2.1. The separateness of persons objection to totalism

Classical aggregative utilitarianism (totalism) was said by Rawls to fail to take into account the ‘distinction between persons.’Footnote ³ Exactly what the separateness objection comes to and whether it is probative have been much discussed in recent years.Footnote ⁴ We might, however, consider Rawls’ point not a definitive objection against totalism but rather a rough diagnosis that directs our attention to the mechanism by which totalism produces results we think are clearly false in – among others – simple equality and additional person cases.

According to that diagnosis, the problem with totalism is that it evaluates a given world by simply aggregating across the population whatever it is – happiness, preference satisfaction, capability, resources, whatever – that gives existence its value for each person who does or will exist at that world. I will call that thing well-being and understand that for a person to have more well-being in one world than another is just for the one world to be better for that person than the other. Then, according to totalism, for one world to be better than another is just for the one world to contain more well-being in the aggregate than the other does.

It’s true that, under that principle, a reduction in any one person’s well-being level will automatically mean (other things equal) a reduction in aggregate well-being – in, that is, overall value. For that reason, totalism can’t credibly be accused of ignoring the plights of anyone at all. But perhaps what the ‘separateness of person’ objection is meant to point out is that totalism doesn’t take those plights into account in the right way. For example, it doesn’t put limits on how far a particular person’s well-being might fall as long as, inch by inch, that fall is offset by gains in well-being that might themselves be thinly spread across a sizeable number of people who will remain very well off however the future unfolds. Nor does it put limits on how far a given person’s well-being might fall as long as, inch by inch, that fall is offset by gains in well-being accrued by additional people, that is, people who need never have existed at all.

Consistent, though, with what we’ve just said is that the problem with totalism isn’t that it aggregates well-being and thus fails to distinguish one person from another but rather that it aggregates well-being and thus fails to distinguish one sort of loss from another. Thus, let’s imagine that losses of (or gains in) well-being come in units. We might accept that an additional unit of well-being makes things no less better for a better off person than that additional unit makes things for the less well off person. We might accept that the one unit has the same personal value, whoever happens to get it. But consistent with that we might also accept that the unit that makes things better for the better off person may have no moral value at all. It may not make things morally better. But allocated to the less well off person – well, that’s a different story. That one unit may then have full moral significance; it may then easily make things morally better.Footnote ⁵

Ditto when the trade-offs are between an existing person who will either incur or not incur a loss of well-being and a possible person who will either incur or not incur a loss of well-being as a function of whether he or she happens ever to exist. Perhaps the lost unit of well-being matters morally when the loss is due to how the world unfolds for the existing person but not when it’s due to the possible person’s never existing at all. Perhaps, correspondingly, the gain in well-being that makes things better for the possible person by way of bringing that person into existence doesn’t make things morally better, whereas the gain in well-being that makes things better for the existing person does make things morally better.Footnote ⁶

Rawls’ comment thus can be subjected to at least two distinct analyses. Whichever we choose, however, the conclusion may be the same. The problem with totalism is that it aggregates well-being. Either we should not aggregate to begin with or, if we do aggregate, we should aggregate something other than well-being.Footnote ⁷

2.2. Some cases

Consider the simple equality case. Agents have exactly two options. They can produce a population of 99 people p1 … p99 who enjoy unimaginably wonderful lives along with one person q who has a life of complete squalor, degradation, pain, misery and suffering. Whatever we think it takes to bring it about that q’s life is far less than worth living, that’s what we should imagine happening to q. Alternatively, agents can produce exactly the same population of individuals, each of whom, including q, enjoys a life that is just short of the unimaginably wonderful.

Totalism declares that the first option, or outcome or possible future or world, is just as good as the second. We agree that that’s a mistake. It’s a mistake to deem the loss to q in w1 to be completely offset – whitewashed away – by gains to p1 … p99. It’s a mistake not to accord to q’s very deep loss in w1 a moral significance that the relatively shallow losses p1 … p99 suffer in w2 lack, and it’s a mistake not to recognize that q’s gain in w2 has a moral significance that the gains p1 … p99 enjoy in w1 lack. But the offsetting itself seems just a function of the particular aggregative principle that totalism embraces. Hence, the diagnosis: totalism produces results that we think are clearly false in virtue of the fact that it aggregates well-being across the population. So, we shouldn’t aggregate well-being across the population.

That totalism is oblivious to the distinctions we have just made is one strike against totalism. Perhaps, of course, the problem with totalism is not that it’s impervious to the value of equality; perhaps its real failing is that it’s impervious to values that prioritarianism, or perhaps leximin, are designed to capture. Whatever are the finer-grained features of our diagnosis of the problem with totalism, the account it offers of the simple inequality case should make one point clear: that more aggregate well-being doesn’t necessarily make for a morally better world.

Totalism produces results that seem just as clearly false in the additional person case – and it seems to generate those false results using the same defective mechanism that was at work in the simply equality case. Here, boldface indicates a person who does or will exist, and italics paired with the asterisk a person who never exists, at the indicated world.

Totalism concludes that w1 and w2 are equally good – that neither world is better than the other, and that the acts that end in those different worlds are both permissible – by way of simply aggregating well-being across the populations at w1 and w2. But surely we can agree that w1 is better. Totalism fails because it ignores the distinction between q’s loss in w2 and p’s loss in w1. It leaves us with no room to say that q’s loss in w2 is morally significant in a way that p’s loss in w1 isn’t. But that seems exactly what we want and need in the end to say.

2.3. The person-affecting intuition as antidote to totalism

Let’s suppose, going forward, that either we should not aggregate to begin with or, if we do aggregate, we should aggregate something other than well-being. That we should reject totalism doesn’t mean that we should reject the many features totalism has that we find compelling, including its insistence that agents maximize the good and do the best they can for people and that acts are to be evaluated on the basis of their consequences.Footnote ⁸

But can a theory retain those features but determine whether one world is morally worse than another without aggregating well-being across the population?

Perhaps it can, by shifting from talk about maximizing well-being in the aggregate to talk about maximizing well-being for each person as an individual.Footnote ⁹ Thus, a person-affecting account of the additional person case might start with the core principle that, where the well-being of each and every person who does or will exist at a given world has been maximized, nothing at that world counts against what has been done there or makes the world itself morally worse than any other world. In other words:

A plausible, person-affecting account of the simple equality case would need to say more since such an account must address how conflicts are to be resolved and trade-offs made. But the underlying goal of the person-affecting intuition – and of the person-affecting approach more generally – for both the additional person case and the simple equality case would be to attend carefully in a way that totalism does not – attend, that is, in the right way – to whether a particular loss – q’s loss in w1 in the simple equality case; q’s loss in w2 in the additional person case – may have a moral significance that other losses lack.

It’s the additional person case that we need to focus on for purposes here. A few notes are in order. First, P* is restricted to people who do or will exist at the world under evaluation. I believe that that restriction is critical. For the claim that q’s loss in w2 is just as deep as p’s loss in w1 – that w2 is worse for q by the same amount that w1 is worse for p – seems to me both plausible and, though controversial, hard clearly to refute. If that claim stands, and if we don’t restrict P* to people who do or will exist, then P* cannot help us make the distinction between q’s plight in w2 and p’s in w1. But the whole point – or at least, a big chunk of the point – of the person-affecting approach is to make exactly that distinction.Footnote ¹⁰

Second, P* provides only a necessary condition on when an act is morally wrong and a world morally worse than another. P* isn’t, that is, definitional.

And finally one addition. We can now just add that well-being is maximized at a world w for a person p if and only if there is no other world w′ available to agents from w, that is, accessible to w, such that p’s well-being in w′ is greater than p’s well-being in w.Footnote ¹¹

For purposes here, it’s useful to separate out P*’s deontic, or act-evaluating, component, and its telic, or outcome-evaluating, component.

Taking the contrapositive of P* and borrowing from Parfit, we can put the deontic component this way:

The telic component, in contrast, is rarely so well-constructed. Thus, it’s often formulated elliptically, as follows: for one world to be worse than another, the one world must be worse for a person who does or will exist at that world.Footnote ¹³ But worse for a person who does or will exist at that world than what?

P* itself provides an answer to that question. Before licensing the critical inference, the wide-angled P* requires a more far-reaching inquiry, an inquiry not just into how a person p fares in w as compared against w′ but against all other accessible worlds w″ as well. P*, in effect, makes the test for when one world is worse than another harder to fail. Thus, again taking the contrapositive, we can write:

I believe that the ellipsis must be filled out in just this far-reaching way if the resulting principle is to stand up to the nonidentity problem – or indeed to have any chance of acceptance at all.

It’s thus worrisome to me that no one does it this way. Rather, in cases where the ellipsis is filled out at all, it’s filled out in a way that forces us to conclude that w isn’t worse than w′ on the basis of a very narrow inquiry. Specifically, the industry standard for filling out the ellipsis is to require that w″ be identical to w′ – to, that is, in effect omit w″ from the picture altogether.

But where worlds w and w′ are the worlds that we are to rank, and the case is one in which the person who seems to have been disadvantaged exists in w but not in w′, that way of filling out the ellipsis forces us to conclude that w is at least as good as w′ is unless we find that w is actually worse for that person than never existing at all would have been.

But that’s a pretty easy way of jacking up the overall good of w. In fact it’s too easy. How a person p is treated in w may be an absolute travesty – compared, not against how p is treated in w′, but against how p is treated in w″. Why shouldn’t that latter fact count against w? Why should that latter fact be deemed immaterial to the analysis? Why should the person-affecting theorist accept a requirement that dictates that a person-affecting comparison between w and w′ can be completed without reference to what is going on in any third world w″ even in a case where what is going on in w″ seems painfully relevant to any accurate evaluation of w?Footnote ¹⁴

That philosophers standardly opt for such a sure-to-fail formulation of the telic component of the person-affecting intuition – a principle that, as we shall see, bars a plausible person-affecting solution to the nonidentity problem and a principle that on other grounds we will be forced to reject – in place of what P* itself suggests of course gives us pause. And pause we shall.

But first some of the plusses of P* should be at least noted. For example, P* opens the door to an entirely plausible account of the additional person case. Thus, P* instructs that w1 is without any moral deficiency – that is, that the choice, or act, a1 is permissible, and w1 isn’t worse than w2.

Now, P*, as noted, provides only a necessary condition for when an act is wrong and a world worse. But P* is perfectly consistent with further person-affecting principles, including certain sufficient conditions on wrongdoing and moral worseness, that seem entirely uncontroversial. Such principles can be expected to tell us, among other things, that what happens to the existing q at w2 does count against what is done at w2 and does make w2 worse than w1 is.Footnote ¹⁵

We can put these points another way. P* gives us the resources we need to say that p’s loss at w1 just doesn’t count – and also gives us the room we need to say, now putting still other person-affecting principles to work, that q’s loss at w2 does count. And these are just the things that we want to say about the additional person case.

2.4. The moral status of merely possible people

The fact that P* makes moral worseness, and wrongdoing, a matter of how people who do or will exist fare at the world in question may seem to suggest that accepting P* requires a commitment to the position that existing and future people have some sort of a moral status that the merely possible lack.

Thus, it may seem that anyone who accepts P* must surely accept the Pareto-like Px as well.Footnote ¹⁶

It’s worth pointing out that Px actually does a better job with some additional person cases than certain other principles that have been explored in connection with the person-affecting approach do. I am thinking here of the prior existence view (PEV).Footnote ¹⁷

Consider, for example, the procreative asymmetry. According to that two-part intuition, it is (other things equal) wrong to bring a miserable child – a child whose life is less than worth living; a child whose well-being level falls into the negative range – into existence but permissible not to bring a happy child into existence. The second half of that asymmetry aligns nicely with the account P* suggests for the additional person case. But it’s the first half, the miserable child half, of the asymmetry we want to focus on here.

According to Px, if it would have been better for the miserable child for that child never to have existed at all, and if leaving that child out of existence would not have made things worse for any other existing or future person, then it’s wrong, and makes the world worse, to bring that child into existence. That result seems right.Footnote ¹⁸

Consider, in contrast, the account PEV offers of the miserable child case. According to PEV, we ‘should promote the welfare of existing beings and those who will exist, independently of our decisions.’Footnote ¹⁹ PEV thus finds ‘no grounds for objecting to’ the decision to bring the miserable child into existence.Footnote ²⁰ And that result seems clearly false.

Of course, that Px does a better job with the miserable child half of the asymmetry than PEV does doesn’t mean that Px is true. Px is susceptible to its own counterexamples and must be set aside as well.

Consider the following case. Let n be very large.

One might even change the example by eliminating ‘qn’ and letting the class of q-persons be infinitely large. Either way, Px tells us that w1 is worse than w2. But that result seems clearly false, and so we reject Px.

There’s a larger lesson here. It won’t do to oversimplify our discussions of obligations in respect of the merely possible. It might seem tempting to say that any third child I might have produced in addition to Annabel and Tommy must be, as a never-existing or indeed non-actual being, devoid of moral status – and that that’s why it was permissible for me not to produce such a third child. But that temptation should be resisted. If my not producing a third child was indeed permissible, it’s not because that additional, never-existing, non-actual child has no moral status. Rather, something else is going on, something that holds me accountable for many of the losses Annabel and Tommy might incur but not for the very deep loss incurred by any such third child. We shall return to what that something else might be in Section 2.5 below. But for the moment the important thing is to underline that it won’t do to connect moral status to modal status – that is, to say that some people (actual people, or perhaps existing people) have full moral status while other people (non-actual, never-existing people) have none at all. Px is false.

Another problem case for Px is addition plus, a variation on Parfit’s mere addition paradox.

Px implies that a1 is wrong and that w1 is morally worse than w2 is. The better view, however, is that the plight the never existing q faces in the non-actual world w2 counts against what is done at w2 and makes w2, not just worse than w3, but – I believe – worse than w1 as well. What is uncontroversial is that what happens to q in w2 makes w2 worse than w3 is. After all, w3 makes things better for q than w2 without making things too much worse for p. What is perhaps controversial – though I think correct – is that what happens to q in w2 also makes w2 worse than w1 is even though w1 is worse for q than w2 is. What happens to q in w2, in other words, counts not just against w2 as compared against w3 but also against w2 as compared against w1. But that means that q, after all, matters morally, even though q is merely possible relative to w1, for purposes of both evaluating a1 and comparing w1 against w2.

Such an account of addition plus suggests, then, the following: never-existing, non-actual merely possible people matter morally, just as you and I matter morally, and they matter radically: they matter morally, not only at the never-existing and non-actual worlds where it so happens that they do or will exist, but also at worlds where they never exist at all.

2.5. The moral status of the merely possible; implications for P*; variabilism

We can itemize some general points regarding the moral status of possible people, a class that includes you and me and people who don’t now but will at some future time exist and people who don’t now and will never exist.

1. (a) Each person has the very same moral status as each other person. People who are, relative to a given world, merely possible matter morally in just the same way that people who do or will exist in that world matter morally.

2. (b) Actual people have no special moral status. It’s not the case that actual people, that is, people who do or will exist at the uniquely actual world, have a moral status that others lack. And it’s not the case that the moral status actual people have is greater than the moral status other people have.

3. (c) Existing (and future) people have no special moral status. It’s not the case that the only people who matter morally for purposes of evaluating a given act at a given world are the people who do or will exist under that act and at that world. In addition plus, q matters morally for purposes of evaluating a1 even though q never exists under a1.

4. (d) Moral status not ‘conditional’ on existence. It’s not the case that the moral status of a given person is conditional on that person’s existence. Again, in addition plus, q has moral status relative to w1 and a1 even though q never exists at w1.

5. (e) Moral status necessary. We take for granted that people have moral status in worlds where they do or will exist. Bad things that happen to them there count against what is done there and shift the balance in the overall comparison of one world against another. But there’s more: the person who does or will exist in one world has moral status in both that world and in every other world. The bad thing that happens to a person in a world where that person does or will exist bears on the evaluation of what is done at worlds where that person never exists.

6. (f) No discounting of the losses incurred or the gains accrued by the merely possible. It’s just as important that the lives of the merely possible go well at a non-actual world where those people have lives as it is that our lives go well at the actual world. Thus, in addition plus, if improving q’s plight in w2 counted for only 10% of what improving p’s plight in w1 counts for, then agents are obligated to improve p’s plight. But we reject that result.

The foregoing compendium of claims regarding the moral status of the merely possible might well lead one to think that we must reject P* – along with the happy child half of the asymmetry and, in the additional person case, the idea that nothing at w1 counts against what has been done at w1 or against w1 itself. If (1) the merely possible matter morally in exactly the same way you and I matter morally, and if (2) at no cost to anyone else bringing a merely possible person into an existence worth having makes things better for that person, then doesn’t it follow that (3) failing to bring that person into existence (other things equal) is wrong and makes the world worse and (where other things aren’t equal) at least counts against that act and that world?

No. We can accept (1) and (2) but still insist on P*. There’s space between premises and conclusion. Thus, we can accept that all people matter morally and in exactly the same way but that there are limits to how any given person matters morally. In respect of each of us, what’s important, from the moral point of view, is that diminutions in our well-being at worlds where we do or will exist be avoided. And, in respect of each of us, what’s not important, from the moral point of view, is that diminutions in our well-being at worlds where we never exist at all be avoided. In respect of each of us, in other words, some losses of well-being – we might even say some harms – have full moral significance while others have no moral significance at all. The losses a person incurs at worlds where that person does or will exist have full moral significance – they count against those worlds and against the acts that bring those worlds about and in favor of alternate worlds and alternate acts. In contrast, the losses a person incurs at worlds where that person never exists have no moral significance at all.

That view, which I have elsewhere called variabilism, is pretty simple.Footnote ²¹ It’s not exactly common sense since we might not be able to rely on common sense to recognize that merely possible people have the same moral status that you and I have. But it has many commonsensical implications, including that the losses Annabel and Tommy incur as existing people matter morally – and that I, as an important agent in their lives, was obligated to work to avoid (both before and after their conception, the nonidentity problem, as we shall see in Section 3 below, notwithstanding). But the loss some third child I never produced at all incurs in virtue of his or her never existing at all simply doesn’t show up on the moral radar, either here, where that child never exists, or in an alternate world, where that child exists and is happy. The loss here doesn’t count against what was done here or in favor of what was done there; the gain there doesn’t count in favor of what was done there or against what was done here.

Ditto for you and me. Our parents weren’t obligated to produce us or anyone else (other things equal) to begin with. But given that they did produce us, they had many obligations (we think!) to work (both after and before we were conceived, nonidentity problem, again, notwithstanding) to make our lives go well.

A word on what is called the complaint view is in order. We might – though it’s kind of a waste of energy in view of how things turned out – rail against the prospect of our never coming into existence at all. We might, that is, complain at this world about our nonexistence at an alternate world at which we never exist. But it doesn’t follow from that that our complaint has any moral significance. Similarly, we might at this world applaud our parents (express ‘thanks’) for their having brought us into existence. But it doesn’t follow that that applause has any moral significance. Variabilism takes advantage of that logical failure, that gap. According to variabilism, our applause – our thanks – has moral significance only if what we are applauding reverses a morally significant complaint; and our complaint has moral significance only if the loss complained about is a loss incurred at a world where we do or will exist.Footnote ²²

Applied, then, to addition plus, variabilism yields the following: q’s ground for a complaint at w2 is based on q’s being worse off in w2 than q is in w3 – it’s not based on q’s being worse off in w2 than q is in w1. It follows, as a conceptual matter, that q has a ground for applause – for thanks – at w3 that is based on q’s being better off at w3 than q is in w2. But it doesn’t follow that q has a ground for applause at w3 that is based on q’s being better off in w3 than q is in w1.

3.1. Two types of nonidentity problems

The nonidentity problem purports to be a counterexample against P* (in both its deontic and its telic forms). I have argued elsewhere that the person-affecting approach is capable of solving the nonidentity problem – but only if we first recognize that the problem comes in distinct types.Footnote ²³ Cases that give rise to the nonidentity problem – whatever their type – have the following pattern: (a) we think the act is wrong and the world worse but (b) we just can’t see how any person who does or will exist under that act or at that world has been made worse off. However, it seems that for a certain subset of nonidentity cases – the Type I cases – the line of reasoning that results in (b) is based on a fallacy. Type I cases include Kavka’s pleasure pill and slave child cases, Parfit’s depletion and risky policy cases, cases involving historical injustices and environmental malfeasance (including do-nothing policies on climate change) and many others. In all such cases, I argue, closer inspection shows that we can readily identify just how an existing or future person has been made worse off. If that argument is correct, then the Type I problem simply dissolves.Footnote ²⁴

Other cases – the Type II cases – resist that analysis. They leave us no room to dispute (b); there is no fallacy that we can simply identify and then avoid.Footnote ²⁵ But on reflection it also seems to me that the Type II cases do not describe a choice that is clearly wrong or a world that is clearly worse.Footnote ²⁶

An example of a Type II case would be a case in which agents choose to bring into existence a child with a genetic or chromosomal disorder but whose life nonetheless will be unambiguously worth living. To insure that we are reacting properly to the case – that is, that our minds are clear of any presupposition that might prejudice the evaluation – it’s critical that we assume two further points: that it’s certain that bringing the one child into existence will place no other existing or future person at all at any risk of any well-being loss, and that it’s certain that the one child’s own well-being has been maximized (that is, agents had no ability whatsoever, including by acts they may have performed before the child is ever conceived, to have made things better for that child). In such cases, it’s not clear to me that the act under scrutiny is wrong or the world worse. If that’s correct then, for the Type II problem, (b) stands but (a) can be rejected.

3.2. The pleasure pill case

My particular concern in this paper, however, is not the Type II problem but rather the Type I problem. For it is my proposed solution to the Type I problem that is put in jeopardy by the issues I noted at the beginning of this paper. If an axiological constraint we are compelled to accept rules out P* as an acceptable formulation of the person-affecting intuition, then the sort of solution I have proposed for the Type I problem shall inevitably fail.

But first I need just to sketch how that solution is supposed to work. The pleasure pill case works well for that purpose. There, a couple, prior to conceiving a child, ‘paus[e]’ to take a pleasure pill; their choice immediately ‘change[s]’ – from what, we might ask – the identity of the child then conceived.Footnote ²⁷ But it also damages the embryo or fetus in some way, causing the child to be born ‘mildly handicapped.’Footnote ²⁸ As such, the child will not be as well off as any nonidentical child who might have come into existence instead had the couple not paused to take the pill.

Let’s call the child in fact brought into existence Andy and call any arbitrary, better off child the couple might have conceived had they not paused to take the pleasure pill Ruth.Footnote ²⁹

Does the couple’s choice to pause to take the pill make things worse for Andy? The argument is that it does not – that, had the couple not taken the pill, Andy would never have existed at all.Footnote ³⁰ And it’s surely better to exist impaired – provided the existence is worth having; and here it’s part of the case that the existence is worth having – than it is never to have existed at all.

The story itself – not the graph, but the story – gives us enough information to decide that what the couple has done is wrong – and that the world where the impaired child Andy exists is worse than the world where the nonidentical but better off child Ruth exists instead. The usual view is that the story also gives us enough information to conclude that what the agents have done in taking the pill is maximizing for Andy. And from there we are to infer that P* – in both its deontic and telic forms – is false.

In fact, however, that latter point – that the choice is maximizing for Andy – is not what the story tells us at all. The story conveys a good deal more detail than the modally impoverished graph itself suggests.

We need only think about the story in a little bit more detail to realize that it’s clearly possible for Andy to have come into existence without the couple having taken the pleasure pill. Nor is the future where Andy exists and is better off simply a logical or conceptual possibility for agents but rather something agents had the power, ability and resources to bring about. It’s, in other words, a possible future that was accessible to agents at the critical moment just prior to choice, a future that agents had the actual power, ability and resources to bring about, a future that is not barred, for example, by the laws of nature. After all, the pleasure pill isn’t a fertility pill. The couple could have paused to take an aspirin instead – and then have proceeded toward conception on exactly the same schedule they in fact follow in w1.

Kavka agrees. The world where Andy exists and is better off is not completely out of reach. It’s just, Kavka notes, enormously improbable that such a better-for-Andy world would have unfolded had the couple done otherwise than just as they did in w1.Footnote ³¹

The view that probabilities bear on how choices are to be evaluated is, of course, entirely plausible. And we’ll need to understand how Kavka and others think that that point plays out in the pleasure pill case – why, that is, they think we are forced to conclude that what the couple has done doesn’t make things worse for Andy – that is, that it is maximizing for Andy. But first let’s correct the graph.

One other preliminary point is in order before we turn to probability. It might seem that any discussion of probability will surely remain just a sideshow given that the case stipulates that, had the couple not taken the pleasure pill, Andy would never have existed at all. It might seem, that is, that, in view of that stipulation, w3, though accessible, must be irrelevant to the question whether a1 makes things worse for Andy. It might seem that a1 can make things worse for Andy – colloquially, harm Andy – only if had a1 never been performed – only if but for a1 – Andy would have been better off. And here by stipulation but for a1 Andy would never have existed at all. Hence, whatever role probability might sometimes play in our analyses of other cases, we may seem forced in this case to conclude that Andy hasn’t been made worse off. We may seem forced to say ‘no harm done.’

But that way of understanding what it is for a given act to make things worse for, or to harm, a person has long been discredited. Suppose we are trying to figure out whether my shooting you in the arm makes things worse for you – whether it harms you. It’s part of the case – I prove, we can imagine, in court – that, had I not shot you in the arm, I would have shot you in the head. Does that mean that I didn’t harm you when I shot you in the arm – really, that I have benefited you, even perhaps that you owe something to me under a theory of unjust enrichment for my having so kindly saved your life? No, not, at least, on the further entirely natural and appropriate supposition that I also had the further option of not shooting you at all. That I would not have availed myself of that further option says something about me – about, perhaps, my character. But it doesn’t show that I haven’t harmed you when I shoot you in the arm. I have made you worse off, not compared against where you would have been had I not shot you in the arm, but against another baseline altogether: where you could have been and indeed would have been had I refrained from shooting you altogether.Footnote ³²

3.3. Argument from probability

The real obstacle, then, to any finding of ‘harm done’ in the pleasure pill case emerges not from a stipulation as to what would have been but rather from a line of reasoning that exploits the surely correct fact that, for any person, that person’s own coming into existence ordinarily remains, at least until the point of conception, a highly improbable event.

In the shoot-you-in-the-arm case, it’s easy to see how I might have both tried to make and indeed succeeded in making things better for you. I could simply have stood there and not shot you at all. But what would have happened had the couple in the pleasure pill case tried to make things better for Andy? What would have happened had they diverted from their actual course in w1 and tried to conceive the very same child they in fact conceive in w1 by trying to pause to take the aspirin on the very same schedule they in fact follow in w1?Footnote ³³ What would the probability have been that w3 would have unfolded? What would the probability have been even that a3, a highly detailed act and not a generic choice, would have been performed in place of the many, many alternate acts that would have just as well implemented the couple’s generic choice to pause to take the aspirin?

Almost nothing; close to zero. But surely it’s better for Andy to have what he has in w1 than it is for him to have only a very tiny probability of obtaining the better existence that he has in w3. Which surely in turn means that a1 and w1 are, after all, maximizing for Andy – and that P* is therefore false.

It’s the inference to the result that a1 and w1 are maximizing that we should resist. We can’t infer that a1 and w1 are maximizing for Andy on the basis of a comparison between the actual well-being Andy has in w1, given the couple’s choice to pause to take the pleasure pill, against Andy’s expected well-being in w3 or in any other world, given the couple’s choice to pause to take the aspirin instead. But expected well-being is what Andy’s very low probability of ever existing at all, given the couple’s choice to pause to take the aspirin in place of the pleasure pill, goes to. We might debate at length whether personal betterness is a matter of having more actual, or more expected, well-being. What we can’t cogently do, however, is to mix and match actual well-being and expected well-being and then conclude anything about personal betterness at all.Footnote ³⁴

Philosophers who have considered my proposed strategy for dissolving the Type I problem have generally resisted it, pointing out that we’d surely all prefer to retain the flawed but good existence we in fact have rather than relinquish what we have in exchange for a very small chance at a still better existence.Footnote ³⁵

If there were any point in the case where the temporal ordering of Andy's alternatives lined up in just that way, then that point might gain some traction. But there isn’t. The cases are structurally distinct.

Consider, e.g., the moment just prior to choice. The couple’s choice to pause to take the pleasure pill at w1 doesn’t secure Andy’s existence. Just prior to choice, it’s no more probable that Andy will exist given the couple’s choice to pause to take the pleasure pill than that Andy will exist given the couple’s choice to pause to take the aspirin.Footnote ³⁶ I concede that, just prior to choice, the probability that Andy will come into existence given the couple’s choice to pause to take the aspirin is very small. But clearly, just prior to choice, the probability that Andy will come into existence given the couple’s choice to pause to take the pleasure pill is also very small and indeed just as small. But the outcome for Andy, if he does exist, is considerably better under the one choice than under the other. A standard expected value calculation then tells us that the choice to pause to take the pleasure pill doesn’t, after all, maximize expected well-being on behalf of Andy.

The upshot? The couple’s taking the pleasure pill maximizes neither actual nor expected well-being on Andy’s behalf. Thus, we can, whether we think that making things worse for Andy is a matter of diminished actual well-being, diminished expected well-being or both, find ample room to conclude that their taking the pleasure pill does, after all, make things worse for Andy.Footnote ³⁷

I believe, accordingly, that we have dissolved the Type I problem. We have a clearly wrong act and a clearly worse world. But that act and that world are also worse for Andy. We are thus left with no grounds for rejecting P*.

4.1. Why P* is critical to proposed solution to the nonidentity problem

I noted earlier that virtually all theorists formulate the core person-affecting intuition in a way that allows us to reach the critical inference after a significantly narrower inquiry than the more far-reaching inquiry that P* demands.Footnote ³⁸ We can put the telic component of the standard formulation – which I’ll call Ps (telic) – as follows.

Thus, P* requires us, before we can determine that w isn’t worse than an alternate world w′ is, to canvas all accessible alternates w″ to w. If w is worse for any person p who does or will exist in w than any such w″, then the necessary condition on outcome worseness that P* sets forth is satisfied and we can’t infer from P* that w isn’t worse than w′ is. In contrast, Ps generates the result that w isn’t worse than w′ is far more readily – whenever there is no p in w such that w is worse for p than w′ is.

The distinction here is not merely technical. P*’s more far-reaching inquiry is essential to the person-affecting solution to the Type I problem that I have proposed.

Thus, about the pleasure pill case, I want to say that w3 shows that w1 is worse not just than w3 is but also worse than w2. We need to say w1 is worse than w3 to solve the nonidentity problem. But we also need to say that w1 is worse than w2. Why? It’s highly implausible that w2 is either better or worse than w3 is. (Notably, P* concurs.) But if that’s so, then, since w1, we agree, is worse than w3, we must also agree that w1 is worse than w2.Footnote ³⁹ But Ps itself says that it’s not; Ps implies that it’s not the case that w1 is worse than w2 since there’s no one in w1 such that w1 is worse for that person than w2 is. In contrast, P* opens the door to our saying exactly what we need to say about that case: that w1 is worse than both w2 and w3 and that w2 is exactly as good as w3 is.

4.2. Objection based on axiology; inconsistency argument

But does the more far-reaching P* run afoul of a certain axiological constraint that Ps itself survives? The clearest argument that it does comes in the form of an inconsistency argument.

Consider the following two-outcome case one in which only w2 is accessible to w1:

P* (telic) implies that w1 isn’t worse than w2 since there’s no existing or future person in w1 such that that person is worse off in w1 than that person is in any alternate accessible outcome. But now consider a three-outcome case that leaves everything just as it is in the two-outcome case but adds a third accessible outcome.

P* implies that w2 is at least as good as w3 is – indeed, that w2 and w3 are equally good. Those results are perfectly plausible. But as noted earlier we want and need to say as well that w1 is worse than w2 is. Since there is someone, p, who is worse off in w1 than that same person is in some alternate accessible outcome, namely, w3, the condition P* (telic) spells out on w1 being worse than w2 is satisfied. P* thus allows us to avoid the result that w1 is at least as good as w2 is, which in turn leaves us room to say that w1 is worse than w2.

Of course, we draw upon a further person-affecting principle to generate that latter result; we can’t on the basis of just P* conclude that w1 is worse than w2. But that further principle will, like P*, exploit the fact that w1 is worse for some person p who does or will exist in w1 than w3 is for p.Footnote ⁴⁰

But now we have an inconsistency. We have said both that w2 is at least as good as w1 – that’s the result we obtained when we applied P* (telic) to the two-outcome case. But applying both P* along with our further principle to the three-outcome case, we then obtained the result that w1 is worse than w2.

Ps safeguards against such an inconsistency. It makes w3 irrelevant to our comparison between w1 and w2. It insures that what we say about how w1 compares against w2 in the two-outcome case just is what we say about how w1 compares against w2 in the three-outcome case. It insures that what we say in both cases is that w1 is at least as good as w2 is.

Now that seems a problematic result – and sufficient grounds, I think, on which to reject Ps. But Ps is not my concern here. My concern here is to determine whether P* indeed involves us in inconsistency – a result that would render what I consider a highly plausible solution to the Type I nonidentity problem impossible.

Is there then a way to retain P* but avoid the inconsistency? In the foregoing, I have been explicit that what we are comparing are not mere distributions of well-being across identified populations but rather full-blown, fully detailed, possible worlds, or futures, or outcomes. So let’s think about those details for a moment. In the two-outcome case, agents lack the power, the ability, the resources, to make things better for p than they are in w1. Their only options in w1 are to bring p into existence at the lower well-being level and to leave p out of existence altogether. In contrast, in the three-outcome case, agents have exactly that power, that ability, those resources, in w1. The agents there are fully capable of avoiding the loss p incurs in w1 on p’s behalf in the three-outcome case.

But worlds are finely distinguished; any little ‘change’ in a given world means we are talking about a different world altogether. If I have an ability, a power, a resource in one world that I lack in another, then we can be sure that the worlds referred to constitute two distinct worlds.Footnote ⁴¹

We are thus free to say, indeed it seems must say, that w1 and w2 in the two-outcome case just aren’t the same worlds as w1 and w2 in the three-outcome case. We avoid the inconsistency by introducing a more exacting vocabulary. In the three-outcome case, the fact of the third outcome is in effect entailed by our descriptions of the first two worlds. In the two-outcome case, our descriptions of the first two worlds rule out that third world.

The upshot? It remains a complete mystery to me why theorists associate Ps with the person-affecting view. If Ps is part of the person-affecting view, then the person-affecting view can’t solve the nonidentity problem. But why not go with P* instead?Footnote ⁴²

4.3. Objection based on accessibility

A further concern about P* arises from the fact that P* – along with the family of person-affecting principles that P* seems naturally to fit in with – ties its worseness rankings to a notion of when one world is accessible to another. I have here adopted a fairly liberal notion of accessibility. For purposes here, accessibility has been defined to include, not just those outcomes the individual agent, working on his or her own and independently of what other agents do, has the power, the resources and the ability to bring about, but also those outcomes the agent working together with other agents (whether in concert or collaboration or by agreement or simply coincidentally) has the power, the resources and the ability to bring about.Footnote ⁴³

The problem with building the narrower notion into our formulation of the telic form of the person-affecting intuition is that the resulting principle generates multiple worseness rankings – rankings that we would need to relativize in order to avoid inconsistency – depending on whether we (a) focus on just the outcomes that the individual agent has the power, the resources and the ability to bring about, or (b) focus on outcomes that agents working together have the power, the resources and the ability to bring about.Footnote ⁴⁴ And it seems plausible that relativizing our rankings in that way would be a mistake.

The worry is that our more liberal notion of accessibility will mean that the connection between the permissibility of a single individual agent’s act and outcome worseness – between the deontic and the telic projects – may not be as clear-cut as we may have thought (or hoped) it was. On the other hand, the defense ‘Whatever act I performed, given the acts of other agents, the same horrible outcome would have obtained; therefore, my act was permissible’ often fails in spectacular ways. Participating in a firing squad consisting of nineteen other armed agents isn’t necessarily permissible, even if not participating would not have produced a better outcome.

So perhaps it’s not, in the end, clear that a more liberal notion of accessibility is problematic. Perhaps indeed an approach that casts a wider net on outcomes that are accessible – and then carves out from the class of otherwise impermissible acts those acts performed by an agent as an individual in cases where what that agent has done by participating in bringing about the bad outcome actually produces a better outcome than the outcome produced by that agent’s simply walking away – will in the end seem a more workable approach.Footnote ⁴⁵