the problem for him I think was that the notion of human flourishing was just too vague in order to give us substantial requirements of reason beyond the ones we already get from the categorical requirements which exist for all rational beings.

  17. Jamie,
    not really. I can see that people have many motivation independent reasons for going to Harlem. In that case if they want to go there, they ought to. But that ought only has the force their reasons for going there have. If they had no such reasons, then there would not be anything they ought to do to go to Harlem. I know, this is still quite confusing, but I’m working on it.

  18. Campbell, I think there has to be some accessibility relation built in, too. Otherwise after you have been poisoned, and we all know it, it will not be the case that we ought to inconvenience ourselves to pour some of the antidote into your mouth (because in the best worlds there would be no point in doing such a thing, since in those worlds you are not poisoned).
    I tried to link to Kai von Fintel’s paper in the comment above, but unfortunately I linked to a collection of powerpoint slides or something like that. But anyway, it’s surprisingly difficult to make those deontic conditionals — the anankastic ones — come out right.
    Jussi, I see the motivation for the view; I was just pointing out that it is obviously false. You’re welcome. (Although sadly, the actual lyrics to “A Train” do not include any anankastic conditional.)

  19. Jussi,
    As I understand modern neo-Aristotelians such as Hursthouse and Foot, they would say that our species membership determines our reasons whether or not we have as a necessary or contingent end of ours to further the flourishing of our species. Thus I think Kant’s own distinction between hypothetical and categorical imperatives is not exhaustive.

  20. Jamie,
    Don’t look at me, I don’t know who that moron Sobeld is? Sure, he has the same e-mail address as me, but still…

  21. Thanks Jamie. You just might have saved me years of redundant work… : )
    David,
    good point. Kant probably hadn’t read Foot or Hursthouse. But, I take it that Foot and Hursthouse would not take the species related reasons to be completely independent of the end of human flourishing. I thought that in their view it is the notion of human flourishing that is supposed to do the work of specifying the content of those reasons. It may be that all humans do not have that end in their motivational set and that having such an end is not even required for being a human (even though lacking it might constitute a deficiency of some sort). Anyway, I think Kant would argue against them that the notion of human flourishing is too vague for doing much work in specifying the content of our reasons.

  22. Thanks Jamie,
    The name might be the least of the worries here. One of the painfully unfortunate features of conditionals like p ⊃ O~q (whatever we decide to call them!) is that strenghtening antecedents is valid for them. So, the inference from (p ⊃ O~q) to (p & r) ⊃ O~q is valid. Maybe this is where the unconditionality is supposed to come in. But then we won’t find many categorical or moral oughts. In this case let r = but you never would have sex with children or so much as bother a child and you would give each child 1 mil. for his education, etc. With sufficient imagination we can certain find conditions that would defeat this obligation (and almost any other). So again (1) does not seem to express a categorical or moral ought.

  23. Jussi,
    My point was not that Foot and Hursthouse are correct, but rather they occupy a position that is awkward to classify using Kant’s distinction between hypothetical and categorical imperatives (esp. the passage you quote). To successfully classify such views, we must note the ambiguity in his notion of a categorical imperative that I mentioned above.

  24. Mike,
    Hey, it’s not my view. I’m just telling you what Steve Darwall meant to be saying about Michael Humer’s conditionals.
    Still, to be fair, the Broome/Dancy/Darwall view is mainly concerned with how to understand the other conditional, (2). Maybe the connective in (1) is some non-monotonic conditional. Maybe it’s just an English indicative conditional, whatever that is. I don’t think that would bother them.

  25. “. . . But it does not follow that
    O(~p/q), i.e. you ought not to be given the antidote, given that you are poisoned.”
    Why think it is false that the best worlds in which you are poisoned are worlds in which you do not receive the antidote? The conditional O(~p/q) might well be true. The best worlds in which I am poisoned might be those in which I am resistant to all poisons. I need no antidotes. In any case, there’s no doubt that those conditionals are not equivalent, as you note. Indeed, I think it’s more or less hopeless to try to formulate conditional obligation with the horseshoe (and it ain’t much more hopeful with the primitive dyadic operator).

  26. David,
    I didn’t mean to argue against you – didn’t think you were thinking that they were right. And, I see your point. I’m not sure though that Kant was using CI ambiguously. I think he would also have had another strategy for explaining away the Foot/Hursthouse species related categorical reasons without making them a distinct category. Many of them he could have made derivative of the reasons for all rational beings. All rational beings must treat their own rationality as an end. Part of that involves maintaining the natural being one happens to be as a rational being – your end after all is to keep being a rational being. This creates some species related subtstantial reasons for humans that are still ultimately categorical. Many of these would be the kind of reasons Foot for instance talks about. If you are rational being of some other species, then you need to do different things to live as that kind of a being. So, we do get some kind of species relative categorical reasons within the Kantian distinction. I think this version of Kant would come close to Scanlon’s parametric universalism whereas the neo-aristotelianism as such often sounds more like a version of bening relativism.

  27. Jussi, David, et al:
    On the Kant/neo-Aristotelian question: I doubt Kant would say that our species nature doesn’t provide us reasons. I think his position is that it’s effectively inevitable that we will pursue our species-specific wants, and hence that any imperative grounded on such reasons can’t be the source of a moral duty and cannot provide us a categorical imperative, i.e., an imperative that applies to rational beings just insofar as they are rational. A quote from the Metaphysics of Morals:
    “For his own happiness is an end that every human being has (by virtue of the impulses of his nature), but this end can never without self-contradiction be regarded as a duty. What everyone already wants unavoidably, of his own accord, does not come under the concept of duty, which is constraint to an end adopted reluctantly. Hence it is self-contradictory to say that he is under obligation to promote his own happiness with all his powers.” (MM 6:386)
    So it’s not just the vagueness of the notion of human flourishing that leads Kant to conclude that it can’t be the basis of any categorical imperatives. His point seems to have to do with the normative logic of duty: that one can only have a duty to do that which it is possible but not inevitable that one will do.
    In light of such passages, I’ve often thought that Kant denied that there is such a thing as *the* hypothetical imperative, i.e., the command to pursue one’s subjective ends. This would be consistent with there being hypothetical imperatives, specific commands linked to our subjective, individual, variable ends.

  28. Michael,
    that’s really interesting. I’ve thought about that claim before as it’s been puzzling me. There is something odd about it. It seems empirically false. There’s just so many people who act self-desctructively. I actually would be quite happy if we could get a categorical imperetive for them from his other commitments.

  29. Mike C,
    This passage in MM is consistent with Kant’s further claims that we have both a duty to promote the happiness of others and a duty to promote our own happiness. See (MM 388, p. 152 ff. in the Gregor volume). It is just that we do not have a duty to promote these for their own sake. With respect to our own happiness, we have a duty to promote it, Kant says, as a necessary means to perfecting ourselves. For dependent beings such as we, happiness and some level of well-being are a necessary conditions of virtue.

  30. Regarding Huemer’s sentences 1 and 2, I agree with his diagnosis: on their most charitible interpretations, they involve differenct senses of ‘should’. In particular, ‘should’ is normative in 1 but not in 2. The most plausible explication of 2, I think, is something like: you’re more likely to have sex with children if you work at a daycare centre.
    But then this seems, to me at least, an odd way to illustrate Kant’s distinction between categorical and hypothetical imperatives. On my understanding of the distinction (and I should confess I don’t know Kant very well), both kinds of imperatives are normative, but the hypothetical ones are conditional in a way that the categorical ones are not. Indeed, if 2 is interpreted as non-normative, it doesn’t seem to be an ‘imperative’ (or appropriately related to an imperative) at all.
    Anyway, setting aside Kant exegesis, the important point is that there are two distinctions in play here: one between normative and non-normative, the other between conditional and unconditional.
    (By the way, I worry that if we keep talking about having sex with children, Google will be sending some very dissapointed visitors our way.)

  31. Campbell,
    I think you are on the right lines. I quite like the Kolodnian way of putting it that the hypothetical, non-normative ought is drawing attention of the agent on what she ought to do by her own lights (this is supposed to be an alternative to the wide scope interpretation). This would make the understanding of oughts ‘imperatives’ still in a way – commands on what they agent ought to do if she let her own willings guide her. Yet, the advice remains non-normative. We do not necessarily say that the agent ought to do what she ought to do by her own lights. Therefore, saying this does not commit us to saying what the agent ought to do, full stop, in the normative categorical sense.
    Sorry about excessive commenting tonight. Should be working on a reply to a paper on Brandom but cannot quite get myself to it yet.

  32. Apologies again and thanks, again, to Jamie for figuring out what I was trying to say. (I won’t waste your time explaining the mistake.) As to whether the term “hypothetical imperative” is appropriate or not, there is a reading on which it is exactly right. Suppose we have the hypothetical imperative, “if your end is A, then you should take (necessary means) B.” One way to turn a hypothetical imperative into a conditional imperative (i.e., something of the form “if A, then OB”) is to substitute for the antecedent “A is your end,” something you are rationally committed to in being in that state of mind, viz., something like “A is to be brought about” or “A should be your end.” This then yields the legitimate conditional imperative: “If A is to be brought about or should be your end, then you should take (necessary means) B.” We might say that “A is to be done” or “A should be your end,” is a “hypothesis” that you implicitly accept, or should accept, in having A as your end.
    Similarly, suppose that p entails q. Then if you should believe that p, then you should believe that q. In believing p, you accept the hypothesis that p, and since q follows from that hypothesis, you should believe q. So we say, “If you believe p, you should believe q.” We don’t mean that you should believe q conditional on your believing p. We mean that you should believe q conditional on what you accept in believing p, namely, p.

  33. I wasn’t clear about what I think about my sentences (1) and (2)–

    1. If you want to have sex with children, you should not work at a day care center.

    2. If you want to have sex with children, you should work at a day care center.

    The most natural reading of my post is that I think that (1) and (2) have the same logical form, and everything else is the same, except that “should” has a different sense in (2) than in (1). But I don’t think that; actually, I think the “should” in (2) isn’t to be taken seriously at all. The best way of explaining the intended meaning of (2) is to paraphrase the entire sentence into a descriptive statement that doesn’t contain any “if-then”s or “should”s:

    3. Pedophilia would be facilitated by working at a day care center.

    (2) is just an idiomatic way of saying that. Unlike (1), (2) is not really conditional, or imperative, or normative.
    Compare a statement like

    4. If you want some beer, there’s some in the refrigerator.

    The intended meaning of this isn’t really conditional. (4) would not be made true, for instance, by your failing to want some beer (nor does it say that there’s some sort of connection between your wanting beer and its being in the refrigerator). Really it just means, “There’s some beer in the refrigerator (which might be relevant to you if you want some).”
    Now here’s an argument: I think it’s generally agreed that (3) entails (2). Obviously, this entailment is explained by my view. But I see no way of explaining it on other views according to which the meaning of (2) involves a real, normative or evaluative “should”. Thus, for instance, (3) obviously doesn’t entail or even provide good evidence for the claim that the best worlds in which you want sex with children are worlds in which you work for day care. It also doesn’t support the claim that the best worlds are ones in which (if you want sex with chilren, you work at day care). (Though the best worlds are in fact ones in which that conditional holds, because the antecedent fails, this doesn’t follow from (3).) Nothing at all follows from (3) about what the best worlds are like, or in general what the correct value system is.
    In general, I can’t think of any natural interpretation of (2) such that (2) follows from (3) and (2) contains a real “should” (roughly, one that is normative or evaluative).

  34. In response to Steve Darwall’s last post: I like the interpretation “If you should aim at X, then you should do Y” better than the O(x -> y) interpretation. But consider this example:
    5. If you should avoid stealing, then you should also avoid telling your little brother to steal.
    I think this is true (I’m recalling Geach’s observation that if it’s wrong to do something, it’s also wrong to get your little brother to do it), but I don’t think it entails
    6. If you want to avoid stealing, you should avoid telling your little brother to steal.
    where this is read as a conditional imperative along the lines of “If you want to get to Harlem, you should take the A train.”

  35. Jussi,
    “It seems empirically false. There’s just so many people who act self-desctructively.”
    Yes, but Kant “deals” with this by claiming that self-destruction is itself motivated by an inclination, since it can’t, arguendo, be motivated by respect for one’s nature as a rational being. And since (for Kant) there are only two sources of motivation, inclination and respect for law, self-destuctive behavior belongs in the former category. Even if Kant is right that there can be no duty (or at least no direct duty — nod to Mike Almeida) to do what we will inevitably do, I think the problem here is Kant’s somewhat simplistic bivalent psychology, one that fails to recognize the possibility that self-destructive behavior might fall into neither category. (Or so I’ve argued in my as yet unpublished work!)

  36. Michael,
    I hear your (3) “If you want to have sex with children, you should work at a day care center.” as saying that the truth of the antecedent provides some reason to make the consequent true.
    I think we could understand (3) to be short for:
    a) wanting to have sex with children is a reason to have sex with children.
    b) working at a day care center facilitates having sex with children.
    c) therefore if one wants to have sex with children, one has a reason to work at the day care center.
    I would say that (b) above is just a tacit premise of the statement to the effect that working at the day care center is a means to the end of having sex with children.
    I would contrast this with your (4) “If you want some beer, there is some in the fridge.” I hear this as not literally claiming that wanting beer is a reason to get some (although perhaps in context it is perhaps often best taken that way). This seems to me the difference between pointing out the means to something an agent wants and saying that they ought to take the means.

  37. Sorry, I got the numbers wrong in the above. I meant “2” where I used “3”, but likely this was obvious.
    But I meant part of the upshot of my post to be that it does not seem obvious that your (real)
    (3) “Pedophilia would be facilitated by working at a day care center”
    entails your
    (2)”If you want to have sex with children, you should work at a day care center”
    since a person could well think that 3 is true while thinking this provides the agent who wants to have sex with kids with no reasons to work at the day care center. Thus perhaps it is only clear that 3 entails 2 after someone has accepted your (to my ear unnatural) reading of 2.

  38. Michael H., you say that (2) is just an idiomatic way of saying what (3) says. I’m sure that’s not true.
    First, we use a similar construction in English with genuine imperatives instead of oughts. We use one with must, with have to, and can. And we import novel ways of expressing the same modalities:

    It’s gotta be rock-n-roll music if you wanna dance with me.

    Second, the same construction occurs in many other languages. I haven’t done an inventory, but you can find it in some languages not closely related to English. For example, I have it on good authority that the same construction occurs in Basque. If (2) (along with its structural relatives) was an idiom, this would probably be the most spectacular coincidence I’ve ever heard of.
    There is definitely something ‘funny’ about the conditional, but I don’t see how it could be an idiom.

  39. Thanks everybody for the input.
    Suppose (as is plausible) that there are two “should”s, a normative and a non-normative. All statements featuring the non-normative “should” are not CIs. But not all statements featuring the normative “should” are CIs. Some are HIs. So the following proposal would return the wrong results:
    (*) A requirement R is a CI iff R features the normative “should.”
    That can’t be right.
    Michael H, are you proposing this?:
    (**) A requirement R is a CI iff R is not entailed by a statement that features no “should.”
    This is kinda cool. Whether it’s also true I’ll let you and David and Jamie figure out while I sit back and press “refresh” every few minutes.
    Another proposal, which does not strike me as prima facie plausible, would be:
    (***) A requirement R is a CI iff R features the categorical “should.”
    This would involve there being a distinct sense of “Should” reserved to CIs.

  40. Not sure what ‘entails’ means in Huemer and Kriegel’s sense. Is it conceptual? It’s analytic that if A then B?
    I do think it’s obvious that if (3) is true, (2) is. But I might have to go with Dave S. on the question of analyticity. These are very difficult judgments to make, I find.
    In any case, Uriah, (**) strikes me as hopeless, using whichever sort of entailment you want. Arthur Prior disillusioned me. Gillian Russell at Wash. U. has a paper that explains the problem. I mean, she has it on her web site.

  41. A couple of thoughts re sorting out the difference between CI and HI. First, one reason not to like normative principles of the form
    x–>Oy
    and prefer principles of the form
    O(x–>y)
    is that the former appear to derive an ought from an is. (OK, obvious to most of you, and fixable with higher-order principles, but still worth pointing out. Some don’t mind this, perhaps, but applies equally to moral and nonmoral oughts, too. It also leads to problems, but they may be preferrable to the ought-is problem.)
    Another thing: principles of the second kind are used by at least Broome (I don’t know if Steve agrees) to rule out pairings of mental states, not pairing of desire and act. That is, ‘y’ refers to an intention, not an act. ‘Rational’ seems to be a modifier that belongs just to the mental realm. May not be a critical point, but worth keeping in mind in discussing what principles of rationality cover, and, in particular, whether such principles ever generate reasons for action. According to many wide-scopers, these principles concern only intentions.
    Further thing. If you prefer that principles of rationality that concern the relations of desires and intentions to be principles that bind and rule them out according to wide-scope (as I do), that does not imply that you cannot also hold there are intentions that are forbidden tout court, as in
    You ought not intend to have sex with children.
    Last thing. Why can’t we say that the difference between the CI and HI is just that the former can rule out intentions tout court in this way. The ‘ought’ here does not on its face apply to a conditional, though (obviously) compatible with principles that do. You could hold that it is not derived anyway from ‘You ought not desire to have sex with chidren’ plus a conditional connecting them. One theory that might hold this is an intuitionistic theory. But you can (well, ok, there are *other* objections) derive that from principles from the well-known alternatives. In any case, one could begin from this mundane point (that CIs alone rule out intentions to act tout court).

  42. BTW, the last suggestion is not to deny that you can reformulate any obligation as a conditional (since you can). It’s just to deny that a conditional is hidden somehow in the logical structure of the claim, one requiring reformulation to to make sense of the ‘ought’. It seems necessary to cash out prudential oughts in terms of conditionals that invoke desires to make sense of the bindingness, but (according to those who think there are genuine CIs) not necessary for moral oughts. Not necessarily the Pritchard view, that it is somehow a mistake to do so. But the simple view seems right, that CI ‘oughts’ apply directly to intentions to act, no need to reformulate to make sense, and that is the difference.

  43. We often use ‘should’ to make non-normative claims about probabilities. For example, we sometimes say,
    (1) It should rain,
    meaning that the probability that it will rain is greater than 1/2. Similarly, we might say
    (2) If you want to have sex with children, you should work at a daycare centre,
    meaning that the probability that you will have sex with children given that you work at a daycare centre is greater than 1/2.
    I take this to be broadly sympathetic to the position that Huemer defends above, except I think there is a sense in which (2) is conditional because it reports a conditional probability.

  44. Let me add, I think (2) is ambiguous, because ‘should’ is ambiguous between a normative and a non-normative (probabilistic) sense. But the non-normative reading is more charitible because it makes (2) true, or at least plausible.

  45. Wow, what a great thread. Here’s my $0.02:
    (1) and (2) encapsulate two different kinds of practical reasoning. Both encapsulate arguments for actions (or intentions, if you like): that one [not] work at a day care center. The first encapsulates an argument whose premise is a fact:
    Fact: I want to have sex with children,
    So, Intention: I do not work at a day care center.
    The second encapsulates some means-end reasoning:
    End intention: I have sex with children
    So, means intention: I work at a day care center.
    We use conditionals to encapsulate practical arguments just like we do for theoretical arguments. Unfortunately, there’s no good way to put the content of an intention into language without making it sound like a report of the intention. Thus we get conditionals of the same syntactic form with different logical forms. The reason that “I want to have sex with children” detaches from (1) is that in the reasoning it encapsulates, “I want” is mentioned in the premise. The reason it does not detach in (2) is that in the reasoning it encapsulates, “I want” is not mentioned in the premise.
    OK, here’s my third cent. I think we can relieve ourselves of any inclination to posit two different senses of ‘should’ by coming up with persuasive arguments of the form A -> O(B -> C), A, so O(B -> C). The same “ought” will have the detaching function in the premise but the non-detaching function in the conclusion; if the argument is valid then it can’t change sense. Here’s a suggested argument:
    A. If you’re eligible to work in a day care center, then if you want to have sex with children you should work at a day care center.
    B. You’re eligible to work at a day care center.
    C. So, If you want to have sex with children you should work in a day care center.
    This seems as valid as any argument could be to me; but an argument that continued, “D. You want to have sex with children. E. So, You should work in a day care center” would not be valid. I conclude that the “should” has only one sense.

  46. Robert, just btw (may not be all that important to your point) I’m not sure there is an Is-Ought problem or, rather, as I think Jamie was hinting at (in response to Uriah above), Arthur Prior has shown pretty convincingly that there ain’t much trouble in getting an ought from an is. For my money, it’s done even more convincingly in David Johnson’s recent _Truth without Paradox_ chp. 4 (R&L, 2004)

  47. This seems as valid as any argument could be to me; … I conclude that the “should” has only one sense.

    How does that follow? Would the argument be invalid if ‘should’ had more than one sense?

  48. I mistakenly posted this on a different thread:
    Mike,
    Sure, the Prior maneuver shows that it’s easy to get an ought from an is. I suspect (can’t prove it…ok, maybe for now just have faith?) that these really are only trivial cases, and that some general principle can be found true in some suitably restricted cases (no oughts from a set of completely non-ought statements). I just hink there’s *something* to St. David’s principle, (though I’ll have to just rest it on prejudice at the moment).

  49. Campbell,
    I agree that we sometimes use “should” in a predictive sense, but it seems implausible to me that that is what is going on with 2. I would say that it is clear that in the intended sense, many other shoulds would follow–e.g. if you want to have sex with children you should work at an elementary school. But it can’t be likely that people that satisfy the antecedent are likely to work at both day care centers and elementary schools (indeed the former empirical claim is surely wildly implausible). Surely the only natural reading of the conditional is that working at a day care center facilitates having sex with children, and not that most people that want to have sex with children work at day care centers.

  50. David,
    I stated my position poorly above. What I want to suggest is (a) that some sentences of the form If you want it to be the case that p, you should make it the case that q mean only that Pr(p/q) > Pr(p/~q), and (b) that sentence (2) is most charitably interpreted in this way. Perhaps this commits me to saying that the proportion of daycare workers that have sex with children is greater than the proportion of non-daycare-workers that do. But that does not imply that ‘most people that want to have sex with children work at day care centers’.

  51. Campbell,
    It seems to me that that reformulation costs you the examples that motivated the proposal. What, on the new formulation, does “It should rain” mean? It would have to be of the form, given BLANK, it is more likely that it rain than given non-BLANK, right? But the most familiar formulations where we use “should” predictively (It should rain. The sun should be rising soon. Suzi should get an A.) seem best understood along the lines of your earlier proposal.

  52. I’m with Robert Johnson on is-ought and A.N. Prior. There’s a great article by Toomas Karmo in Mind (1988, 252-7), “Some Valid (but no Sound) Arguments Trivially Span the Is-Ought Gap”, that provides a proof of a useful version of the is-ought gap. I’m going to post about it in a separate thread, since it’s a sufficiently distinct topic.

  53. Jamie, a note about “idiom”: “idiom” might be the wrong word for what I intend. I don’t mean that there’s no connection at all between the meaning of “if you want x, you should A” and the meanings of its component terms, just that the meaning of the sentence isn’t what one would expect from putting together those terms in the most straightforward way, or what one would expect from comparing it to sentences like “If you drop the glass, it will break.” I don’t think that it’s a complete coincidence that other languages have equivalents of “if you want x, you should do A”.
    Again, compare statements like, “There’s some ice cream in the freezer, if you want some.” The speaker knows there’s some ice cream in the freezer whether or not the addressee wants some. (In contrast, the utterer of “If you drop the glass it will break” doesn’t think the glass will break whether or not you drop it.) So the intended meaning of the sentence isn’t what one would expect from comparing it to sentences like “If you drop the glass it will break.” But of course the meaning of “There’s ice cream in the freezer if you want some” isn’t just totally unrelated to the meanings of its components, nor would it be a total coincidence to find that other languages have similar sentences.
    Also notice that you could make the point that there are lots of similar ice cream sentences that use different words: “There’s some ice cream in the freezer, if you’re hungry.” “There’s some ice cream in the freezer, in case you were wondering.” But hopefully this doesn’t make you want to say that these sentences really are conditionals in the normal sense, to be understood on analogy to “The glass will break if you drop it.”

  54. Right, I certainly agree (and said in the earlier comment) that there is something funny about the instrumental ‘ought’ conditionals. I just think it must be some kind of generalization from ‘ordinary’ conditionals. (And the ‘biscuit’ conditionals, the “If you’re hungry” ones, must also be generalizations from ordinary ones, for the same reason, I say.) So I’m skeptical that they mean nothing more than the ‘facilitation’ facts, as you were saying. If you (or someone) can explain how those facilitation propositions can be constructed from the conditional sentence’s parts by some kind of generalization from the semantics for ordinary conditionals, that would be a lot more convincing.
    Hm, to make that paragraph clearer I would have to make it a lot longer and bring examples and claims down from earlier comments, but I assume that (a) practically nobody is still reading, and (b) those who are either remember the examples and claims or are willing to scroll up and find them.

  55. I’m very late weighing in here, but the stuff I’m working on intersects here in lots of ways. My suggestions here reprise those I made in (my only) two earlier posts:
    http://peasoup.typepad.com/peasoup/2005/12/semantics_of_ou.html#more
    http://peasoup.typepad.com/peasoup/2005/12/pragmatics_of_o.html#more
    First, I agree with Mike Huemer that hypothetical imperatives are not genuinely conditionals. “If you want… then you ought…” is more perspicuously translated as “In order that…you ought…”. “In order that…” functions to qualify the “ought” rather than present conditions for it, which explains why the “ought” is nondetachable.
    Second, on the possibility of giving a unified semantics of the “ought” in categorical and hypothetical imperatives: all the suggestions floated so far, I think, try to reduce to hypothetical “ought” to the categorical “ought” (e.g. If you ought to pursue the end, then you ought to pursue the means/ You ought to make it the case that (if you pursue the end then you pursue the means). I’m disposed to try the reduction the other way around – it’s the categorical “ought” that I find most suspect. The proposal is this: a “categorical imperative” is a categorical USE of “ought”, which consists in rhetorically presupposing the relevant end (or engaging in a practice parasitic upon this rhetorical use). “You ought not to have sex with children” is then elliptical for something like “In order that you do not harm children, you ought not to have sex with them”, where the end of not harming children is contextually presupposed and conversationally demanded. This captures an expressivist dimension to categorical judgement. Jamie expresses skepticism that (hypothetical) “ought” statements mean nothing but the “facilitation facts” – I’m saying that this is all the semantic content of both hypothetical and categorical imperatives, but that we communicate more by our use of them, especially categorical imperatives.
    Finally, in my previous post I argued (more radically than Campbell, who was as I recall the only sympathetic party) that ALL oughts were “predictive”. First, note that at least in the “hypothetical” cases, it is difficult to distinguish the terms for deontic possibility (might, may, can) and necessity (must, have to) from the modal versions: “In order that you do not harm children, you MUST not have sex with them”. My hypothesis is therefore that deontic terms are just modal terms under an “in order that…” The predictive and (hypothetical) deontic oughts have at least this much in common: they are both probabilistic. What ought (predictively) to happen is what is most likely to happen; the means one ought to take is (all else being equal) the means that is most likely to bring about the end. There are still some problems I have to solve before I can confidently assert that I have a single semantics for both uses, but I’m optimistic.
    Anyway, these are some of the main ideas I’m trying to work out in a paper I’m working on. If anyone’s interested, I’d be happy to share.

  56. Steve,
    So how would you understand “You ought not to harm children”? Would that be something like, “In order that you not harm children, you ought not to harm children”, with the goal of not harming children presupposed? In general, how would you account for judgements of intrinsic wrongness?
    Also, what is it to presuppose an end? (Not, I take it, to assume that the end ought to be pursued.) Usually, I think of propositions as the sort of things that can be presupposed.

  57. Mike,
    In judgements of intrinsic wrongness, I suggest the semantic content is the trivial “In order that p, it ought to be the case that p”, as you speculate. The idea is that since the categorically demanding element is pragmatic rather than semantic, uttering such trivialities still has point and force.
    By “contextually presuppose” I meant leaving the “in order that p” implicit, while by “rhetorically presuppose” I meant behaving as if there was a shared commitment to the end.

  58. Steve,
    could you unpack this:
    ‘The idea is that since the categorically demanding element is pragmatic rather than semantic, uttering such trivialities still has point and force.’
    a little. What is the point of uttering such trivialities? Why is that point, whatever it is, rather implicated and not part of the semantics of the claim?

  59. Jussi,
    Sure thing. The main function of uttering categorical imperatives, according to my view, is expressing a demand that certain ends are honored. Suppose the semantic content of my utterance is “In order that children are not harmed, you ought not to have sex with them”. The “facilitation fact” (following Jamie) being asserted is here itself informationally fairly trivial (although not logically so) — but by uttering it categorically, I express the demands (a) that my audience respects/values the end of children not being harmed, and (b) that my audience not have sex with children. (I accomplish this by the following mechanism: by leaving the end implicit, I behave as if it were uncontroversial and shared. Because it evidently is not, I rhetorically communicate by my behaviour my demand that it be so).
    Now given this picture, making moral assertions that are semantically trivial, like “[in order that you don’t harm children], you ought not harm children”, has practically as much conversational point as making nontrivial assertions.
    Is this clearer?

  60. That’s much clearer, thanks. I guess that I’m again left wondering why would one be pushed to give the predictive account in relation to the assumed end as the linguistic meaning of the categorical claim and leave the demands for sharing the assumed end and complying with this end in action as the implications to be dealt with pragmatics. Why not, for instance, just say that by making categorical claims we make such demands, and that that is what the meaning of such claims is constituted of?
    But, I think we’ve been though this before, and so maybe I’ll look up the earlier thread.

  61. Jussi,
    Good question, and I don’t know that you’ll find any answer to it in the earlier thread.
    Here are a few motivations for this view:
    (1) systematicity of meaning: it is (I claim) able to identify a common semantics for a wide range of uses of “ought”.
    (2) Intuitive appeal: to me (but perhaps to no-one else) it just seems right.
    (3) to identify the expression of a demand as the semantic content of “ought” seems to commit the speech-act fallacy. When “ought” is embedded, it doesn’t express such any demands. So it seems we need an account of the semantics that doesn’t commit the fallacy, but is capable of explaining how we use “ought” to express demands. This account can.
    (4) It is nonmysterious and amenable to naturalism.
    (5) Your proposal struggles to accommodate the amoralist; mine (I claim) does not.

Comments are closed.