§ Introduction

My aim in this post is to argue that the ‘ought,’ predicate, interpreted in a global sense, agglomerates. I’ll henceforth refer to this position as agglomeration. The position is as follows. 

 

Agglomeration: If “I ought to do A,” and “I ought to do B,” then “I ought to do A and B,” where A and B are actions, and “A and B,” is a weak conjunctive action.

 

I use the term ‘weak conjunctive action,’ to mean an action of the form “A and B,” where A and B are actions. For any agent X, the sentence “X did ‘A and B,’” is true at time t when (1) “X did A,” is true at time t and (2) “X did B,” is true at time t. There is a sense in which the action ‘A and B’ is constructed by a very weak form of conjunction since the conjunction of actions A and B does not imply that A and B must be done in a particular order or near each other in space or time. To understand how weak these conditions are, contrast the sense in which I use the word ‘and’ in this paper with the work that ‘and’ does in the common-sense interpretation of the action ‘running and thinking.’ In the common-sense interpretation, ‘and’ implies that the agent runs and thinks simultaneously. However, if we interpret ‘running and thinking,’ as a weak conjunctive action, then it is true on Friday that an agent who ran on Wednesday and thought on Thursday did the action of ‘running and thinking.’ 

I use the term ‘ought statements’ to refer to statements of the form “X ought to do A.”  I distinguish between global and local ought statements. An ought statement is global if and only if the obligation to do A in circumstance C that the ‘ought’ statement points too reflects and incorporates all the relevant moral factors in C. Not all obligations are global. A local obligation to do A in circumstance C is an obligation that is true in virtue of a subset of the morally relevant factors in C. For instance, suppose that I borrowed Alice’s sword and promised to return it. In virtue of my promise, it seems that I ought to return Alice’s sword. However, suppose further that I know with certainty that if I give Alice her sword, she’ll use it to kill Bob. Moreover, if I never return her sword, Alice will instead try to kill Bob by punching him, fail to kill Bob, and be imprisoned such that she poses no further risk to public safety. Considering all these facts, including that fact that I promised Alice I’d return her sword, it seems to me that I have an obligation not to return her sword. Thus, I have a global obligation not to return Alice’s sword but a local obligation to return the sword. 

My argument in favor of agglomeration unfolds in two parts. First, I argue that the counterfactual wrongness position on the meaning of ought statements implies that ought statements agglomerates. Second, I argue in favor of counterfactual wrongness and contrast it with alternative views about the meaning of ‘ought,’ statements. Before my argument, I’ll discuss two reasons why the question of whether ought statements agglomerate is of philosophical interest. 

The first reason agglomeration is of interest is that it is an open question. Although the agglomeration principle is facially plausible, skeptics may point out that agglomeration is not a necessary feature of predicates. Take, for example, the predicate “Alice can afford.” Suppose that “Alice can afford a Prius,” and “Alice can afford a Camry.” These two sentences do not imply the truth of the statement: “Alice can afford a Prius and a Camry.” The skeptic can reasonably withhold belief in agglomeration until it is justified by a positive argument. 

The second reason agglomeration is of interest is that, in combination with another principle of deontic logic, the principle that ‘ought implies can,’ agglomeration negates the possibility of moral dilemmas. 

 

Ought implies can: If “I ought to do A,” then “It is possible for me to do A.”

 

A moral dilemma occurs when it is true of someone that “They ought to do A,” and “They ought to do not-A.” ‘Not-A,’ is an action that an agent does when they do not do action A. We can thus rewrite the latter sentence in the definition of moral dilemmas as: “They ought to not do A.” 

 

For the sake of argument, assume that ought implies can, agglomeration, and that Alice is in a moral dilemma: She ought to do A and she ought not to do A. By agglomeration, she ought to do A and not-A. Per the principle that ought implies can, it is possible for Alice to do A and not-A. Recall that “A and not-A” is a weak conjunctive action. The sentence “Alice did ‘A and not-A’” cannot be true at any time t, since if it were true at time t, it would imply that Alice both did action A and did not do action A. Thus, it is logically impossible for Alice to do “A and not-A,” yielding a contradiction with the earlier derivation that it is possible for Alice to do ‘A and not-A,’ from ought implies can. Thus, we must reject one of the three: agglomeration, ought implies can, or the possibility of moral dilemmas. If we believe in agglomeration and ought implies can, then we must disbelief in the possibility of moral dilemmas. 

 

 

§ Counterfactual wrongness implies agglomeration  

The counterfactualwrongness view claims that the sentence “X ought to do A” is true if and only if the sentence “it would be wrong for X not to do A” is true. In this section, I identify the types of ‘ought’ statements that counterfactual wrongness applies to, give an alternate formulation of the counterfactualwrongness in terms of possible worlds, and finally argue that agglomeration of global ought statements holds if we accept counterfactual wrongness of global ought statements. 

First, I claim that counterfactualwrongness only applies to sentences that refer to ‘ought,’ in a global sense. Recall the case of Alice’s sword from the introduction: I promised to return Alice’s sword. However, if I return it, Alice will use the sword to kill Bob. It is true that, in a local sense, “I ought to return Alice’s sword.” However, my intuition is that it would not be wrong of me to fail to return Alice’s sword, precisely because this is a local obligation that does not take in account every relevant factor. This example demonstrates why counterfactualwrongness only applies to global ‘ought’ statements and not local ‘ought’ statements.

Second, we can restate counterfactualwrongness in terms of possible worlds as follows:

 

Possible worlds formulation (PWF): “X ought to do A,” if and only if, in every possible world where X does not do A, X acts wrongly.

 

The PWF shows us is that the crucial feature of a global obligation is that it ‘picks out’ a set of possible worlds in which the agent acts wrongly by virtue of violating the obligation and a complementary set of worlds in which the agent satisfies the obligation. This gives us a criterion we can use to infer the existence of an obligation. 

Now, I’ll argue that agglomeration of global ought statements holds if we accept counterfactual wrongness for global ought statements. Recall that agglomeration claims that the sentences “I ought to do A,” and “I ought to do B,” imply that “I ought to do A and B,” is true. Suppose for the sake of the argument that “I ought to do A,” in a global sense and I ought to do B,” in a global sense. On the PWF, the sentence “I ought to do A,” implies that in each world in the set  of possible worlds in which I do not do A, I act wrongly. Similarly, “I ought to do B,” implies that in each world in the set  of possible worlds in which I do not do B, I act wrongly. 

Moreover, on the PWF, “I ought to do A and B” if and only if in each world in the set  of possible worlds in which I do not do A and B, I act wrongly. Recall that since ‘A and B’ is a weak conjunctive action, it is true that “I did ‘A and B’” at time in world if and only if “I did A” is true at time in world x and “I did B” is true at time in world x. We can thus partition  into three disjoint sets: , the set of worlds in which I did A, but not B, , the set of worlds in which I did B, but not A, and , in which I did neither A nor B. :  is a subset of  , the set of all worlds where I do not do B. Since I act wrongly in every world in   , I act wrongly in every world in  . By symmetric argument, since  , I act wrongly in every world in  Finally, since   and  , I act wrongly in every world in  . Since every world in  the set of worlds where I do not do “A and B,” is in one of , I act wrongly in every world in . Therefore, the sentence “I ought to do A and B” is true in a global sense, since in every possible where I fail to do ‘A and B,’ I act wrongly. Therefore, if you accept counterfactual wrongness of global ought statements, then you must accept agglomeration of global ought statements. 

 

§ Defense of counterfactual wrongness

Before offering a positive argument in favor of counterfactual wrongness and responding to objections, I want to explain the work that counterfactual wrongness does for my argument, and what this requires me to prove. 

Obviously, my argument in favor of agglomeration uses the truth of counterfactual wrongness as a premise. However, readers should not confuse the assertion of counterfactual wrongness with an attempt to define the meaning of the word ‘ought’ or to describe the factors that ground the existence of moral obligations. Counterfactual wrongness does not even offer an account of the sort of thing that an obligation is. These questions are left up to a more specific theory about the contents of morality. Counterfactual wrongness merely imposes the constraint that any moral theory’s view of what is ‘wrong’ to fail to do and what one ‘ought’ to do must coincide. Different theories may forward different views about what makes failing to do an action wrong. Moreover, my argument does not impose the burden of disproving other accounts of what an obligation is or about the truth conditions of ought statements. It is sufficient to show that such accounts are compatible with counterfactual wrongness.

Why should a skeptic accept the counterfactual wrongness view? I’ll break this question down into two parts. First, I’ll argue that “X ought to do A” implies that failing to do A is wrong, and second, that the fact that failing to do A is wrong implies that “X ought to do A.” The first part is straightforward. My intuition tells me that it is wrong to violate one’s obligations unless one has a sufficiently compelling moral reason to do so. However, as I stipulated earlier, counterfactual wrongness only applies to global obligations, which consider all the relevant moral factors. If there were a moral reason against doing my duty which was sufficiently compelling to justify not doing it, then I would not be under a global obligation to do it, as this reason would have been considered in the determination of my global obligation. Thus, it is always wrong to fail to do what one has a global obligation to do, since there can never be a sufficiently compelling moral reason to violate a global obligation. The second part of my positive argument in favor of counterfactual wrongness is to defend that “It would be wrong of X to fail to do A” implies that “X ought to do A.” It is self-evident to my intuition that one ought to avoid doing wrong. 

My argument is, of course, heavily motivated by my intuitions about the nature of wrongness and obligations. However, I’ll next show that in cases where we are faced with two mutually exclusive obligations and we are tempted to reject agglomeration, counter-factual wrongness shows us that we can instead collapse two local obligations into a single global obligation. Consider Sophie’s choice: Sophie’s son and daughter are being held in a concentration camp. Sophie is told that she can choose one of her children to save from death. If she does not choose, both will be killed. 

A skeptic of my argument can make the following claims: It would be wrong of Sophie to fail to save her son, and therefore, counter-factual wrongness implies that “Sophie ought to save her son.” Moreover, it is possible for Sophie to save her son, so there can be no objection that it is only wrong to fail to do something you had the possibility of doing. By symmetric argument, it is true that “Sophie ought to save her daughter.” However, it is impossible for Sophie to do the weak conjunctive action ‘save her son and save her daughter.’ Therefore, ought implies can implies that it is not true that “Sophie ought to save her son and save her daughter.” However, agglomeration implies that “Sophie ought to save her son and save her daughter.” We are led to the conclusion that we must reject either ought implies can or agglomeration. A skeptic can reasonably claim that their intuition that ought implies can is stronger than their intuition in agglomeration and reject agglomeration.

While compelling, the skeptic’s argument goes wrong because it confuses local obligations for global obligations. Recall that our argument in favor of agglomeration only holds in the case of ought statements where it is the case that failing to do the action demanded by the ought statement is always wrong. Moreover, we claimed that such ought statements are always global ought statements. However, I claim that “Sophie ought to save her son/daughter,” is a local statement. Why might it be true that “Sophie ought to save her son/daughter.” Presumably, it is because there is a set of compelling moral reasons to save the son/daughter. However, the very reason why Sophie’s choice is a moral dilemma is because there are equally strong reasons to save the other child. If the reasons in favor of saving one child were stronger than the reasons in favor of saving the other child, than the obligation to save the morally preferred child would override the other, and there would be no dilemma. 

Therefore, the sentence “Sophie ought to save her son/daughter,” should be interpreted as asserting a local obligation, since it fails to consider the morally relevant factor that, in saving the child the obligation points too, Sophie allows her other child to die. Since these are local ought statements, by contrapositive, counter-factual wrongness implies it is not always wrong to fail to satisfy them. In what cases would it not be wrong for Sophie to fail to satisfy her obligation to save her son/daughter? In the cases where she saves her other child instead! This conclusion jives with my intuition: Sophie is not doing wrong by letting her son die and saving her daughter, since she is doing one of the two best possible acts (the other being saving her son and letting her daughter die). Every reason why it might be wrong to let her son die, such as the fact that her son’s last memory would be that of seeing his mother pick his sibling over him, is an equally strong reason to save her daughter. While it is deeply regrettable for Sophie to be put in this situation, Sophie is not acting wrongly when she chooses one child, since there simply was no better option. 

We can complete our account of Sophie’s choice-like situations by specifying what Sophie’s global obligation actually is and seeing that this obligation poses no threat to the agglomeration principle. It would be wrong of Sophie to fail to save either her son or her daughter, so by counter-factual wrongness we infer that “Sophie ought to save her son or her daughter,” in a global sense. Note that we treat the weak conjunctive action “save her son or her daughter,” as a single action. As with the weak conjunctive actions that are conjoined by ‘and,’ we say that Sophie did the action ‘saving her son or her daughter,’ at time if Sophie did the action “saving her son,” at time t or if Sophie did the action “Saving her daughter,” at time t. Since Sophie only has a single global obligation, there are no obligations to be agglomerated, and our account of Sophie’s choice avoids the skeptic’s worry that agglomeration and ought implies can are incompatible. 

Finally, I’ll explain how counterfactual wrongness is compatible with the motivation view, which, stated imprecisely, asserts that “X ought to do A,” if and only if, supposing X were perfectly rational and fully grasped the grounds upon which the obligation is justified, then X would be motivated to do A. Motivation claims that a necessary component of an obligation is an attached justification and that the justification must be sufficiently compelling so as to motivate action. 

Motivation is of interest because it seems to generate cases where agglomeration fails. Consider the sentences “Alice ought to donate to charity,” and “Alice ought to volunteer at the soup kitchen.” It is plausible that, taken individually, the justifications for each of these obligations are motivating, but taken together, they impose a large enough sacrifice on Alice that she cannot be motivated to do both. In this case, it appears that motivation implies that it is not true that “Alice ought to donate to charity and volunteer at the soup kitchen.” This violates agglomeration.

However, counterfactual wrongness resolves this objection. Suppose “I ought to do A,” and “I ought to do B,” in the sense that I can be rationally motivated to do either A or B, but I cannot be rationally motivated to do ‘A and B,’ because the sacrifice is too great. If I do A, then I can no longer be motivated to do B. Therefore, after I do A, it is no longer true that “I ought to do B.” Therefore, failing to do B is not prima facie wrong. Therefore, there are possible worlds where I fail to do B and am not acting wrongly. Counterfactual wrongness of global ought statements implies that I must not have a global obligation to do B. By symmetric argument, I do not have a global obligation to do A. Therefore, it is perfectly consistent with my view that “I ought to do A" and “I ought to do B,” do not agglomerate, since neither is a global ought statement. However, consider possible worlds where I fail to do both A and B. Since I have not done B, I can still be motivated to do A, and am thus acting wrongly. By symmetric argument, I am acting wrongly in virtue of failing to do B. Thus, I act wrongly in every possible world where I do neither A nor B. Thus, counterfactual wrongness implies that I globally ought to do ‘A or B.’ 

New to LessWrong?

New Comment
1 comment, sorted by Click to highlight new comments since: Today at 2:26 AM

I think this gets a LOT simpler and more straightforward when you use weights rather than binary imperatives.  You could model this as importance compared to all other things, or as uncertainty about "ought-ness" of an action.  

If you're 95% ought to do A and 95% ought to do B, you're 90% ought to do A and B (assuming A and B are independent actions.  Different calculations apply if they're correlated somehow).