For my first post, here’s something on the semantics of ‘ought’ that I’ve been working on. I’m pretty excited about it, so maybe someone can burst my bubble.

It’s striking that the deontic terms (‘ought’, ‘must’, ‘have to’, ‘might’, ‘may’, etc.) are also modal terms. My idea is that the deontic terms are simply the modal terms occurring in a particular propositional context: under an ‘in order that…’ (or ‘in order to…’) operator.

Consider

(1) In order that he evades arrest, Max must mingle with the crowd.

Is the ‘must’ here deontic or modal? I think it’s both. Supposing we can analyze the modal ‘must’ as approximately

MODAL MUST: It must be the case that p =(df.) Every possible world state is such that p,

then I believe we can analyze the deontic ‘must(e)’ as

MUST(e): In order that e, it must be the case that p =(df.) Of the possible world states in which circumstances C lead to its being the case that e as a result of some member of Â, every such possible world state is such that p.

C is the background conditions, Â is the set of relevant alternatives (mutually exclusive states of affairs which are potential causes for its being the case that e).

We can pull the same trick with ‘may’:

MODAL MAY: It may be the case that p =(df.) At least one possible world state is such that p;

MAY(e): In order that e, it may be the case that p =(df.) Of the possible world states in which circumstances C lead to its being the case that e as a result of some member of Â, at least one such possible world state is such that p.

What about ‘ought’? I think we can do the same thing, although it’s a bit trickier. The modal (or ‘predictive’) ‘ought’ is the one we use in, for example,

(2) It ought to rain tomorrow.

I propose the following:

MODAL OUGHT: It ought to be the case that p =(df.) There are more possible world states, compatible with circumstances C, such that p, than there are such that any other member of Â obtains;

OUGHT(e): In order that e it ought to be the case that p =(df.) Of the possible world states in which circumstances C lead to its being the case that e as a result of some member of Â, more such possible world states are such that p than are such that any other member of Â obtains.

To me, at least, these analyses seem strikingly successful. An obvious objection is that this can’t account for ‘categorical’ uses of deontic terms. I disagree, but that’s too complicated a matter to get into with this post. So consider these merely as analyses of the instrumental use of deontic terms.

What does everyone think?

## 23 Replies to “Semantics of ‘Ought’”

1. Jussi Suikkanen says:

Well, if you ask me, I think that ‘ought’ is an ambigious term here. It seems to be that we use ‘ought’ in very different meaning in the sentence ‘it ought to rain’ than in the ‘one ought not to murder’ sentence. The latter is, I think, connected in virtue of meaning to necessarily having motivation not to murder (in rational agents). No such motivation at all is necessarily connected to the first use. That seems more like another way of just saying ‘it will probably rain tomorrow’.
So, I don’t think that any purely extensional possible world analysis of ‘ought’ will capture the meaning of the practical ought. Of course one could go all externalist (fetishism problems lurking there) or try to account for the motivation in terms of Gricean implicatures. But so far I’m more convinced about internalism and do not really feel the need for possible world semantics of the practical oughts – I’d much rather try an analysis in terms of reasons.

2. Jamie says:

Steve, it’s hard for me to get a feel for your suggestion. For one thing, you haven’t said how all of those tacit parameters get their values. And in the ought part, what does ‘more world states’ mean, given that the cardinalities are sure to be infinite? Maybe a couple of examples would convey the idea.
Jussi, I tend to agree that the ‘predictive’ ought just means something different, but I bet your internalist concerns don’t worry Steve much. First, he does go for that Gricean stuff. Second, it seems quite possible that the motivational force of ought could be accounted for by a story about how the value of e gets determined.

3. Jussi Suikkanen says:

I agree with Jamie that examples would be helpful to get the full idea of the proposal.
Come to think of it – if this is an account of instrumental ‘ought’, then there is something to be said for it maybe. There is a feature of instrumental ‘ought’ that seems to be shared, at least for me, by the predictive ought, and this is non-normativity. I’ve always thought that ‘ought’ in the sentence ‘In order to x, one ought to y’ cannot be normative – it does not imply that anything is to be said for doing y. So take the case of having the end of digging eyeballs out of other people’s skulls. In order to do that one really ought to use a sharp knife rather than a blunt spoon, but this is not to say anything for the said activity of using the knife. With this respect the ought in ‘it ought to rain tomorrow’ is similar, it does not imply that there is necessarily anything to be said for raining tomorrow. Well, sometimes it can be said to imply that ‘it would be good if it rained tomorrow’. But you can take the sentence ‘the bomb ought to have exploded by now’ – this does not imply anything good in many situations.
And instrumental oughts do not seem to require motivation either. I can judge that ‘in order to dig the eyeballs, I ought to use the knife’ without being motivated or irrational. So, I guess I’ll take my objection back.
Of course, it’s still an open question whether this works for the deontic oughts. They seem to be normative and require motivation. I also wonder if one deals with motivation there with Gricean implicatures whether this leads to ambiguity – to distinct concepts of which one has the implicature and one does not.

4. Steve Finlay says:

Jussi:
As Jamie says, I’ve argued at length (in two published papers) for implicature accounts of internalist data. I agree that ‘ought’ is used to communicate something different, but anyone who concludes from this that it means something different I would accuse of making the speech-act fallacy.
Jamie:
The infinite cardinalities objection is a troubling one — I’m not sure what to say about it. (Help, anyone?) But I clearly need to make more proposal clearer and more intuitive, so here goes.
In the case of ‘must’, the idea is that I must A (in order that e) just in case A-ing is the only way to make it the case that e. In the case of ‘may’, the idea is that I might A (in order that e) just in case A-ing is one possible way to make it the case that e. These analyses seem to me straightforward and compelling.
To say that I ‘ought’ (instrumentally) to A in order that e is, I think, to say that A-ing is the most promising or likely way to make it the case that e – in other words, that given all the ways I could try to make it the case that e, A-ing is the one most likely to succeed. I see a clear link to the predictive ‘ought’, and the symmetry with the other modal/deontic terms is enticing. I thought I had captured this thought in my OUGHT(e) principle, but I’m going to have to do some swotting on infinity and probability.
To explain the parameters: whenever you make a claim of the form, [in order that e, it ought to be the case that p], you presuppose some background state of affairs — this is supposed to be captured with parameter C. And when we talk about what ‘ought’ to be the case, there is some implicit contrast class. I intend R to be the class of states of affairs/actions that are potential partial causes of e, and that are not compossible with p (but including p). For example, if at time t Max attempts to mingle with the crowd, then it is not possible that at t he tries to slip out the back door. Both are possible ways to evade arrest, but rule each other out. Instrumentally, Max ought to perform the one that is most likely to bring about his end.
To Jussi’s second post: I appreciate the help. What you see as the marks of ‘normativity’ I would attempt to capture by pragmatics, but I would rather not get into those details here. If you accept this as an analysis of the instrumental ‘ought’, then my mission here is accomplished.
I work this view out in more detail, including the stuff Jussi is worried about, in a paper I’m working on. I’m happy to share, if anyone is interested.

5. Steve,
Here’s a quick probabilistic account, which seems to be in the spirit of your proposal. (Note: P(p|q) is the conditional probability of p given that q.)
1. In order that e, it may be the case that p iff P(e|p) > 0.
2. In order that e, it ought to be the case that p iff P(e|p) > P(e|~p)
3. In order that e, it must be the case that p iff P(e|p) > 0 and P(e|~p) = 0.
I’ve written these in order of increasing strength. On this account, `must’ implies `ought’, that is, if it must be the case that p, then it ought to be the case that p (relative to a given end e). Similarly, both `must’ and `ought’ imply `may’. But neither `ought’ nor `may’ implies `must’.

6. Further to my last comment, let me note that, if we restrict our attention to possible ends, so we assume that P(e) > 0, we can give the following parallel definitions of modal and deontic terms.
(1a) It may be the case that p iff P(p) > 0.
(1b) In order that e, it may be the case that p iff P(e|p) >0.
(2a) It ought to be the case p iff P(p) > P(~p).
(2b) In order that e, it ought to be the case that p iff P(e|p) > P(e|~p).
(3a) It must be the case that p iff P(~p) = 0.
(3b) In order that e, it must be the case that p iff P(e|~p) = 0.

7. Jamie says:

Probability does seem like the best stand-in for number of worlds. But Campbell’s biconditionals have lost contact with Steve’s view because they are in no way causal. So, for example, I think (3b) makes it come out true that in order for the sun to be shining, I must cast a shadow. What responsibility!
I can see several other things to worry about. Briefly:
1. I am not familiar with the expression, “In order that e, it may be the case that p.” I don’t think I’ve ever heard the construction and I have no intuition about what it means. “In order for the Red Sox to win the pennant, they must get a good closer.” Okay. “In order that the Red Sox will win the pennant, they may get a good closer.” Huh?
2. Neither (2b) nor any causal replacement seems to be on the right track to me. We don’t want this coming out true, I assume: in order to avoid traffic on the way home tonight, I ought to kill myself immediately.
3. The parameter I want help with is C, I think. Can you tell us what value C might take in the example of Max that you gave? (By the way, am I right in thinking that the ‘R’in “I intend R to be the class of states of affairs/actions…” was supposed to be an ‘A’, and that you can’t put its hat on in the comments?)
4. I said in my first comment that I agreed with Jussi that the predicitive ought has a different meaning, but I’m definitely on the side (fashionable now) that thinks there are not many senses of ought otherwise. (Madison workshop attenders will remember Judy Thomson expressing this view; I agree with her.) (By the way, that was the deontic ‘will’ in that last parenthetical.) So I doubt that there is any ‘instrumental ought‘. Are you committed to there being such a thing, or is that just a way of speaking?

8. Steve Finlay says:

Campbell, thanks for getting on board the program! These are much cleaner than my efforts, but they’re also vulnerable to some problems that I was trying to avoid.
First, to capture the ‘in order that…’ operator, there has to be a causal (or at least more than a logical or statistical) link between p and e. Otherwise we get results like the following: in order that I am rested tomorrow, I ought to snore for 8 hours tonight. (Because if I snore for 8 hours tonight, it’s more likely that I sleep tonight).
Second, while your treatment of ‘ought(e)’ is cleaner, I don’t think it succeeds. P(e|p) cannot be contrasted simply with P(e|~p): suppose there are five relevant alternatives, and that p is only slightly more likely to result in e than any of the other four. I think it is still the case that it ought(e) that p, but it is not the case that P(e|p)>P(e|~p). Furthermore, your suggestion would have the result that for each of those alternatives, it ought to be the case that it obtains, given the assumption that for each q, P(e|q) is greater than the probability P(e|q) [i.e. the class of relevant alternatives is the class of noncompossible means to e, not the class of noncompossible states of affairs per se.]
Third, I abandoned earlier efforts to explicitly use a probability operator in my analysis of ‘ought’ because I’m trying to get the following result: that the deontic terms are nothing but the modal terms occurring under an operator. You’re offering me, instead, closely related terms. I thought I had accomplished this by quantifying over possible world-states, but then Jamie had to oblige my request to have my bubble burst…
Actually I’m fairly optimistic that I can respond to Jamie’s concern, but it won’t be pretty. How about this: suppose we have two possible means, p and q. There are infinitely many possible world-states in which e as a result of p, and infinitely many possible world-states in which e as a result of q. But it may be possible to find a many-to-one correspondence between these sets of world-states…I’m still mulling this one over.

9. Steve Finlay says:

1. Granted ‘in order that…they may..’ sounds unnatural. But how about this: ‘in order to win the pennant, the Red Sox might get a new closer’. This sounds fine to me, and I think it means just the same thing. They might get a new closer, or they might upgrade their infield, etc. — all various ways they might go about making it the case that they win the pennant.
2. In order to avoid traffic, ought I to kill myself? I deal with cases like this in my paper. Speaking strictly instrumentally, I think cases like this can be true. We do allow noninstrumental considerations into our assessment of means, however, which complicates the picture greatly. In my paper I also suggest that we can filter such possibilities out of the class of relevant alternatives.
3. The text I pasted in for my original post had “R” instead of “A” – I hadn’t noticed the change. In Max’s case, C consists in Max being in a certain location, the police being in a certain location, etc. — all the facts relevant to whether mingling with the crowd is the best way for him to evade arrest.
4. I definitely disagree with Judy, but I think my view is actually semantically simpler than hers: ‘ought’ is no more ambiguous than ‘tall’ and ‘good’, and is the same in instrumental , modal, and categorical uses (although I haven’t argued for that in this post). I don’t think there is a distinct, ‘instrumental’ ought, therefore. But it seems intuitive to me that the ‘oughts’ of prudence, etiquette, morality, law, and epistemology all have different senses. It is possible for what I morally ought to do to conflict with what I legally, or prudentially ought to do.

10. Steve,
I don’t think the second reason you give for rejecting the probabilistic account is correct. I’m pretty sure the following is true. Suppose the propositions p1, p2 … pn form a partition of logical space (so they’re mutually exclusive and jointly exhaustive). And suppose P(e|p1) > P(e|pi) for i = 2, 3 … n. Then P(e|p1) > P(e|~p1).

11. Jamie says:

Steven,
1. Oh, okay, I understand that construction. I think I would say ‘could’ instead of ‘might’.
2. Traffic Story

In order to avoid traffic, you could take Route 10 or you could kill yourself. It is much more effective to kill yourself, since in that case you will most certainly avoid traffic, while taking Route 10 is only reasonably likely to allow you to avoid traffic. Nevertheless, you ought not to kill yourself.

Do you have an explanation for why that paragraph is obviously acceptable? You do think the last sentence is false, right?
3. When someone is judging that Max ought to slip out the back door, she is presupposing that the police are in a certain location, that the door is not locked, and so on? This seems very implausible. There are tests for presupposition that the background conditions pretty obviously fail. Why are we supposed to think that they are presuppositions?
4. It’s true that what you morally ought to do often conflicts with what you prudentially ought to do, but that doesn’t show that there are different senses. What you eat for breakfast is different from what you eat for lunch, but that doesn’t show that ‘eat’ has different senses.
Campbell,
That is true as long as the partition is countable, anyway.

12. Steve Finlay says:

Campbell, you’re absolutely right. An elementary error on my part.

13. Steve Finlay says:

Jamie,
I think the ‘ought’ in ‘you ought not to kill yourself’ is not ‘in order that you avoid traffic’. So it’s not the sort of ‘ought’ I’m trying to account for in this post. It does contaminate our judgements of what we ought to do in order to avoid traffic, but I’ll again suggest that we can filter those possibilities out of the class of relevant alternatives.
On presuppositions: I retract that ill-advised characterization of C (they’re not ‘presuppositions’.) Thinking it over now, what seems relevant is the world-state prior to the occurrence of a member of R. How does that sound?
I accept that there are different ways of explaining apparently conflicting ‘oughts’. My claim is merely that there is some prima facie evidence for the claim that ‘ought’ has different senses. I’d have thought it was the default, common-sense view. (And I’m sure that what I eat for breakfast doesn’t CONFLICT with what I eat for lunch. Certainly they’re compossible).

14. Jussi Suikkanen says:

Steve,
concerning the last point I think what many people would say currently is that in the ‘oughts’ of prudence, etiquette, morality, law, and epistemology ‘ought’ is used in the very same sense to denote the same kind of ought. The conflict is then explained by the fact that these are in first instance pro tanto oughts which are just grounded on different things (or different things explain them). The fact that we can assess what we ought to do over-all seems to support the idea that we are comparing same sort of things – and then we get the over-all ought, full stop, as a result.
I have a vague question though on an issue that has started to bother me lately generally. I call these my ‘Bernard Williams moments’. I’ve started to loose the plot on what philosophical theories are supposed to be doing, when they are succesful, and so on. I know that this is a silly metalevel worry. I like Williams’ idea that we do philosophy in order to understand things that puzzle us in our lives. The question of what we are doing when we are talking about ‘oughts’ can be one such issue about which we would like to have more self-understanding.
Now take your view. I’m being told that when I think to myself that in order to get food, I ought to go to the kitchen, what I mean is that of the possible world states in which circumstances C lead to its being the case that that I get food, as a result of some member of R, more such possible world states are such that p than any other member of R obtains. wow. Now I really understand myself. The point is that the more I think about this analysis, the more puzzled I am. And I didn’t start being this puzzled. I’m sure there is something in your view that helps me to understand new things, but it still escapes me.
I would like some guidance in what the account is supposed to help me to understand. We do not want complicated analysis for their own sakes. If we had the infamous telescope to other possible worlds of course the account could enable me to find out when I ought to do things. But then we also get other silly worries. What if there are no possible worlds, are there then no things I ought to do? Does what I ought to do depend on what coherent stories we can tell? Is the content of my ‘ought’ thoughts really sets of possible worlds and isn’t that an awful lot of information in my mind?
I guess there are answers to many of these questions. Sorry about them. My mind just boggles.

15. Steve Finlay says:

Jussi,
On your first point: I’ve conceded to Jamie that there are different ways of explaining the data. But (a) if, as I think, it can be the case that you morally ought to do what you prudentially or legally ought not to do, then it can’t be the case that these qualifiers merely indicate the source or basis for the ‘ought’: they must qualify the sense, somehow. (b) Perhaps they are just ‘pro tanto’. But then you have a distinction between at least two senses of ‘ought’, in any case. In my paper I address the case of ‘ought’ simpliciter, but this is one of the complications I’m trying to avoid in this post.
Your second concern: I share your caution towards possible worlds — that’s why I talked in terms of possible world states instead. But I don’t think that the analyses are actually as complicated as they sound. If ‘in order that…ought’ essentially involves alternative ways the world might be, causation, and the likeliest way to a goal — which I think it quite clearly does — then no simpler analysis could possibly be adequate.

16. Jamie says:

Steve, the adverbs no doubt qualify the sense. Similarly, ‘for breakfast’ and ‘for lunch’ qualify the sense of ‘eat’. But ‘eat’ itself does not change sense, so there is no case, even prima facie, for thinking that the sense of ‘ought’ itself changes.
For lunch, I ate only salad. For breakfast, I ate toast and no salad. Those would certainly conflict if the adverbs (or prepositional phrases, or whatever they are) didn’t modify the sense of ‘ate’. They do, obviously, but there aren’t two senses of ‘ate’.

17. Steve Finlay says:

Jamie, I think you may be conceding me everything I want. After all, I am arguing that ‘ought’ contributes the same content in every use. The sense in which I am claiming it has ‘different senses’ is just that sentences with unqualified ‘oughts’ are all elliptical for sentences with ‘oughts’ qualified in various ways, such that an unqualified sentence ‘A ought to X’ may be used to express different propositions with different truth values. And you’ve granted this.

18. Kris McDaniel says:

“To me, at least, these analyses seem strikingly successful. An obvious objection is that this can’t account for ‘categorical’ uses of deontic terms. I disagree, but that’s too complicated a matter to get into with this post. So consider these merely as analyses of the instrumental use of deontic terms.”
I hereby request that Steve discuss this issue in a seperate post (if he desires to do this and has the time), since this is what is bothering me. I don’t see how to naturally extend the view to what he calls categorical uses of deontic terms. I’m really curious to see what the anlaysis looks like.

19. Steve Finlay says:

Kris, I’d be happy to oblige. It’s too late tonight. I’ll try to post on Thursday, perhaps just pasting in the relevant section of the paper.

20. Quick question about the predictive ‘ought’. You say,
“The modal (or ‘predictive’) ‘ought’ is the one we use in, for example,
(2) It ought to rain tomorrow.
I propose the following
MODAL OUGHT: It ought to be the case that p =(df.) There are more possible world states, compatible with circumstances C, such that p, than there are such that any other member of Â obtains.
The same holds, presumably, for “the train ought to be here at 6pm”. But it is not clear (to me) why that would be analyzed as there are more possible world states compatible with C in which p than for any other member of that set. Suppose what has happened is that (unknown to those waiting)the train tracks have been destroyed and there is no chance of the train arriving anywhere near 6pm. In the circumstances C there are very few that include p (the train arriving at 6pm). It is nonetheless correct for those waiting to utter “the train ought to be here at 6pm”. That is, someone who utters that sentence in those circumstances is making no mistake: that is what is reasonably expected to occur.

21. Steve Finlay says:

Mike,
Good question. Defining the predictive ‘ought’ is probably trickier than I’ve acknowledged here. I think you’ll agree that if those waiting knew of the damage to the tracks, then they couldn’t appropriately say “the train ought to be here at 6pm” (unless the ‘ought’ is deontic). One possible reply, then, is that in virtue of their epistemic situation, they have every reason to think that there are more possible world-states in which the train arrives at (by?) 6pm than any relevant alternative, and that is why the utterance is appropriate. Another possible reply is that C consists only in the known circumstances. This is a tempting option for me, as I’ve always been inclined to the view that the ordinary concept of probability is epistemic rather than metaphysical.

22. Steve,
Restricting C to known circumstances will yield an unusual logic for your predictive O, if that is something you’re worried about. For instance, you’re probably not going to get as a thesis OA -> <>A, where
‘<>‘ represents metaphysical possibility, since epistemic possibility exceeds metaphysical. You likely won’t have closure
under implication (OA & OB)->O(A & B), since the expectation of A and B individually might be quite high while the expectation of A and B together might be very low. So the logic will probably be pretty weak. Still, that wouldn’t be the worst news. One advantage is that certain important distinctions wouldn’t collapse.
I have one other worry. How will you handle expectations of tautologous states of affairs? Certainly there are some tautologies T (take a complex theorem, if you like) that I’d reasonably expect–not knowing that they are tautologies and(even)having strong evidence that they are not–not to be true. Yet on any interpretation (epistemic or otherwise) the probability of T on any evidence is high. But then I should expect all tautologies to be true. But I don’t.

23. Steve Finlay says:

Mike,
My thoughts on the nature of the predictive ‘ought’ are definitely half-baked, and I will be going back to the drawing board. But the logical consequences you mention don’t worry me much – I’ve never believed in the ‘axioms’ of deontic logic. Against OA-> <>A, the following case occurs to me. Scrutinizing some geneological data, I might say “C ought to be the father of D”. In using ‘ought’ rather than ‘must’, I accept that it is (epistemically) possible that C is not the father of D. But the identity of one’s parents is a metaphysical necessity (right?) What ‘ought’ to be the case here may be metaphysically impossible.
On tautologies — I’m not sure what the relevance of what I ‘should expect’ is. I’m not identifying this as the meaning of the predictive ‘ought’, although there’s obviously a close relation. My inclination is to say that any claim that some tautologous state of affairs ‘ought not’ (predictively) to obtain is false (although possibly justifiable given epistemic limitations).