I. The procreation asymmetry

My aim in this article is to reconcile the following two common-sense claims: that, when all else is equal, (a) we are morally obligated not to create miserable children, but (b) we have no obligation to create happy children. (In the cases I will discuss, I assume that nothing apart from the welfare of one's future child is at stake in the choice whether to procreate. This includes effects on the child's parents, siblings, and any third parties.) Despite the plausibility of the two claims, they appear to be in tension with each other. This is because the most obvious ways of justifying each claim entail the falsity of the other. For instance, if we try to justify (a) by arguing that we have a moral duty to maximize total happiness, we will be forced to reject (b). Alternatively, if we try to justify (b) by arguing that we have a moral duty to maximize the happiness only of those who already exist or who will exist whatever we do, we will be forced to reject (a).Footnote ¹ In virtue of this prima facie tension, the two claims are often taken to form an ‘asymmetry’.Footnote ²

To make the asymmetry vivid, consider the following three worlds. The three worlds are exactly similar except for the varying presence in them of two numerically distinct individuals: Happy, who has positive lifetime welfare, and Sad, who has negative lifetime welfare.

Following Broome, I use the omega symbol (Ω) to represent the welfare of a person who doesn't exist in a given world.Footnote ³ Numbers represent the lifetime welfare of those who do exist in a world; I assume that these numbers measure welfare cardinally and that they allow for interpersonal comparisons of welfare. Departing from Broome, however, I assume that a person has no welfare – just as she has no mass – in worlds where she doesn't exist. This assumption entails (a) that Happy is better off in Wh than she is in Wn and Ws (in other words, that Happy has a ‘life worth living’), and (b) that Sad is worse off in Ws than she is in Wh and Wn (in other words, that Sad has a life ‘worth not living’).

Broome has raised the following objection to (a):

We might put Broome's argument this way: ‘Happy can't have zero welfare in Wn or Ws, because this would entail that she exists in these worlds. Only existing entities can have properties. However, Happy, by hypothesis, does not exist in either Wn or Ws.’ In response, I am inclined to bite the bullet and concede that Happy does exist, in a way, in Wn and Ws. There are independent reasons to accept this claim. For instance, the claim is entailed by necessitism which says that necessarily everything is necessarily something.Footnote ⁵ Necessitism entails that possible things exist necessarily, so in so far as Happy is possible, she will exist in all worlds, including Wn and Ws. However, Happy doesn't exist as a concrete entity, or even as a person, in worlds where she was never born; ‘a merely possible person’, on this view, is a non-concrete entity which could have been a person. So, if we take necessitism on board, ‘Ω’ will refer to the zero welfare held by each merely possible person, understood as existing in this way.

Let Φ(Wx) be an action such that a world, Wx, would obtain if Φ(Wx) were performed. We can now present the procreation asymmetry as consisting in the following two claims:

1. (a) Given a choice between Φ(Ws) and Φ(Wn), only Φ(Wn) is permissible.

2. (b) Given a choice between Φ(Wh) and Φ(Wn), either choice is permissible.

The asymmetry presents a challenge for any moral theory which aims to justify our common-sense moral judgements. Existing attempts to justify the asymmetry are often unsatisfying. For instance, consequentialists tend to appeal to ad hoc asymmetries in value: claiming, for instance, that while it is impersonally worse for the world to contain an additional miserable person, it is not impersonally better for the world to contain an additional happy person; or, that while a person is harmed by being brought into existence with a miserable life, she is not benefited by being brought into existence with a happy life.Footnote ⁶ Non-consequentialists tend to appeal to ad hoc asymmetries in our obligations: claiming, for instance, that while we are obligated not to harm others, we are not obligated to benefit them.Footnote ⁷ While it is perhaps plausible that our negative duties trump our positive duties, this weaker claim entails merely that we should prioritize the non-creation of miserable children over the creation of happy children; it does not entail that we lack any positive duty to create happy children.Footnote ⁸ My own solution is a broadly consequentialist one. But while I seek to explain the asymmetry without appealing to any deontological constraints, my approach of course remains available to those who deny that permissibility in general is determined by the value of outcomes.

By appealing to a distinctive kind of moral actualism, I will offer a novel justification of the procreation asymmetry which avoids any appeal to ad hoc asymmetries in value or obligation. I will not attempt to give a full defence of the theory but simply aim to show how it provides the basis for an attractive account of the asymmetry. I argue, first, that there is a distinct set of permissibility facts for every possible world, W: whether an action is permissible, relative to W, depends only on how it affects those who exist in W. Second, I argue that an action is objectively permissible iff it is permissible relative to the actual world. But in cases where one doesn't know which world is actual, one can aim only to do what is subjectively permissible. I argue that an action is subjectively permissible iff its ‘expected permissibility’ (the degree of objective permissibility one should expect it to have, if performed) is at least as great as that of every alternative.

In the next section, I will give an initial characterization of moral actualism, describing the problems which have led many to think that the view is untenable.

II. Moral actualism

Think of everyone who ever existed and who will ever exist. This includes us (the people who exist now), our ancestors, and our descendants. These are the actual people. This group does not include those who would have existed had we or our ancestors acted differently. Nor does it include those who could exist in the future given actions we and our descendants could, but won't, perform. These people are merely possible.

Now it is very tempting to think that the welfare of merely possible people is morally irrelevant. If so, then in so far as an action's moral permissibility is determined by the amount of welfare it produces, only the welfare of actual people is relevant. For the purposes of this article, I will assume that the only factor which determines an action's moral permissibility is its effect on welfare. This allows me the convenience of articulating moral actualism as a kind of utilitarianism, according to which an action is morally permissible iff it maximizes the welfare of actual people.Footnote ⁹ Or, more precisely:

Note that the permissibility of a merely possible action is not determined, on this view, by its effects on those who would be actual if it were performed (I will consider such a view in Section V, below). Rather, it is determined by its effects on those who are in fact actual (given the actions that are, in fact, performed). This has curious implications for the morality of procreation. Recall our three worlds, Wh, Wn, and Ws. According to moral actualism:

What does this show? First, moral actualism, thus characterized, does not vindicate the procreation asymmetry: on this view, if one actually creates a happy person, then doing so is obligatory; and, if one doesn't create anyone, then the merely possible action of creating a miserable person is permissible. Second, it shows that moral actualism violates the following plausible constraint on moral theories:

But even if one is prepared to reject normative invariance, we are still left with a difficult problem, which is that moral actualism can offer no guidance in procreative choices. Because one cannot know whether a future child is actual before one decides whether to create it, one cannot know, before deciding, whether creating it is permissible (because this depends on whether it is actual). For example, before I have decided whether to create Sad, I cannot know whether doing so is permissible because it is permissible to create Sad only if I don't actually create her. And before I have decided whether to create Happy, I cannot know whether doing so is obligatory because it is obligatory to create Happy only if I do actually create her.Footnote ¹²

I will defend actualism by relying on a distinction between ‘objective’ and ‘subjective’ permissibility. Any moral theory might characterize the objective permissibility facts as being difficult to know or even unknowable. But this is not a fatal problem for such a theory, as long as it can be paired with an appropriate account of subjective permissibility, which says what to do when one is uncertain about the objective permissibility facts. So, we should not worry that the objective permissibility facts are sometimes unknowable according to moral actualism. Once we supplement actualism with a theory of subjective permissibility, we will get a moral theory which is action-guiding, which respects normative invariance and, moreover, which explains the procreation asymmetry.

However, before offering my account of subjective permissibility (in Sections VI and VII), we first need a more explicit characterization of actualism's underlying axiology: its account of how worlds are ordered in terms of betterness.

III. Axiological actualism and axiological relativism

There are two ways in which we might reconstruct actualism's underlying axiology. As I will explain, moral actualism might presuppose either axiological actualism or, instead, axiological relativism. While actualism's normative extension is the same either way, my account of subjective permissibility (which says what to do, when one is ignorant about the objective permissibility facts given by moral actualism) will require the latter interpretation.

So far, we have assumed that moral actualism presupposes just one ordering of worlds, where the value of a given world is determined by the welfare of actual people in that world:

As we have seen, if there is just one ordering of worlds, just one set of permissibility facts is determined by this ordering.

But we might instead understand actualism as presupposing a distinct ordering of worlds for each possible world W, as determined by how those in W fare in different worlds:

Given axiological relativism, we must posit the existence of a distinct set of permissibility facts for every possible world:

According to axiological relativism, how do Wh, Wn, and Ws compare in value? We need to give three answers, relativizing separately to each world:

Now, if I must choose between Φ(Wh), Φ(Wn) and Φ(Ws), how should I make my decision? Moral relativism generates three sets of world-relative permissibility facts, as follows:

But these world-relative permissibility facts don't, by themselves, tell me how to make my decision. For I must now ask: ‘which world-relative permissibility facts am I governed by?’. In other words, we need an account of how the world-relative permissibility facts determine our non-relative permissions.

IV. From moral relativism to moral actualism

So, which actions are non-relatively permissible, given the world-relative permissibility facts given by moral relativism? My answer is that an action is permissible iff it is permissible relative to the actual world (W@):Footnote ¹³

Because Φ(Wx) is permissible_W@ just in case actual people have at least as much welfare in Wx as they have in any other available world, it follows that we have found an extensionally equivalent way of defining moral actualism. I will rely on this new formulation, however, because its explicit appeal to moral relativism will be crucial for my account of subjective permissibility.

As we have seen, moral actualism can't, by itself, function as an action-guiding moral theory. For instance, if I am choosing between Φ(Wh), Φ(Wn), and Φ(Ws), I can't know – in advance of deciding – which of the three worlds is actual. And because the permissibility of an action, according to moral actualism, depends on how it affects actual people, I can't thus know whether an action is permissible before I decide whether to do it.

My solution, as I have indicated, is to interpret moral actualism as a theory of objective moral permissibility. While we can't always know whether an action is objectively permissible, we can nevertheless know whether it is subjectively permissible. I will show how we might supplement actualism with a theory of subjective permissibility below, in Sections VI and VII. But let's first consider a tempting alternative to moral actualism, one which promises to do away with the need for a theory of subjective permissibility. Instead of saying that an action is permissible iff it is permissible relative to the actual world, this view says that an action is permissible iff it is permissible relative to the world in which it is performed.Footnote ¹⁴

V. Weak actualism

While I can't know, in advance of choosing between Φ(Wh), Φ(Wn), and Φ(Ws), whether any of these options is obligatory, merely permissible, or prohibited (according to moral actualism), it is striking that I can know what normative status each option would possess (according to moral actualism) if I performed it. Thus, I can know that if I chose Φ(Wh), then Φ(Wh) would be obligatory; that if I chose Φ(Wn), then Φ(Wn) would be merely permissible; and that if I chose Φ(Ws), then Φ(Ws) would be prohibited. It is tempting, moreover, to try to use facts about the counterfactual normative status of an option to determine its actual normative status. For instance, one might propose the following view which, following Caspar Hare, I will call weak actualism:Footnote ¹⁵

Weak actualism entails the following accounts of obligatoriness, permissibility, and impermissibility (where the bracketed alternatives follow straightforwardly, assuming axiological and moral relativism):

Let's now see weak actualism in action, in a situation where one must choose between Φ(Wh), Φ(Wn), and Φ(Ws). According to weak actualism:

The verdicts concerning Φ(Wn) and Φ(Ws) seem promising. Indeed, one may argue, on this basis, that weak actualism is able to vindicate the procreation asymmetry. But the verdict concerning Φ(Wh) threatens the very coherence of the view, as it doesn't make sense to say that Φ(Wh) is obligatory while Φ(Wn) is permissible. If Φ(Wh) is obligatory, then it is one's only permissible option (this follows simply in virtue of the meaning of ‘obligatory’), and so Φ(Wn) would have to be impermissible; on the other hand, if Φ(Wn) is permissible, then, for the same reason, Φ(Wh) can't be obligatory (i.e. one's only permissible option): it would have to be either merely permissible or prohibited. This is a first reason to reject weak actualism.

Let's introduce a new world, Ws*, which contains another equally miserable individual, Sad*; now suppose that you are forced to choose between Φ(Ws) and Φ(Ws*). Weak actualism entails that both Φ(Ws) and Φ(Ws*) are impermissible because the Ws people have less welfare in Ws than in Ws*, while the Ws* people have less welfare in Ws* than in Ws. But if an option is impermissible, then some alternative option must be permissible. So the theory again comes out as incoherent. This is a second reason to reject weak actualism.Footnote ¹⁷

Some have argued that all of one's options could be impermissible; that is, they defend the possibility of ‘moral dilemmas’.Footnote ¹⁸ Weak actualists might thus address the second challenge by arguing that there is no contradiction in saying that Φ(Ws) and Φ(Ws*) are both impermissible. But this move does not help address the first challenge.

One might try to salvage weak actualism by taking its definition of permissibility as primary (‘Φ(Wx) is permissible iff it is permissible_Wx’), and then defining impermissibility and obligatoriness in terms of it, by saying that an action is impermissible if it is not permissible, and that an action is obligatory iff it is the only available permissible action. By doing this, we get the following alternative definition of obligatoriness:

The resulting view is attractive (a) because it supports the asymmetry: given Φ(Wh), Φ(Wn), and Φ(Ws) as options, it implies that Φ(Wn) is permissible and that Φ(Ws) is impermissible, and (b) because, given the same options, it avoids the implication that Φ(Wh) is obligatory. However, we are still left with the problematic implication that, given a choice between Φ(Ws) and Φ(Ws*), both options are impermissible. A further problem is that it is arbitrary whether we take the weak actualist definition of permissibility or, instead, its definition of obligatoriness, as primary. But this choice cannot be arbitrary because it has normative implications. For instance, if we take the weak actualist account of obligatoriness as primary, then weak actualism implies that Φ(Wh) is obligatory.

One reason for dwelling on the problematic implications of weak actualism is that this view forms the basis for many seemingly attractive accounts of the procreation asymmetry. For instance, it has been put to me that we can explain the asymmetry by appealing to the contractualist principle that ‘acts are morally permissible iff they are justifiable to the people they affect’.Footnote ¹⁹ The idea is as follows: because Φ(Wh) is as justifiable to the Wh people as Φ(Wn) is to the Wn people, both actions are permissible; but because Φ(Ws) is less justifiable to the Ws people than Φ(Wn) is to the Wn people, Φ(Ws) is impermissible if Φ(Wn) is an option. But this is just a version of weak actualism which employs a contractualist account of an action's world-relative normative status, as follows:

So, despite its appeal, this contractualist explanation of the asymmetry inherits all the problems with weak actualism which we have already seen.

VI. Subjective actualism

Having rejected weak actualism, let's now return to moral actualism, according to which whether an action is objectively permissible depends on whether it is permissible relative to the actual world.

In order to salvage moral actualism, we need an account of what to do in cases where we are ignorant about the objective permissibility facts, but have access to the world-relative permissibility facts described by moral relativism. Specifically, we need a theory which says what is subjectively permissible in cases where: (a) I am choosing between Φ(Wx) and Φ(Wy), (b) I don't know whether Wx or Wy is actual (and, thus, don't know which options are objectively permissible), but (c) I do know whether (and to what extent) Φ(Wx) is permissible_Wx and whether (and to what extent) Φ(Wy) is permissible_Wy.

I propose the following generic account of subjective permissibility:

If we take moral actualism as our theory of objective permissibility, then an action is objectively permissible iff it is permissible_W@. It follows that an action is subjectively permissible iff its expected permissibility_W@ is at least as great as that of every alternative. But how do we determine an action's expected permissibility_W@? The answer is simple: if I were to perform Φ(Wx), then Wx would be the actual world; so, the expected permissibility_W@ of Φ(Wx) is its permissibility_Wx. So, moral actualism generates the following account of subjective permissibility:

VII. Explaining the asymmetry

In order to employ subjective actualism, we first need a way of quantifying world-relative permissibility; that is, a way of expressing the degree to which an action, Φ(Wx), is permissible_Wx. I make two assumptions: (a) that permissibility is a maximal notion: no permissible action is more permissible than any other, and (b) that actions can vary in their degrees of impermissibility: some actions are more impermissible than others. I thus stipulate that all ‘fully’ permissible actions have a ‘permissibility value’ of 0 and that impermissible actions have varying permissibility values below 0 corresponding to their degrees of impermissibility. (The degree of an action's permissibility is thus either 0 or some number below 0 representing its impermissibility.)

So, what is the degree of permissibility, relative to the world where it is performed, of each of our options: Φ(Wh), Φ(Wn), and Φ(Ws)?

Now, subjective actualism says that an option, Φ(Wx), is subjectively permissible iff its degree of permissibility_Wx is at least as great as that of every alternative. It follows that Φ(Wh) is subjectively permissible, that Φ(Wn) is subjectively permissible, and that Φ(Ws) is subjectively impermissible. Thus, by appealing to subjective actualism, we can explain the procreation asymmetry.

VIII. The problem cases

Let's now consider how subjective actualism deals with the two cases, raised earlier, which proved problematic for weak actualism. The first case concerned a choice between Φ(Wh) and Φ(Wn). (Weak actualism says, implausibly, that Φ(Wh) is obligatory, while Φ(Wn) is permissible.) The second case concerned a choice between Φ(Ws) and Φ(Ws*), where Ws* is a world containing a distinct but equally miserable individual, Sad*. (Weak actualism says, implausibly, that both options are impermissible.)

First, consider the choice between Φ(Wh) and Φ(Wn). Moral actualism says (a) that, relative to Wh, Φ(Wh) is obligatory while Φ(Wn) is impermissible; and (b) that, relative to Wn, both actions are permissible. On either relativization, the permissibility facts are perfectly coherent. What one must do, according to moral actualism, depends on which world is actual. But as these ‘objective’ permissibility facts are unknowable, one can only do what is subjectively permissible. So, as I have argued, because the expected permissibility of both options is equal – both would be fully permissible, if performed – both actions are subjectively permissible.

Now, consider the choice between Φ(Ws) and Φ(Ws*). Moral actualism says (a) that, relative to Ws, Φ(Ws*) is permissible while Φ(Ws) is impermissible, and (b) that, relative to Ws*, Φ(Ws) is permissible while Φ(Ws*) is impermissible. Again, on either relativization, the permissibility facts are perfectly coherent. But again, when faced with the choice between Φ(Ws) and Φ(Ws*), one can't know which action is objectively permissible (because one doesn't know which choice one will make and, thus, which world is actual). So, as I have argued, because both options have equal expected permissibility – one should expect both actions to be equally impermissible, if performed – both are subjectively permissible.Footnote ²²

But this last claim requires further argument. Why should one think that the degree of permissibility_Ws of Φ(Ws) is exactly the same as the degree of permissibility_Ws* of Φ(Ws*)? In order to answer this question, we need to make the idea of the degree of permissibility of an action more precise; that is, to introduce a measure of impermissibility.

Luckily, there's a natural way of deriving this further information. The permissibility_Wx of Φ(Wx) depends on how it affects the Wx people. For instance, Φ(Wx) is impermissible_Wx iff, had I performed some alternative action, the Wx people would have been better off. So, how permissible Φ(Wx) is depends on how much better off the Wx people could have been. In other words, the ‘permissibility value’ of an action, Φ(Wx), is equal to the difference between the welfare of the Wx people in Wx and the highest possible welfare of the same people.

This implies that when an action, Φ(Wx), is ‘fully’ permissible_Wx – i.e. when the Wx people couldn't have had higher welfare, given any alternative actions – its degree of permissibility_Wx is 0. (I stipulated this earlier, but without argument.) This resonates with the plausible idea that permissible actions are all equally permissible. But because impermissible actions can have varying degrees of permissibility – interpreted as values below 0 – it also follows, plausibly, that some impermissible actions are more impermissible than others.

According to subjective actualism, Φ(Wx) is subjectively permissible iff its degree of permissibility_Wx is at least as great as the degree of permissibility_Wy of Φ(Wy), for every available Φ(Wy). I have now proposed that the degree of permissibility_Wx of Φ(Wx) is equal to the extent to which the Wx people could have been better off. It follows that we can simplify our formulation of subjective actualism:

Let's now return to the choice between Φ(Ws) and Φ(Ws*). The degree of permissibility_Ws of Φ(Ws) is determined by how much better off the Ws people – specifically, Sad – could have been. This is the difference between Sad's welfare in Ws – which we've stipulated to be −10 units of welfare – and her welfare in Ws*, which we've stipulated to be 0 units of welfare. The degree of permissibility_Ws of Φ(Ws) is thus −10. But if we assume that Sad* also has −10 units of welfare in Ws*, it follows that the degree of permissibility_Ws* of Φ(Ws*) is also −10. So, given a choice between Φ(Ws) and Φ(Ws*), both options are subjectively permissible, because both have equal expected permissibility.

More generally, if given a choice between creating two miserable individuals, then, other things being equal, one is subjectively required to create the less miserable of the two. Curiously, however, given a choice between creating two happy individuals, there is no subjective obligation to create the happier of the two, according to my theory. For instance, consider the following case:

(This table, and others to follow, depicts a choice between certain mutually exclusive and exhaustive options, listed in the first column. Subsequent columns show, for each option, (a) the welfare of those who would exist, were that option to be taken (with Ω representing the zero welfare of someone who doesn't exist, given an option), and (b) the expected permissibility (EP) of each option.)

As both options have equal expected permissibility, subjective actualism implies that both are subjectively permissible. Many have the intuition, however, that choosing Φ(Wh) would be impermissible. Subjective actualism thus appears unable to solve Parfit's ‘non-identity problem’, i.e. the problem of justifying the intuition that we are morally required, other things being equal, to create the happier of two possible individuals.Footnote ²⁴ However, see the Appendix, for an attempt to solve the non-identity problem within the constraints of subjective actualism.

IX. Subjective obligatoriness

I have argued that Φ(Wh) is not subjectively obligatory. However, it seems that a similar argument could be given for the claim that Φ(Wh) is subjectively obligatory. At least one of these two arguments must be unsound.

My argument for the claim that Φ(Wh) is not subjectively obligatory depended on the principle that an action is subjectively permissible iff the degree of objective permissibility one should expect it to have, if performed, is at least as great as that of every alternative. As the expected permissibility of Φ(Wh) is at least as great but no greater than that of Φ(Wn), both options are subjectively permissible and neither is subjectively obligatory.

But perhaps we should instead employ an alternative principle, which says that an action is subjectively obligatory iff the degree of objective obligatoriness one should expect it to have, if performed, is greater than that of every alternative. Because one should expect Φ(Wh) to be objectively obligatory, if performed (relative to Wh, Wh is better than both Wn and Ws), this alternative principle entails that Φ(Wh) is subjectively obligatory.

I will argue, however, that the alternative principle is false. While the expectation that an action would have at least as much objective permissibility, if performed, as any alternative makes that action subjectively permissible, the fact that an action would have more objective obligatoriness, if performed, than any alternative does not make it subjectively obligatory. My argument depends on the following case:

Knowing these facts about yourself, you should expect that your handing over $1000 would be objectively obligatory (assuming you are obligated to return money you have borrowed), and you should expect that your keeping the money would be objectively permissible (assuming it is permissible not to give someone money if they are not owed it). But your expectation that handing over $1000 would be objectively obligatory, if performed, clearly gives you no reason to hand over the money, and so does not support the claim that doing so is subjectively obligatory.

In contrast, the expectation that an action would be objectively impermissible, if performed, clearly gives one a reason not perform the action. Consider the following case:

Let's assume that it is objectively impermissible to lend money to someone who will use it for evil purposes. If so, you should expect that your lending the money would be objectively impermissible. This expectation gives you a clear reason not to perform the action, and so supports the claim that doing so is subjectively impermissible.

The upshot of this discussion is that the subjective obligatoriness of an action is not determined by its expected objective obligatoriness. Rather, because an action is subjectively permissible iff its expected objective permissibility is at least as great as that of every alternative (as previously argued), and because an action is subjectively obligatory iff it is the only available subjectively permissible action, it follows that an action is subjectively obligatory iff its expected objective permissibility is greater than that of every alternative.

X. Retrospective evaluation

Suppose you have already faced a choice between Φ(Wh), Φ(Wn), and Φ(Ws), and that you picked Φ(Wh). You now know that Wh is the actual world and, consequently, that Φ(Wh) was objectively obligatory all along. Can you infer that Φ(Wh) was subjectively obligatory all along? (The implication that Φ(Wh) was subjectively obligatory all along is, of course, inconsistent with the account presented in this article.) This inference is valid, however, only if P is true:

I will offer a counterexample to P. But first, note that Frank Jackson's famous ‘drugs case’ is a counterexample to a related principle, P*:

In Jackson's case, you don't know whether your patient has condition X or condition Y, and can prescribe one of three drugs: A, B, or C. B will cure the patient if she has X, but kill her if she has Y. C will kill the patient if she has X, but cure her if she has Y. A will partially cure the patient either way.Footnote ²⁵ In this case, you know that you are objectively required either to prescribe drug B or to prescribe drug C, and so that you are objectively prohibited from prescribing drug A. But, clearly, you are subjectively required to prescribe drug A. So, P* must be false.

Let's suppose that you do, sensibly, prescribe drug A. Your patient is partially cured, and it emerges that she had condition X. However, the discovery that you were objectively obligated, all along, to prescribe drug B does not allow you to infer that you were subjectively obligated, all along, to prescribe drug B. (You shouldn't kick yourself, or feel guilty, for having prescribed drug A.) We thus have a counterexample to P.

XI. Expected permissibility: aggregative versus individualist views

According to subjective actualism, an action, Φ(Wx), is subjectively permissible iff its degree of permissibility_Wx is at least as great as the degree of permissibility_Wy of Φ(Wy), for every available Φ(Wy). In Section VIII, I argued that the degree of permissibility_Wx of Φ(Wx) is determined by the extent to which the Wx people could have been better off, given the alternatives. However, there are two ways in which we might cash out this thought. On the first way, we look for an available world, Wy, where the total welfare of the Wx people is as great as possible. The degree of permissibility_Wx of Φ(Wx) is the difference between the total welfare of the Wx people in Wx and their total welfare in Wy:

An alternative approach first considers the extent to which each Wx person could have been better off, and then aggregates these extents:

It turns out that there are various cases where the two views diverge. I will survey some of these cases and will show that the individualist view systematically has more plausible implications. To start, the aggregative and individualist views can have different implications for an action's expected permissibility in cases with more than two options. For instance, consider the following case:

On the aggregative view, the expected permissibility of each option is −10, which entails (given subjective actualism) that each option is subjectively permissible. This result is obviously unfortunate. In so far as our aim is to vindicate the idea that it is impermissible to create miserable people, our theory should say that creating two miserable people is impermissible when one has the option of creating just one miserable person. The individualist view avoids this result as, on this view, Φ(W3) has lower expected permissibility than either of the other options.

Consider another multi-option case. Take our original three worlds – Wh, Wn, and Ws – and imagine that a third party, Bob, is slightly better off in Ws than in the other two worlds (perhaps because he takes some small pleasure in Sad's suffering). Now suppose you are presented with a choice between the three worlds.

Intuitively, Bob's addition should not change the original permissibility of the three options. However, the aggregative view implies, weirdly, that Φ(Wh + b) is obligatory and that Φ(Wn + b) is impermissible. The individualist view avoids this implication.

Here are some other cases where the individualist view avoids absurd implications of the aggregative view:

The aggregative view entails that both options in Case 4 are subjectively permissible. This result is puzzling. The starting point of our inquiry was the idea that creating happy people is not obligatory, and that creating miserable people is impermissible. It is therefore surprising that simultaneously creating both a happy and a miserable person should be permissible. Why should the value of an otherwise non-obligatory outcome (that of creating a happy person) cancel out the value of an otherwise impermissible outcome (that of creating a miserable person)? The individualist view captures the common-sense judgement that Φ(W1) is impermissible.

Things are similar in Case 5. On the aggregative view, it appears that we can make P1 as miserable as we like, just so long as we create enough other marginally happy people along with her. This is implausible. Again, the individualist view captures the more palatable judgement that Φ(W1) is impermissible.

Our intuitions concerning Case 6 are perhaps less clear. However, because the cost to P1 as a result of one's choosing Φ(W2) (−8) is less than the cost to P2 of one's choosing Φ(W1) (−10), it is arguable that Φ(W2) is obligatory. Only the individualist view captures this intuition.

In so far as we have no obligation to create happy people, it seems that P1's welfare should be the only relevant consideration in one's choice in Case 7. As P1 exists in both worlds, but is better off in W1 than in W2, it seems that Φ(W1) should be obligatory. Only the individualist view captures this judgement.

Because P1 now exists only given the first alternative in Case 8, it seems that P1's welfare is now an irrelevant consideration. In contrast, in so far as we are obligated not to create miserable people, P2's and P3's welfare are each relevant. But given that P2 and P3 are equally miserable, the reasons generated by their welfare intuitively ‘cancel out’, so we are left with the intuition that both options are permissible. Only the individualist view captures this judgement.

Interestingly, the aggregative and individualist views give different verdicts regarding Parfit's case of the ‘Two Medical Programmes’, which is represented in Case 9.Footnote ²⁶ While neither view supports Parfit's intuition that both options are permissible in this case, the individualist view captures the idea that Φ(W2) is preferable to Φ(W1) because someone has a ‘complaint’ in W1 (specifically, P1 can complain that he would have been better off in W2), while no one has any complaint in W2.

This final case stands out because, here, the implications of the aggregative view are more plausible. In particular, only the aggregative view agrees with the intuitive judgement that Φ(W1) and Φ(W2) are each permissible.Footnote ²⁷

XII. Choice-set dependence

One might worry that subjective actualism violates a requirement that permissibility facts not be choice-set dependent. Permissibility facts are problematically choice-set dependent, according to a theory, iff the introduction of a new option can make a previously impermissible option permissible, or a previously non-obligatory option obligatory. Unfortunately, subjective actualism (on both the aggregative and individualist variants) has this feature. For instance, Case 3 shows how the aggregative view can lead to choice-set dependence. And the following case illustrates how the individualist view can lead to choice-set dependence.

Given just the first two options, both options are permissible (on the individualist view). But if we add the third option, Φ(W1) becomes obligatory. However, choice-set dependence in Case 11 is not implausible. Before we introduce Φ(W3), the considerations for and against Φ(W1) and Φ(W2) are arguably symmetrical. However, this symmetry is broken upon the introduction of Φ(W3) because it makes Φ(W2) less attractive for one person (the extent to which P1 could have been better off is now greater), without making Φ(W1) less attractive for anyone. (The introduction of Φ(W3) does, of course, make both Φ(W1) and Φ(W2) more attractive for P3 (because the extent to which he could have been worse off is now greater). But this doesn't affect the relative attractiveness of Φ(W1) and Φ(W2).)Footnote ²⁸