3.1 Background; the nature of my complaint

In this part 3, I turn to the question of why MacAskill favors the existentially unrestricted form of longtermism over the existentially restricted form.

To his great credit, what has changed his mind, and compelled him to accept a position that he earlier on found “bizarrely unintuitive” (169), is the arguments. But what is it that makes those arguments themselves so compelling? In what follows, I take a close look at two of those arguments – and, specifically, two of the principles those arguments rely on – and make the case that they can be trusted only by those theorists who (wittingly or not) already take for granted that the value of existence controversy is to be resolved against Narveson – i.e., that the existence-insensitive way of resolving that controversy is correct.Footnote ¹⁶

Before proceeding to the arguments, however, I should clarify that I am taking MacAskill’s arguments to be working toward a single conclusion – that is, MacAskill’s own existentially unrestricted form of longtermism. I should note, however, that on occasion MacAskill couches his arguments not just as arguments favoring that particular form of longtermism but also as arguments favoring the existence-insensitive way of resolving the value of existence controversy (see e.g., 169). Those two positions being quite obviously connected, it might seem that it’s MacAskill’s intention to kill two birds with one stone: the existentially restricted form of longtermism and the existence-sensitive way of resolving the value of existence controversy.

If the two-birds-with-one-stone reading is the better reading of MacAskill’s arguments, then my complaint would just be that the very construction of the arguments quietly assumes that the existence-sensitive way of resolving the value of existence controversy is false. Under that way of reading the arguments, my complaint would then be that they beg the question.

Perhaps, however, the better reading is that the main purpose of the two arguments – and it’s certainly their ultimate purpose – is to favor just the existentially unrestricted form of longtermism. On that reading, there’s no begging of the question.

For purposes of this present paper, I will go with that second reading. My complaint, then, will be that the arguments he has given us are, on close inspection, very gappy. To fill the gap – to complete the argument; to avoid the invalidity – we must write in principles that we have no reason at all, beyond the bare assumption that the value of existence controversy is to be resolved in a particular way, that is, the existence-insensitive way – to accept. Or so I shall argue in what follows.

Now, save, perhaps, for that critical point, none of what I say will be news to MacAskill. He understands the pros and cons of the arguments – which he, in the main, credits to other theorists (169) – at least as well as I do. He is simply saying that his considered opinion is that the population ethics mainstream has got things right – that their arguments, regardless of any deficiencies, are on balance compelling. And I’m simply reminding the population ethics mainstream that the existence-insensitive way of resolving the value of existence controversy really is just an assumption. Once we recognize that particular way of resolving the controversy as the bare assumption that it is, the force of MacAskill’s arguments favoring his own existentially unrestricted form of longtermism begins to fade.

3.2 The three option argument

John Broome has fashioned a devastating argument against Narveson’s original – and now widely quoted – formulation of his own underlying intuition: other things equal, we are “neutral” about making happy people.Footnote ¹⁷ MacAskill discusses Broome’s argument with approval – and then, on the basis of the conclusion of that argument, goes on to argue that the existentially unrestricted form of longtermism is correct (176–189).Footnote ¹⁸

Let’s, with Broome, call that formulation of Narveson’s underlying intuition the “neutrality intuition” (“NI”). NI has two parts.

Broome’s three option case (Table 2) then launches the argument against NI:

Table 2.

Three option case

Charlotte* never exists in future f1;

Charlotte does or will exist in futures f2 and f3;

Charlotte is better off in f3 than in f2

1. 1. If NI, f1 is exactly as good as f2 and f1 is exactly as good as f3. (NI in combination with conceptual principlesFootnote ¹⁹ )

2. 2. If NI, f2 is exactly as good as f3. (1 in combination with conceptual principlesFootnote ²⁰ )

3. 3. f2 is worse than f3. (standard Pareto principleFootnote ²¹ )

4. 4. If NI, inconsistency. (2 and 3)

Therefore,

1. 5. Therefore, NI is false; i.e., Part I or Part II or both Part I and Part II of NI are false. (4)

2. 6. But surely f2 isn’t worse than f1; surely, Part II of NI is true. (5)

Therefore,

1. 7. f2 is, after all, better than f1; Part I of NI is false. (5 and 6)

Therefore,

1. 8. The existentially unrestricted form of longtermism is correct.

Once we determine, thanks to Broome’s argument, that NI is false – once we reach the intermediate conclusion in line 5 – we then identify – in line 6 – which part of NI is false (that it’s Part I; after all, how can the mere addition to f2 of the happy child Charlotte make f2 worse than f1?). We then reach the further intermediate conclusion – in line 7 – that f2 is, after all, better than f1 (while f3, we can add, is better still). Other things equal, making happy people makes things better.Footnote ²² From there, it’s a short step to MacAskill’s ultimate conclusion in line 8: “yes, it is a loss if future people are prevented from coming into existence – as long as their lives would be good enough” (187–188). And: the “loss of those future people who will never be born if humanity goes extinct in the next few centuries” is a “moral loss” (36, 169; 188). And: “the early extinction of the human race would be a truly enormous tragedy” (182–183). And: the “practical upshot of this is a case for space settlement” (189). Finally: though MacAskill doesn’t “think that we should scold those who choose not” to have children, he does think that not having children imposes a “moral loss” (36, 169, 188).

Now, it seems that Broome himself concludes his own argument at line 5. He doesn’t, in the relevant text, try to work out just which part of NI fails.Footnote ²³ For my part, I consider the argument compelling – indisputable, I thinkFootnote ²⁴ – all the way through to the intermediate conclusion that we find on line 5.

MacAskill, in contrast, continues the argument. A notable plus of doing so is that, by the time we get through line 6, we’ve nicely avoided the inconsistency we see in line 4.

But there’s a problem. It’s not always recognized that, once we get through line 5, we have options for avoiding the inconsistency that go well beyond line 6. In other words: line 5 on its own doesn’t compel us to accept line 6. Nor does logic on its own compel us to accept line 6. Once we show in line 5 that NI is false, we, in other words, have options: we can reject Part I of NI, or we can reject Part II, or we can reject both.

What is it, then, that does compel us, given that we accept line 5, to move to line 6? What compels us to avoid the inconsistency by accepting Part II of NI and rejecting Part I? Why not reject Part II? And retain Part I – that is, the more intuitive part of NI, the very part that explains why Narveson’s famous witticism remains so widely quoted even today – and seems to come very close to something very important?

Nothing in the argument itself answers that question. Once we get to line 5, we can’t – absent additional work – validly move to line 6.

For MacAskill’s part, the additional work that completes the argument comes in the form of an examination of a handful of population theories – the “average view,” the “total view,” the “critical level view” (177–186). Then, without purporting to have resolved the issues each such view independently gives rise to, he concludes that the “balance of arguments” favors the position that, informally, “bigger is better,” and, more formally, that (as noted above) “yes, it is a loss if future [happy] people are prevented from coming into existence” (186; 189–190).Footnote ²⁵

Thus the additional work that aims to justify the inference from line 5 to line 6. But MacAskill himself recognizes that work as incomplete (177). That we can get from that work – in combination with line 5 – to line 6 thus remains more a matter of conjecture than of reasoned argument.

What form might that conjecture take? What principle, in other words, is quietly being put to work in advancing the argument from line 5 to line 6?

Plausibly, the principle that may be lurking in the back of many theorists’ minds, whether or not in MacAskill’s own – to justify the inference from line 5 to line 6 is the mere addition principle.Footnote ²⁶ The modest idea that the mere addition of a happy person doesn’t make things worse certainly may seem perfectly innocuous. It might seem the kind of principle that doesn’t even bear stating – a principle that we almost have no choice but to accept as a legitimate, if unstated, premise of the argument.

In fact, however, the mere addition principle will seem innocuous to us only so long as we continue to operate under the assumption that the value of existence controversy is to be resolved in one particular way – in the existence-insensitive way, that is, against Narveson. We are obviously happy to accept the principle that the existence of the additional happy person doesn’t make things worse so long as we continue to operate under the assumption that the existence of the additional happy person in fact makes things better. Once, however, we identify that particular way of resolving the controversy as the mere assumption that it is – once, that is, we recognize as an option the existence-sensitive way of resolving the controversy – we see that the mere addition principle, without more, can’t be relied on to justify the inference from line 5 to line 6. Thus the door is open to our accepting line 5 and rejecting line 6.

Interestingly, there’s a deeper point here: what, with Broome’s help, we can understand the stated argument to show is that the seemingly innocuous mere addition principle is actually a wolf in sheep’s clothing. Once we are on the hook for the idea that, other things equal, the mere addition of the happy person can’t make things worse, we are then on the hook for the highly contentious intermediate conclusion we see in line 7 – otherwise known as the Pareto plus principle, the principle that, other things equal, the mere addition of the happy person doesn’t merely not make things worse, but actually makes things better.

But let’s go back. I’ve claimed that, once we get through line 5, we have options beyond simply accepting line 6. To see that that’s so, consider what the existence-sensitive proposal for resolving the value of existence controversy would have to say about Broome’s case. More simply, what does the EC have to say about that case?

EC quickly instructs that f1, being worse for no one who does or will exist in f1, isn’t worse than f3. Nor, according to EC, is f3 worse than f1 (f3 is, after all, maximizing across all accessible worlds for Charlotte). Conceptual principles then support the result that f1 is exactly as good as f3.Footnote ²⁷ We can then apply the Pareto principle that Broome himself put to workFootnote ²⁸ to show that f2 is worse than f3. Given that result in combination with a further conceptual principleFootnote ²⁹ , we then infer that f2 is worse, not just than f3, but also than f1. The upshot: f1 is exactly as good as f3, and f2 is worse than both.Footnote ³⁰

The distinction between ordinary and existential losses is another way of articulating the existence-sensitive solution. Charlotte’s ordinary loss in f2 has full moral significance (she exists there, and is worse off than she is in f3), her existential loss in f1 – and I’ve already conceded that it is a loss – is devoid of any moral significance. Building on those points, we can say that her morally significant loss in f2 does count against f2, while her morally insignificant loss in f1 doesn’t count against f1. With no loss at all in f3, it’s plausible then to conclude as above: f1 is exactly as good as f3, and f2 is worse than both.

***

Thus

The fact that the mere addition principle is so widely accepted and may well be quietly at play in justifying the inference from line 5 to line 6, together with the fact that we now understand that we have options – that there exists an alternate, credible, way to avoid the inconsistency we see in line 4 that rejects the mere addition principle – is what makes it fair to call that principle a dogma.

To be clear: Narveson’s original formulation of his underlying intuition – that making happy people is morally “neutral” – fails. Broome’s argument, through line 5, makes that point: Narveson overshot. But what we’ve now seen is that we can reject NI and still retain the more intuitive part of Narveson’s underlying intuition: that making happy people doesn’t, other things equal, make things better. And we can now add that, on occasion, making happy people can actually make things worse.

3.3 The argument based on expected value

Standard expected value theory (SEVT) plays an important role in MacAskill’s thinking (36–40, 259–260). And it quickly generates its own argument in favor of the existentially unrestricted version of longtermism. Thus SEVT implies much of what MacAskill lists as being of special importance (9–28):

Indeed, MacAskill’s list easily could have been generated from SEVT and SEVT alone.

SEVT’s aim is to provide a way for us to take probabilities into account (to rationally place our bets). And probabilities are surely critical to moral evaluation.

What is in dispute is whether SEVT is the right way to take probabilities into account. It’s certainly a widely accepted add-on to many forms of maximizing consequentialism. It is fair to say it is the mainstream view.

Perhaps most salient for purposes here is that SEVT lends robust support to MacAskill’s own existentially unrestricted form of longtermism. As soon as he recognizes that a person who never exists at all but could have existed and been well off sustains a moral loss (36, 169), he is well on his way to making the “moral case for space settlement” (189) – not just for we who do or will otherwise exist, but also for all those many, many additional people we could work very hard (and potentially at great sacrifice to others as well as ourselves) to bring into existence as well.

But as widely accepted as SEVT is – among population ethicists, maximizing consequentialists and perhaps economists – it doesn’t itself come without its own challenges. It’s not a conceptual or morally obvious truth.

Things get worse. Though widely accepted, we can identify cases in which SEVT seems clearly to produce a false result and still other cases in which the result SEVT generates can seem plausible only to those theorists who have already assumed the existence-insensitive way of resolving the value of existence controversy.

As stated earlier, I believe that probabilities are critical to moral evaluation. What I now just want to note is that what probabilities are critical to is the evaluation of the choices we face – as, e.g., permissible. In contrast, many of the principles noted in part 3.2 above, e.g., the standard Pareto principle and the mere addition principle, though not EC itself, focus exclusively on ranking alternative accessible futures in respect of their overall betterness. Under conditions of certainty, the two sorts of moral analysis plausibly come together. But once we add probabilities into the mix – and specifically probabilities based on information available to agents prior to choice – it’s the choices that exist for us as options that we are aiming to evaluate rather than the accessible futures we are aiming to rank.

***

In the first case we need to consider – all-but-known disaster/additional-people version (Table 3) – agents must choose between A, rescuing a very large number of people — say, a billion people — from a terrible disease, and B, bringing a vastly larger number of additional people into existence in the very distant future.Footnote ³¹

Table 3.

All-but-known disaster/additional-people version

Pop. I contains 1 billion existing or future people;

Pop. II contains all those people and a great many additional people

Table 3 Long description

A table comparing two options, A and B, based on probability, accessible futures, and per person wellbeing. The table has three rows and four columns. The columns are labeled Option A, Option B, and two additional columns for populations. The rows are labeled probability, accessible futures, per person wellbeing, +100, and +0. For Option A, the probability is 1.0, accessible futures is f1, per person wellbeing is not specified, +100 is Pop. I, and +0 is Pop. II. For Option B, the probability is 0.999999999...9, accessible futures is f2, per person wellbeing is not specified, +100 is Pop. II, and +0 is Pop. I.

Let’s now stipulate that the Table 3 case is one-off; the pattern doesn’t repeat itself again and again at the particular future such that, if things don’t turn out well in an early trial, they may well turn out well over the long run.

If the numbers, including the probabilities, are sufficiently stark, then the expected value of choice B will be greater than the expected value of choice A. And the moral principles we can anticipate in light of SEVT, accordingly, will instruct that B is morally obligatory and that A is wrong.

Even MacAskill notes about a similar case that, “intuitively” for him, choice A, as the “safe bet,” is “the right thing to do.” (259). “Taking the low probability [of success] action seems wrong.” (259)

What MacAskill doesn’t do, however, is reject SEVT on those grounds. Rather, he writes that “[u]nfortunately, there is no good solution to this problem …. If we wish to avoid the idea that tiny probabilities of enormous amounts of value can be better than guarantees of merely large amounts of value, then we run into other problems that seem just as bad.” (259). And anyway, he notes, in the cases he wants to talk about in his book, the “probabilities under discussion aren’t at all very tiny” (259–260).

But that’s not enough. If the problems we run into themselves force us to seriously question SEVT, then we surely can’t rely on SEVT to justify MacAskill’s own existentially unrestricted form of longtermism.

Let’s just note that SEVT also condemns choices very like choice A in the additional-people version of all-but-known disaster in far more local, close-to-home cases as well, including where the choice is to have two children rather than ten or no children rather than two and where how the choice is made doesn’t affect anyone else who either does or ever will exist. As noted earlier, MacAskill explicitly notes that he doesn’t “think that we should scold those who choose not” to have children (188). But to say that that choice imposes a morally significant loss on the child, or that it makes the world morally worse (36, 169), is, I think, to “scold” those individuals. I think it also panics us.

As noted earlier, probabilities are critical to the moral evaluation of our choices. But the additional-people version of all-but-known disaster at least raises the question whether SEVT, however widely accepted, is the right way to take probabilities into the account. For those unwilling to bite the bullet, the case does more: it shows that SEVT isn’t the right way to take probabilities into account.

Moreover, we have options. One option would be to insist that our theory reflect the existence-sensitive way of resolving the value of existence controversy – that it, in other words, include EC as a constraint; that it recognize the morally critical distinction between ordinary losses and existential losses. Under such a theory, there is no moral pressure at all to choose B over A in all-but-known-disaster/additional person version: under choice A, the losses the members of population II sustain in f1 are each and every one sustained by persons who never exist in f1. They thus can’t, according to EC, make f1 worse than f2 or f3 or make the choice of A wrong. In contrast, the losses the members of population I sustain in both f2 and f3 are all sustained by persons who do or will exist. The upshot is that, consistent with EC, the door is open for still other moral principles to instruct that f2 and f3 are both worse than f1 and that the choice of B is wrong.

***

SEVT faces further challenges in the context of cases that don’t invoke the value of existence controversy at all. Such cases seem to function as bona fide counterexamples against SEVT. If SEVT can’t be relied on to generate correct results even in easy same-people cases, it surely can’t be relied on to extend support in the seemingly far more complex additional-people cases.

Thus consider the same-people version of all-but-known disaster (Table 4). In that case, the agent – say, the mom – has a one-off choice between A – doing the normal parental things for her child, Tommy – and B – having the very young Tommy take a high risk/high reward pill, one that comes with a truly significant possible upside for Tommy, an upside consisting of, say, a trillion (1T) units of wellbeing.

Table 4.

All-but-known disaster/same-people version

Table 4 Long description

A table comparing options A and B based on probability, futures, and wellbeing levels. The table has 3 rows and 3 columns. Column headers are Option A, Option B, and probability, futures, and wellbeing levels. Row labels are probability, futures, and wellbeing levels. Row 1: probability, 1.0, 0.999999999, 0.000000001. Row 2: futures, f1, f2, f3. Row 3: wellbeing levels, +1 trillion (1T), Tommy, -100, Tommy, +100, Tommy.

Then, the calculation under SEVT:

$$\eqalign{ {{\rm{EV}}\left( {\rm{A}} \right)} & = {\rm{ }}1 \times 100 \cr & = {\rm{ }} \!+ \!100 \cr {{\rm{EV}}\left( {\rm{B}} \right)} & = {\rm{ }}\left( {0.999999999 \times \!\!-\!100} \right){\rm{ }} + {\rm{ }}\left( {.000000001 \times 1{\rm{T}}} \right) \cr & = {\rm{ approx}}. - \!99 + 1000 \cr & = {\rm{ approx}}. + \!900 }$$

SEVT clearly supports the result that choice B is obligatory, whether performed in f2 or in f3.

But at least as performed at f2 choice B seems clearly wrong: the probabilities are overwhelming; we can almost see the miserable child; the hapless mom; the hideous outcome that, given the probabilities, the agent (the mom) all but knew all along that her choice of B would give rise to.

The ad hominem “risk averse” just doesn’t seem adequate here or appropriate. The risk neutral but rational evaluator will surely agree that B – specifically, that B at f2 – is wrong.

***

Thus I propose:

But again we have options – at least, it’s premature to conclude that we don’t have options. In the same-people version of all-but-known disaster, the existence-sensitive way of resolving the value of existence controversy isn’t of any help at all. What we need is another way altogether of taking probabilities into account – a theory that avoids the overreach that we seem to get (too many choices deemed obligatory, too few deemed permissible – but that still recognizes probabilities as critical to moral value.

In reply to this same-people version of all-but-known disaster, perhaps MacAskill would again defend the SEVT analysis by noting that “[u]nfortunately there is no good solution” to the problem; perhaps he would again note that the probabilities he is interested in the book “aren’t at all very tiny” (259–260).

Just as important, MacAskill himself, as noted earlier, explicitly recognizes the challenges SEVT faces. It thus isn’t fair to say, of MacAskill personally, that SEVT functions as a “dogma” for him.

But those points don’t entirely solve the problem. If we accept the same-people version of all-but-known disaster as a bona fide counterexample against SEVT (even if we, though not I, perhaps concede that the additional-people version of the case isn’t a bona fide counterexample) – if we accept that SEVT generates false results in the same-people version of all-but-known disaster – then we can’t rely on SEVT going forward. Specifically: we can’t rely on SEVT to justify MacAskill’s own existentially unrestricted form of longtermism. Rather, the philosophically correct path is to go on to the next theory: to identify a principle that recognizes the relevance of what we can neutrally call anticipatory probability-related value and that tests better than SEVT does. We need, in other words, to identify a better way of taking probabilities into account. Once we have figured that out, we can construct our argument – for or against MacAskill’s existentially unrestricted form of longtermism, as the case may be. But until then we should hold our endorsement of MacAskill’s position in reserve.

Previously in this paper I’ve pointed out that we in fact have theoretical options – existence-sensitive options – that we can go to in place of the existence-insensitive option MacAskill wants to leave us with. As noted earlier the existence-sensitive analysis doesn’t help us to identify any particular way of analyzing the same-people version of all-but-known disaster that would generate results that are more obviously credible than the seemingly clearly false results generated by SEVT. But that fact doesn’t undercut what I said earlier: once we have identified a bona fide counterexample, the philosophically correct path forward is to go on to the next theory – whether that theory constitutes some variation on SEVT or offers some other way of taking probabilities into account altogether – and then the next, until we arrive at one that tests better than SEVT does — that, at the very least, passes our most immediately credible tests, which SEVT does not.