ShouldWeWishWellToAllPenultimateDraft

Caspar Hare March 2016

DRAFT of a paper forthcoming in The Philosophical Review

Should We Wish Well to All?

1. Introduction

Some theories of the rights and wrongs of individual action (henceforth just moral

theories) have a curious feature. They say that sometimes it is wrong for you to act as you

would if you were reasonable and moved solely by individual concern for each one of the

people affected by your actions. In short: they generate conflicts between reasonable beneficence

and morality.

I want to do three things here. First I want to present some examples of moral

theories that generate conflicts between reasonable beneficence and morality (this is the

business of Sections 2 and 3). Second I want to present an argument to the conclusion that

all such theories are mistaken (Section 4). Third I want to suggest that this raises some very

interesting further questions about the nature of beneficence (Sections 5 and 6).

2. Socially Progressive Moral Theories and Reasonable Beneficence

Our first example involves moral theories that direct you to pay special attention to the

less-well-off (henceforth progressivist moral theories, for short). Consider a very simple, very

extreme progressivist moral theory, one that says that your single moral duty is to improve

Many thanks to two anonymous reviewers, and to Michael Caie, Stephen Darwall, Toby Handfield, Richard

Holton, Seth Lazar, Anna Mahtani, Michael Otsuka, Wlodek Rabinowicz, Agustin Rayo, Kieran Setiya, Larry

Temkin, Judy Thomson and Alex Voorhoeve for most helpful comments. More generally, thanks to audiences

at the London School of Economics, SUNY Buffalo, Vanderbilt University, the Philosophy Mountain

Workshop, and the University of Colorado Boulder. And special thanks to participants at the 2015 Ethics and

Decision Theory Workshop at the Australian National University.

the condition of the worst-off people affected by your actions (call it extreme progressivism).

And consider a case:

At the Doctor

You, the doctor, have twenty six patients – Abigail, Benny, Cofi,…, and Zeke.

You know them well. You and the twenty six know that one of the twenty six

has been infected by Maleficitus, a very nasty disease, while the remaining twenty

five have been infected by Tolerabalitus, a tolerably nasty disease. It is too early to

guess who has what, but you and they know that now is your only opportunity to

administer drugs, and that you have only two drugs at you disposal. One

provides mild relief for Maleficitus sufferers, temporarily taking the edge off

some of the symptoms. The other totally cures Tolerabalitus. But the drugs

interact dangerously. Nobody wants to be taking both.

What should you do? If your single moral duty is to improve the condition of your worst-off

patient then you should give them all the Maleficitus-relieving drug. That way none of them

has to suffer through unrelieved Maleficitus. But each patient, insofar as he or she is

reasonably tolerant of risk, does not want you to do that! Each patient, insofar as he or she is

reasonably tolerant of risk, would take a 25/26 chance of cured-Tolerabalitus and a 1/26

chance of unrelieved-Maleficitus, over a 25/26 chance of uncured-Tolerabalitus and a 1/26

chance of (slightly) relieved-Maleficitus. And for you, the doctor, insofar as your concern for

others involves reasonable toleration of risk to them, your concern for Abigail recommends

giving out the Tolerabalitus-curing drug, and your concern towards Benny recommends

giving out the Tolerabalitus-curing drug, … and so on for all your patients.

So in this case extreme progressivism says that is wrong to act as you would if you

were reasonable and moved only by individual concern for each of your patients. It

generates a conflict between reasonable beneficence and morality.

Extreme progressivism is a very uncompromising theory. Only the worst off matter.

Few real philosophers accept that. Do less extreme, more plausible versions of progressivism

generate conflicts between reasonable beneficence and morality? This is a tricky question to

answer. Any progressivist theory specifies a certain way in which you should pay special

attention to the less well off. It avoids the conflict if, in situations in which you are uncertain

about who stands to gain or lose from your action (situations like At the Doctor), it would

not be unreasonable for you, when acting out of concern for a particular person, to pay

special attention in just the same way to possibilities in which things go less well for that

person.

So, in effect, making a judgment about whether a progressivist theory generates the

The example has played an important role in discussions of progressivism. In brief: Parfit (1997) distinguished

two kinds of progressivist theory, welfare egalitarianism, which says that equality of welfare is in itself good, and

welfare prioritarianism, which says that incremental benefits to worse-off people have greater moral significance

than equivalent incremental benefits to better off people. He then pointed out that welfare prioritarianism,

unlike welfare egalitarianism, is not subject to a ‘leveling down objection’ – it does not say, implausibly, that we

can make things in a morally significant way better by making everybody worse off). Rabinowicz (2001) then

pointed out that prioritarianism is subject to a seemingly-similar-in-spirit objection: Though it does not say that

sometimes we should act contrary to the interests of all, it does say that, in cases like At the Doctor, we should

act contrary to the expected interests of all. There has been much subsequent discussion about whether this is a

bad feature of the view. See Otsuka and Voorhoeve (2009), Parfit (2012), Otsuka (2012) and Otsuka (2015).

So, for example, a welfare prioritarian can avoid the conflict if she says that 1) acting morally consists in acting

so as to maximize expected total priority-adjusted well-being, and that 2) acting from reasonable concern for a

person consists in acting so as to maximize her expected priority-adjusted well-being. (Though there is then a

question about whether she should really count as a ‘prioritarian’. If these ‘priority adjusted well-being levels’

matter both morally and prudentially then it is unclear why not to call them ‘well-being levels’ and be done with

it.) A welfare prioritarian faces the conflict if she says that incremental benefits to the less-well-off matter much

more, morally, than incremental benefits to the better-off, and that it would be unreasonable to adjust so

conflict involves making a judgment about just how risk-averse you can be, when acting out

of concern for a particular person, without being unreasonable. But I know of no very

convincing general proposals about just how risk-averse you can be on a person’s behalf,

without being unreasonable. We may all agree that it is unreasonable to be perfectly risk-

averse on a person’s behalf, to care only about minimizing the badness of the worst thing

that can happen to them. I do not keep my children indoors, swathed in Kevlar. I would

think you nutty if you did. But is it unreasonable to be moderately risk-averse on a person’s

behalf? If so, exactly what counts as ‘moderate’? Many philosophers who think carefully

about risk maintain a judicious silence on questions like this.

Our next example will be cleaner. It involves a theory that is plausible enough to

have been endorsed by many real philosophers. We do not need to make any assumptions

about the reasonability of risk-aversion in order to see that this theory generates conflicts

between reasonable beneficence and morality.

3. Constraints and Reasonable Beneficence

The single most famous and well-worn case from all of twentieth century normative

ethics:

Footbridge

While loitering on a footbridge above a trolley track, you see that a trolley is heading

towards a group of five people, immobile on the track. You can prevent their deaths,

heavily for well-being level in assessing risks to yourself or people you care about. In short: she faces the

conflict when she is more left-wing than she is personally risk-averse.

Notably Lara Buchak in Buchak (2013). In Buchak (manuscript.) she outlines some restrictions on reasonable

risk aversion.

but only by pushing a person, presently beside you, down onto the track, into the

path of the trolley.

As every sentient creature in the world now knows, the large majority of people who are

asked about this case say that you ought not to push. Their judgment is endorsed by moral

theories (henceforth constraint-against-bad-killing theories, for short) that say that, for a certain,

bad kind of killing, we ought not to kill in this way, even when killing in this way brings

about what utilitarians would call ‘the greater good’. What is the ‘certain, bad kind of killing’?

Roughly: you kill in the bad way when the death of the person you kill causes the greater good

to come about (getting less rough about this is a major, unfinished task in moral theory

We can easily fill in the details of the Footbridge case so as to show that constraint-

against-bad-killing theories generate conflicts between reasonable beneficence and morality.

Footbridge With Suitcases

The six people involved here are Alexia, Benny, Cate, Dora, Emilio and Frederica. You

know them all and care about them all. Why are they all so immobile? Why can’t the

five just move off the track? Why will the one allow you to push him or her? Because

the six people are trapped in six suitcases, one person per suitcase. The suitcases were

shuffled a few moments ago. You have no idea who is where. Nobody has any idea

who is where.

Constraint-against-bad-killing theories continue to say that you should not push (for this

remains the bad kind of killing – the death of the one will cause the saving of the five), but

Thomson (1985).

The classic challenge is to precisify the theory so as to get the result that it is wrong to kill in Judy Thomson’s

Footbridge case, okay to kill in Judy Thomson’s Switch and Loop cases. I should note that Thomson herself

now rejects this challenge. Her final view, in Thomson (2008), is that all killing in cases like this is forbidden.

now each of the six wants you to push! Each of the six would take a 5/6 chance of living

and a 1/6 chance of being crushed by a trolley (and saving five people by being crushed),

over a 5/6 chance of being crushed by a trolley (and saving nobody by being crushed) and a

1/6 chance of living. And for you, insofar as you care about them, your concern for Alexia

tells in favor of pushing, your concern for Benny tells in favor of pushing,… and so on for

all of them.

Note that this time, in saying that each person, insofar as she has reasonable self-

concern, will want you to push, and that you, insofar as you care about each person, will

want to push, we are making no assumptions about the extent to which it is reasonable to be

risk-averse on your own behalf and on behalf of the people you care about.

4. The Argument From Composition

So what? Is it a problem for a theory that it says that sometimes you should act

against reasonable concern for all people involved? What should we think about cases like At

the Doctor and Footbridge with Suitcases? Well, in both these cases my first inclination is to

say that it is morally fine to act out of reasonable concern for all people involved. Go ahead

– give out the Tolerabalitus-curing pills, push the suitcase on the track. Moral theories that

say otherwise are mistaken.

Some people have shared my first inclination, notably John Harsanyi.

But, oddly, I

find that many contemporary philosophers do not. They say that the important question is

not whether I can cast my action as an expression of concern for all, given what I now know

(in now common terms: whether it is ex-ante justifiable to all), but whether I could cast my

See Harsanyi 1955 and Harsanyi (1976).

action as an expression of concern for all, if I knew everything about its consequences (in

now common terms: whether it is ex-post justifiable to all)

And they have further things to say in support of their view. Derek Parfit, writing

about progressivist moral theories and cases like At the Doctor, suggests that if we find the

idea that it is morally fine to act out of reasonable concern for all appealing, it because we are

making a particular mistake. It might be tempting to think “Of course you should do what

everybody involved reasonably wants you to do. It is in everybody’s interests that you do

that. And we are never morally obliged to act against everybody’s interests.” But that is not

right. Though all of your patients reasonably want to be given the Tolerabalitus-curing drug,

it does not follow that it is in all of their interests to be given the Tolerabalitus-curing drug.

One of them has Maleficitus. It is in his or her interests that you give out the Maleficitus-

relieving drug. And something similar could be said about Footbridge With Suitcases. It

might be tempting to think “Of course you should push. It is in everybody’s interests that

you push.” But that is not right. Though all of the entrapped people reasonably want you to

push, one of them is on the footbridge. It is in his or her interests that you hold back.

And Marc Fleurbaey and Alex Voorhoeve, also writing about progressivist moral

theories and cases like At the Doctor, suggest that if we think that ex ante justifiability is

what matters then we will be liable to predictable changes of mind in response to new

information. So you, the Doctor, not knowing who has what disease, will want to give

everyone the Tolerabalitus-curing drug, but you will know that you would not want do this if

These terms are entrenched, but a bit misleading. The intended contrast is not between what I can say

beforehand and what I can say afterwards. It is between what I can say, given what I know, and what I could say if I

knew more.

Parfit (2012).

you knew who had what disease. You will then have an aversion to information that is highly

relevant to the problem at hand. Such aversions are irrational.

I am not persuaded by either of these arguments. So far as mistakes go, of course

somebody might mistakenly assume that it is in everybody’s interests that, e.g., you push. But

I do not think that I am assuming that. I think that I am well aware that there is someone in

the suitcase, and that he or she will die if you push. And yet the thought that you should

push remains appealing to me. And, so far as aversions to information go, I do not think

that aversions of the sort Fleurbaey and Voorhoeve describe need be irrational. In cases like

this, where things of incommensurable value (like people’s lives) are at stake, it makes

perfect sense to resist the judgments of your better informed self.

So is there anything to be said on the other side, beyond just insisting that it is

obvious and right?

Happily I think there is. I think there are at least three strong arguments

to the conclusion that theories like extreme progressivism and constraint-against-bad-killings

are mistaken, that you ought to act out of reasonable concern for all in some cases in which

these theories say otherwise.

Two of the arguments I am not going to talk about in any detail here. The first is the

argument from presumed consent. Roughly: Suppose you consult the people involved (talking to

Fleurbaey and Voorhoeve (2013).

See Hare (2010) and Hare (2013) Chapter 3.

In Harsanyi’s (1976) axiomatization of utilitarianism, he proposes an axiom that effectively ensures that

morality and beneficence will not come apart: ‘Axiom 3: Pareto Optimality. Suppose that at least one individual

j(j=1,…,n) personally prefers alternative A to alternative B, and that no individual has an opposite personal

preference. Then individual i will morally prefer alternative A over alternative B.’ Glossing the axiom, he says

only: ‘Axiom 3 is a very weak and hardly objectionable moral postulate.’ But, as John Broome pointed out in

Broome (1987), once we understand that ‘is personally preferred to’ and ‘is morally preferred to’ play the role in

Harsanyi’s theory that ‘is better for an individual than’ and ‘is better than’ play in standard utilitarian theory, and

once we understand that the ‘alternatives’ to which he refers are uncertainty-weighted prospects, not outcomes,

this axiom no longer appears weak or ‘hardly objectionable’. It is the sort of thing that needs to be argued for.

them face to face in the doctor case, via mobile phone in the footbridge case). What will they

say? The patients will all say they want you to give out the Tolerabalitus-curing drug. The

people trapped in suitcases will all say that they want you to push. Wouldn’t it be okay for

you to push then? If so, wouldn’t it be okay for you to push in our case, on grounds that

they would request it if asked?

The second is the argument from dirty hands. Roughly: in a case like Footbridge With

Suitcases, if you willingly refrain from pushing and you are rational then you must prefer

outcomes in which everybody, including you, is worse off over outcomes in which

everybody, including you, is better off. But it is indecent to have preferences like that.

The third argument, the argument I want to talk about in detail here, has to do with

composition. The basic idea: Break the action we looking at up into proper sub-actions, each of

which only influences the interests of one person. Then argue that you ought to do each of

these sub-actions, appealing to the idea that, if you know that an action affects only one

person’s interests, then it is morally permissible to act out of concern for that person. Then

infer, by way of way of ought-agglomeration, that you ought to do all of the sub-actions.

Now for the detail. Let’s begin with a warm up. Here is an argument to the

conclusion that in a certain case (not exactly Footbridge With Suitcases) you ought to kill

one person by pushing him or her under a trolley, so as to prevent five people from being

crushed by trolleys. I will try to set it out as clearly as I can, in austere premise-and-

conclusion form. If you reject the conclusion then it will be interesting to see which premise

you reject.

Frances Kamm explores this sort of argument in Chapter 11 of Kamm (2001).

I discuss this in Hare (2013) Chapter 4.

Six Tracks

You see six tracks. On each track is a trolley, heading toward a pile of five

suitcases. Above each track is a footbridge, on which is perched one suitcase.

You can stop each trolley, but only by pushing the suitcase on the footbridge

above it into its path. You learn that Alexia is in a suitcase in the vicinity of

the first track (henceforth Alexia’s track) – either in the suitcase on the

footbridge (henceforth Alexia’s footbridge) or in one of the five suitcases on

the track. The remaining five suitcases in the vicinity of Alexia’s track are full

of sand. Likewise, Benny is in a suitcase in the vicinity of the second track

(henceforth Benny’s track), with the remaining five suitcases in the vicinity of

the Benny’s track filled with sand… and so on for Cate, Dora, Emilio and

Frederico. Furthermore you learn that one and only one of them, you have

no idea who, is on a footbridge.

The first premise:

P1 In Six Tracks, irrespective of what you do later (with respect to pushing the

other five suitcases), you ought to push the suitcase on Alexia’s footbridge

down onto Alexia’s track. And, irrespective of what you did earlier and will

do later (with respect to pushing the other five suitcases), you ought to push

the suitcase on Benny’s footbridge down onto Benny’s track… and so on.

(A clarificatory aside: I am working with a subjectivist notion of oughtness here –

according to which what you ought to do is intimately connected with your beliefs about

what will happen if you do one thing or another. This may make you pause. Some

philosophers insist that the term ‘ought’ is ambiguous between this subjective reading and an

objective reading – according to which what you ought to do is intimately connected with

what will happen if you do one thing or another. If you are one such person then please

substitute ‘subjectively ought’ for ‘ought’ in all that follows. Some other philosophers say

that ‘ought’ only supports an objective reading. Instead of saying ‘you subjectively ought to

do this’ we should say ‘it is rational for you to do this’, ‘it is sensible for you to do this’, or,

when things of moral significance are at stake, ‘if you were decent then you would do this.’ If

you are one such person then please make these substitutions in all that follows.)

Clarificatory aside over, why accept P1? Well, let’s attend (without loss of generality)

to that first suitcase, the one on Alexia’s footbridge. You know that pushing it will affect

Alexia and nobody else. If you push it then there’s a 1/6 chance, from both your and her

point of view, of your killing her and a 5/6 chance of your saving her life. Those odds sound

good to you and her. Why not push?

A reply: ‘Because then you run a risk of killing her!’

True, but this exactly the sort of thing that a brave surgeon does. She operates on her

patient for his own sake, knowing that there is a small chance that she will kill him and a

larger chance that she will save his life. We have reached the point where I have no further

argument, but am I happy to resort to strong language. It is absurd to say that it is never

appropriate to run a risk of killing a person when we intervene solely for that person’s own

sake.

A more sophisticated reply: ‘What matters is keeping your confidence that you kill

somebody at some point acceptably low. If you knew that you weren’t going to push any of

the other suitcases, then you could push now while keeping your confidence that you kill

somebody at some point acceptably low. Then you ought to do it. But once you know that

you have pushed in the past or that you will push in the future, you cannot push now

without allowing that confidence to rise unacceptably high. In the extreme, once you know

that have pushed all five other suitcases, then you cannot push now without becoming near

certain that you kill somebody at some point. That makes it okay to refrain from pushing.’

But again, the brave surgeon does not just operate on one patient. She operates on

many patients. And as she operates on more and more, her confidence that she kills

somebody at some point rises and rises. Does that change her moral obligations to her later

patients? The suggestion seems to be that she should get her medical degree and perform

one or two risky operations. Then she can retire, leaving future risky operations to be

performed by new recruits whose slates are clean.

Again, I am happy to say that this is

absurd.

If you are not yet convinced then please ask yourself what you would do if Alexia

were someone you cared about. Really. Imagine she is your daughter, or mother, or sister, or

old friend. You know that she is in one of the six suitcases in the vicinity of the first track.

You (and she) think it 5/6 likely that she is in one of the five suitcases on the track, in the

path of the trolley. You know there is nobody but you and Alexia anywhere near the track…

What would you do? I would push. And though I would feel vexed about it, though I would

chew my fingers in horror at the thought that I might be killing her, I would not feel guilty

about it. This is just between me and her, and I am acting out of concern for her interests as

best I can. That is what I ought to do.

The second premise is a general principle concerning the relation between the

deontic status of actions and the deontic status of their proper parts. Call it Weak

Agglomeration:

Thanks to Frank Jackson for making a close variant of this point to me.

P2 For any composite action A

…A

, if irrespective of whether you do A

through A

, you ought to do A

, and… and irrespective of whether you do

…A

n-1

, you ought to do A

, then you ought to do A

…A

Why accept P2? I think it compelling because I think that it is the job of a moral

theory to give you a package of advice, and a package of advice that is followable in the

following sense: there is something that you can do such that, if you were to do it, then you

would have done everything that the theory says you ought to do. Any theory that violates

Weak Agglomeration does not give us a package of advice that is followable in this sense. To

that extent it is defective.

(Another clarificatory aside: The ‘Weak’ is in there to distinguish it from a principle

that is controversial in the theory of practical rationality, Strong Agglomeration:

For any composite action A

…A

, if you ought to do A

, and… and you

ought to do A

, then you ought to do A

…A

Philosophers who call themselves actualists reject Strong Agglomeration.

They say that it

fails in cases where A

is the best composite action you can do, but doing either A

-but-

not-A

or A

-but-not-A

is disastrously bad (worse than doing neither A

nor A

), and you

know that, although you can do A

, as a matter of fact if you do A

then you will not do

. In such cases you ought to not-do-A

, and you ought to not-do-A

, but you ought to do

The canonic statement of actualism is in Jackson and Pargetter (1986).

But actualists (fortunately for them, I think) need not reject Weak Agglomeration,

nor need they reject the idea that their job is to give a followable package of advice. In cases

where Strong Agglomeration fails, according to the actualist, although you ought to do A

and ought to not-do-A

, there is something (A

) such that if you were to do it then you

would have done everything that you ought to have done – if you were to do A

then it

would be the case that you ought to do A

, because it would be the case that if you were to

do A

then you would do A

Aside over, the conclusion follows:

C In Six Tracks you ought to push all the suitcases – knowingly killing one

person and saving five.

Fine, but Six Tracks is not exactly like Footbridge with Suitcases, the case we started

with. And maybe one difference is important. In Six Tracks the death of the person you kill

is not a cause of the saving of the five. If the person had not been killed then the five would

still have been saved. Maybe this is not the bad, constrained-against kind of killing. So maybe

a constraint-against-bad-killing theorist can concede that is okay to kill in this way.

Finding a case in which the action of killing-the-one-and-saving-the-five decomposes

into independently performable sub-actions, each of which affects one and only one known

person (as in Six Tracks), but in which the killing of the one causes the saving of the five (as

in Footbridge With Suitcases), is a little trickier. But it can be done. Here is an example:

Two Parallel Tracks – Six Buttons

You see two tracks, running in parallel. On each track is a trolley, heading

toward a pile of five suitcases. Above each track is a footbridge, on which is

perched one suitcase. You learn that Alexia, Benny, Cate, Dora, Emilo and

Frederica are in the six suitcases in the vicinity of the first track, but you have

no idea who is where. And you learn that the six suitcases in the vicinity of

the second track are full of sand. You cannot stop the first-track-trolley, but

you can press any combination of six buttons – marked ‘Move Alexia’, ‘Move

Benny’… and so on. By pressing the ‘Move Alexia’ button you move Alexia

from her position on the first track to the corresponding position on the

second track (though you don’t see the move). By pressing the ‘Move Benny’

button you move Benny from his position on the first track to the

corresponding position on the second track (though you don’t see the

move)… and so on. By pressing any of the buttons you push the suitcase on

the footbridge on the second track, and thereby stop the second trolley (if

you press more than one button then this is symmetrically over-determined

by your button-pressing).

In this case, by pressing all six buttons you move everybody from the first track to

their corresponding position on the second track, and push the person on the bridge

into the path of the second trolley, killing him or her in such a way that his or her

death (or something very close to it – the impact of the trolley against his or her

body) causes the trolley to stop and the five to be saved. This appears to be the bad,

constrained-against kind of killing, but the argument from composition can be

applied to this case too:

P1 In Two Parallel Tracks – Six Buttons, irrespective of what you do later (with

respect to button-pushing), you ought to press the ‘Move Alexia’ button.

And, irrespective of what you did earlier and will do later (with respect to

button-pushing), you ought to press the ‘Move Benny’ button… and so on.

P2 For any composite action A

…A

, if irrespective of whether you do A

through A

, you ought to do A

, and… and irrespective of whether you do

…A

n-1

, you ought to do A

, then you ought to do A

…A

C In Two Parallel Tracks – Six Buttons you ought to press all the buttons –

knowingly moving all six people to the lower track, and pushing one of them

into the path of the trolley in such a way that his or her death causes the

saving of the five.

And the motivation for P1 is much as before. You know that pressing (e.g.) the ‘Move

Alexia’ button will affect Alexia’s survival and nobody else’s. If you press it then there’s a

1/6 chance (from your and her point of view) of your killing her and a 5/6 chance of your

saving her life. Why not press?

A reply: maybe, if you are pressing no other buttons, then it you ought to press the

‘Move Alexia’ button. But suppose that you are pressing other buttons. Then, if you press

the ‘Move Alexia’ button, there’s a 1/6 chance of your killing her in the bad way – killing her

in such a way that her death causes the saving of the five. That changes things. It is never

true that you ought to risk killing someone in the bad way.

One might say that, but I find it very hard to resist this now-familiar thought:

because you know that pressing the ‘move Alexia’ button has no impact on the well-being of

anybody other than Alexia, the decision to press or not is in a certain way private – it is

between you and her. She would take the 1/6 chance of being killed and 5/6 chance of

being saved. It doesn’t matter to her whether the being-killed involves being-killed-in-the-

bad-way (her death saves lives) or being-killed-in-the-okay-way (her death saves no lives). If

anything, she would reasonably prefer that her death saved lives.

And, insofar as you care

about her, you should feel the same, and act accordingly.

Imagine, again, that Alexia is your daughter, or sister, or mother, or old friend.

Would you press the ‘move Alexia’ button? Again I would press. Again I would feel worried

but not guilty. I would be moving her for her own sake, and only for her own sake.

We can call this thought, the thought that I am repeatedly appealing to here, defeasible

privacy: When you know that an action will affect the well-being of only one person, and you

know who that person is, then it is morally appropriate, in the absence of powerful defeating

considerations, to act out of reasonable concern for that person. What would be a powerful

defeating consideration? Maybe something like ‘this person has done horrific things and

does not deserve to reap the rewards of her doing those horrific things’ would do the job.

Maybe then it would be fine not to act out of concern for her. But, the thought is, when you

know that only one person has something at stake in your action, and you know who she is,

there is a strong moral presumption in favor of acting out of reasonable concern for her.

Any defeaters must be powerful. Considerations like, ‘if I do this then there is a chance that I

will harm her in the bad way’ or ‘the worst thing that could happen to her if I do this is

worse than the worst thing that could happen to her if I do not do this’ are much too weak.

That’s the argument. It applies just as well to extreme progressivism and our original

At the Doctor case, just so long as you, the doctor, are not giving out the drugs all in one go,

but to each patient individually. It applies quite generally to cases in which there is a

beneficent-to-all action (one recommended by individual reasonable concern for each person

This observation was made by Derek Parfit: “…when someone else could act in some way that would both

kill us but also save several other people’s lives, we would have no strong reason to prefer being killed as a side-

effect of the saving of these people’s lives rather than as a means. It would be in one way better to be killed as a

means, since our death would then at least do some good.” Parfit (2011) p. 365.

affected by your action) that decomposes into independently performable sub-actions, each

of which affects one and only one, known person, and each of which is recommended by

reasonable concern for that person. In such cases you ought to take the beneficent-to-all

action. Moral theories that say otherwise (e.g. some forms of progressivism and standard

forms of constraint-based deontology) are mistaken.

Stepping back from the details, we can think of the argument as offering a solution

to hard problem in moral theory. Hard problems in moral theory arise when there are prima

facie appealing sets of moral norms that turn out, on careful reflection, to be inconsistent. In

this case there is a prima facie appealing set of moral norms governing our interactions with

groups of people, expressed by platitudes like be fair, attend to the neediest first, don’t leave anyone

behind, don’t use some to benefit others. And there is a quite different, also prima facie appealing

set of moral norms governing our one-on-one interactions with individual people, expressed

by platitudes like be guided by concern for the person’s well being, act as she would reasonably want you to

act if she knew what you knew. But interact with enough individuals and you have interacted

with a group! To avoid inconsistency one set of norms must be restricted. I say it is the first

set.

5. Nearby Cases

Well and good, but just how far must the group-interaction norms be restricted? The

argument from composition applies when a beneficent-to-all action decomposes into

independently performable sub-actions, each of which affects one and only one, known

person, and each of which is recommended by reasonable concern for that person. What

should we say about nearby cases that do not satisfy this intricate condition? Do (e.g.) the

progressivist and deontological moral theories that fail when the condition is satisfied hold

up when the condition is not satisfied? These are the questions I will address in the

remainder of this paper. I will look at some nearby cases in this section. What I say about

them will be more speculative than what I have said thus far, but I hope that it will allow me

to frame an important and interesting problem.

One nearby case involves a beneficent-to-all action that does not decompose in the

right way. An example:

Two Parallel Tracks – One Big Button

Your situation is exactly as in Two Parallel Tracks, but this time there is just one big

button. Pressing it will have the same effect as pressing all six buttons in Two Parallel

Tracks.

Can we apply the argument from composition here? Well, in this case maybe we can say that,

by pressing the big button, you do six things – you move Alexia, you move Benny… etc.

And maybe we can still ask whether you ought to move Alexia, whether you ought to move

Benny… etc. But in this case moving Alexia involves moving them all, so, in order to decide

whether you ought to move Alexia, we need immediately to take a view about whether you

ought to move them all. The claim that you ought to move Alexia cannot then serve as a

premise in a persuasive argument to the conclusion that you ought to move them all. The

argument from composition is not going to work.

Nonetheless my inclination is to think that you ought to press in this case too. In

Two Parallel Tracks – Six Buttons you, the morally contentious person, press all six buttons

because you want to move all six people from their position on the upper track to the

corresponding position on the lower track. Why does it matter, morally, whether you get

what you want by pressing six buttons or one button? This seems to me to be too delicate a

thing to support so much moral weight.

To dramatize the point: Imagine that, in this case, you start off believing that the big

button is not wired to anything. Then you are told that pressing it will move Alexia from her

position on the upper track to the corresponding position on the lower track. You welcome

this news. You decide to press the button. And then you learn that it will do something else

too: it will move Benny from his position on the upper track to the corresponding position

on the lower track. Isn’t this further good news? After all, if the opportunity arose to move

Benny by pressing another button then you would take it. And you would take it because

you want to move Alexia and Benny – which is exactly what pressing the big button does.

Why should you care about whether you get what you want by pressing multiple buttons or

one button?

I do not yet see a very convincing reply to this.

Another nearby case is one in which they, but not you, know precisely whom you are

in a position to kill, whom you are in a position to save. An example:

Two Parallel Tracks – Six Buttons – Knowing Victim

Your situation is exactly as in Two Parallel Tracks – Six Buttons, except this time

each of the six knows his or her place on the track.

What should you do now? I say that, although you cannot anymore take solace in the

thought that each one of them wants you to press, the argument from composition still

applies. Ought you press, e.g., the ‘Move Alexia’ button? There’s a 5/6 chance that Alexia is

thinking “Please, please, press that button – it will save my life and affect nobody else.”

There’s a 1/6 chance that she is thinking “Please, please, don’t press that button – it will kill

me and affect nobody else.” Pressing remains the thing to do.

The really interesting nearby case is one in which you know precisely whom you are in

a position to save and precisely whom you are in a position to kill. What should you do now?

Obviously, if the action decomposes into independently performable sub-actions, as in Two

Tracks – Six Buttons, and you know which sub-action affects which person, then you should

save the five but leave the one alone. But what if the action does not decompose, or you

don’t know which sub-action affects which person? Consider:

Two Parallel Tracks – One Big Button – Known Victim

Your situation is exactly as in Two Parallel Tracks – One Big Button, except this time

you know that Alexia is the person on the upper bridge, the person you are in a

position to kill.

Two Parallel Tracks – Six Buttons – Known Victim – Unknown Buttons

Your situation is exactly as in Two Parallel Tracks – Six Buttons, except this time you

know that Alexia is the person on the upper bridge, and this time you do not know

which button moves which person. You just know that by pressing all the buttons you

will kill Alexia and save Benny, Cate, Dora, Emilio and Frederica.

In these cases pressing is no longer beneficent-to-all. What ought you to do?

The argument from composition will not establish that, still, you ought to press. But

this may not be a failing in the argument, because it may be that the deontic facts have

changed – it may be that it matters whether you know who is who.

For thinking that it is no longer the case that you ought to press: Now, because you

know so very much about Alexia and the rest, you have strong reasons not to press. These

are considerations to do with Alexia’s mother and father, to do with her career, to do with

her funny impressions of old philosophers etc. And you have strong reasons to press. These

are considerations to do with Benny’s son, Cate’s unfinished sonata etc. The reasons are

incommensurable. Faced with such incommensurability you need a tiebreaker, and a policy

of non-intervention in cases like this is as good a tiebreaker as any.

For thinking that you ought not to press: By pressing you express the belief that your

reasons not to press (to do with Alexia’s family, her career, her funny ways etc.) are

outweighed by your reasons to press. This is a mistake. Alexia is not fungible in this way.

And it is a mistake that diminishes us all.

If the second way of thinking is right then classical deontological theories remain

sound in their application to cases in which you know who is who. Is it right? Let’s put this

question to one side. (I have view about it. I am one of those people who thinks that, by

pressing, you need not be taking any more or less of a stand on the relative strength of your

reasons than by not-pressing. But many people do not share my view. This dispute is old and

deadlocked. I doubt we will settle it here.) Let’s just adopt the first, anti-aggregatative way of

thinking. Once you know who is who, it is at least no longer the case that you ought to

press.

This gives rise to a curious problem.

6. What is it to Know Your Victim?

Look hard at the Two Parallel Tracks – Six Buttons – Known Victim – Unknown

Button case. Why doesn’t the argument from composition apply to this case too? In this case

you may dub the person who will be moved by your pressing the first button, whoever he or

she is, ‘Oney’. And you may dub the person who will be moved by your pressing the second

button, whoever he or she is, ‘Twoey’… and so on. In this case, you know that pressing

button one affects only Oney, and, by pressing button one, you raise your confidence that

Oney will survive from 1/6 to 5/6.

So it would seem, by defeasible privacy, that you ought

to press button one. Likewise, you know that pressing button two affects only Twoey, and,

by pressing button two you raise your confidence that Twoey will survive from 1/6 to 5/6.

So it would seem, by defeasible privacy, that you ought to press button two… And so it

would seem, by weak agglomeration, that you ought to press all the buttons.

The natural answer is that, in this case, for each button, you now have a strong

reason not to press it. Yes, pressing button one raises your confidence that Oney will

survive, but pressing button one also reduces your confidence that Alexia will survive. That

matters.

But why can’t the same be said about Two Parallel Tracks – Six Buttons? In that case

you may dub the person on the upper bridge, whoever he or she is, ‘Bridgey’. Yes, pressing

button one raises your confidence that Alexia will survive, but pressing button one reduces

your confidence that Bridgey will survive. Why doesn’t that matter in just the same way? You

don’t know whether Bridgey is Alexia, Benny, Cate, Dora, Emilio or Frederica. But isn’t

Bridgey, in any case, a human being, with dignity, and worthy of respect?

To answer this question in a full and satisfying way we need an explanation of why

the cases differ. And we need an explanation that can be applied quite generally. I have said

that when you know nothing about the person you are in a position to kill, beyond that he or

Putting this thought in a precise way is tricky. On one understanding of propositions, supposing that Oney

is (e.g.) Alexia, the proposition that Oney will survive is the proposition that Alexia will survive, so it cannot be

that you have greater confidence in the proposition that Oney will survive than in the proposition that Alexia

will survive. We must put the thought in terms of modes of presentation, or metalinguistically (you are more

confident in the truth of the sentence ‘Oney will survive’ than in the truth of the sentence ‘Alexia will survive’).

On other understandings of propositions this is not so. We can say there are two propositions in which you

have different levels of confidence. For present purposes I don’t think it matters which way we go.

Eizabeth Harman asks a question very like this in her (2015) review of my (2013) book.

she is one of the six, you ought to kill. When you learn that the person you are in a position

to kill is on a bridge, still you ought to kill. When you learn that the person you are in a

position to kill is Alexia, and you know a very great deal about Alexia, it is not the case that

you ought to kill. What triggers the change? Consider:

Two Parallel Tracks – Six Buttons – Poet Victim

You learn that the person on the footbridge has a secret talent for balladic poetry.

Ought you to press now?

Two Parallel Tracks – Six Buttons – ‘Bugaboo’ Victim

You learn that the person on the footbridge was affectionately nicknamed ‘Bugaboo’

by his or her mother.

Ought you to press now?

Two Parallel Tracks – Six Buttons – Fleshy-Eared Victim

You notice a crack in the suitcase on the footbridge. Sticking out of the crack is the

occupant’s red, fleshy ear. A bead of sweat drops from the helix into the cavum

conchae.

Ought you to press now?

Rather than answering these questions directly, let me offer a way of thinking about

them.

I said earlier that Defeasible Privacy applies when you know who the person you are

in position to affect is. A natural proposal, then, is this: It ceases to be true that you ought to

press at the point at which you know who your victim, the person on the footbridge, is. In

Two Parallel Tracks – Six Buttons – Known Victim – Unknown Button you know who your

victim, Alexia, is. In Two Parallel Tracks – Six Buttons you do not know who your victim,

Bridgey, is. What we need is a general theory of what it is to know who somebody is. Happily,

philosophers of language have worked for many years on this project (typically by way of

addressing the question: What is it to be capable of entertaining ‘singular thoughts’ about a

person?) So maybe our strategy should be to borrow their best theory, apply it case-by-case

and thereby find out what your obligations are.

I do not think this strategy very promising. For one thing, philosophers of language

have not arrived at anything like consensus on what it is to know who somebody is. For

another thing, it is not clear to me that the considerations to which philosophers of language

have been sensitive in accepting or rejecting this or that theory of singular thought have had

to do with the moral significance of singular thought. So it is not clear to me that their best

theory would go very far towards explaining why it matters so much, morally, whether you

know who your victim is. This talk of ‘knowing who the person you are in a position to

affect is’ should be regarded as shorthand for something with explanatory force.

Here is a strategy that I find more promising. We start with this proposal: it ceases to

be true that you ought to press when you have reasons to press, reasons not to press, and

you no longer have most reason to press. What we then need to do is to think about your

reasons to press, and about whether they are decisive.

In the background, I am thinking of reasons for acting a certain way as

considerations that count in favor of your acting that way. I am thinking of a reason as yours

when you are aware of it, and are in a position to be motivated by it. I am thinking that

sometimes you may have reasons to do one thing, reasons to do another, inconsistent, thing.

I am thinking that in some such cases there is most reason to do one or the other thing, in

other cases there is not. I am thinking that the latter is typical of cases in which there are

incommensurable values at stake – roughly when there are ways in which it is good if you do

the one thing, very different ways in which it is good if you do the other thing, and no fact

of the matter about it being overall better if you do either.

When you learn more about your victim you may thereby gain reasons not to press.

Suppose you have drawn up ‘Pro-Pressing’ and ‘Con-Pressing’ columns. Now suppose you

learn about the balladic poetry. That gives you something new to write in the ‘Con-Pressing’

column: ‘If I press then an important poetic voice will be stifled’. Now suppose you see the

sweaty ear. That gives you something further to write: ‘If I press then that ear will soon be

blue.’ As you learn more and more, the reasons against pressing mount up, and after a while

(any proposal as to exactly when will have an unseemly precision to it) your reasons to press

are no longer decisive.

But why doesn’t the knowledge that pressing will kill Bridgey already give you

powerful reasons not to press? After all, you know that, whoever Bridgey is, it will be in

certain ways worse if Bridgey dies and the others live, than if Bridgey lives and the others

die. Why can’t you dub these ways of being better ‘Bridgey-values’ and think to yourself

‘Pressing will be Bridgey-worse than not pressing’?

Of course you can think that to yourself, but I say that you still do not have a reason

not to press. To have a reason to do something you must do more than recognize that there

is a way in which it is better if you do it. You must be in a position to recognize the way of

being better, and thereby be moved to pursue it.

Consider a less vexed case, in which the values in play are of little moral significance:

Oprah or Opera?

In one envelope I place a ticket to Oprah (thrilling, a once-in-a-lifetime opportunity to

commune with the popular zeitgeist). In another envelope I place a ticket to the opera

(sublime, uplifting). I shuffle the two envelopes, place them before you and allow you to

pick one.

In this case, though you know that there is a way in which the first envelope is better than

the second envelope (the opportunity it gives you is either more thrilling or more sublime), I

say you have no reason to pick the first envelope over the second envelope. The

consideration ‘There is a way in which the first envelope is better’ is not a reason you have.

If you pick the first over the second, and I ask you what moved you to do that, you can’t

very well reply: ‘I was moved by the thought that there is a way in which it is better if I pick

it.’ You knew that the very same thing could be said for picking the second over the first –

there is a way in which it is better if you pick it! You do not have a reason until you

recognize the way in which it is better if you pick the first envelope. And just dubbing it

‘lefty-value’ will not help you recognize it. You need to open the envelopes.

Likewise, in Two Parallel Tracks – Six Buttons, the consideration ‘There are ways in

which it is better if I don’t press’, though true, is not a reason you have not to press. You

know the very same thing can be said for pressing – there are ways (more ways, in fact) in

which it is better if you press! You do not have reasons until you recognize the ways in

which it is better if you don’t press. And just dubbing them ‘Bridgey-values’ will not help you

recognize them. You need to learn more about Bridgey.

This is a background assumption of the most-reason argument for prospectist decision theory in Hare (2010)

and Hare (2013) Chapter 3.

7. Wrapping Up

In sum: Many moral theories tell us to act against reasonable concern for all in some

cases. The argument from composition shows that such theories are mistaken. Wish well to

all, and act on your wishing.

References

Broome, John. 1987. “Utilitarianism and Expected Utility.” Journal of Philosophy 84 (8): 405-

422.

Buchak, Lara. 2013. Risk and Rationality, Oxford: Oxford University Press.

Buchak, Lara. (Manuscript). “Taking Risks for Others and Aggregating Risks.”

Fleurbaey, Marc, and Alex Voorhoeve. 2013. “Decide as you Would with Full Information!

An Argument Against Ex Ante Pareto.” In Inequalities in Health: Concepts, Measures and

Ethics, ed. Nir Eyal, Samia Hurst, Ole Norheim and Dan Wikler, 113-128. Oxford:

Oxford University Press.

Greene, Joshua, R.B. Somerville, L.E. Nystrom, J.M. Darley, J.D. Cohen. 2001. “An fMRI

Investigation of Emotional Engagement in Moral Judgment.” Science 293: 2105-2108.

Hare, Caspar. 2010. “Take the Sugar.” Analysis 70 (2): 237-247.

Hare, Caspar. 2013. The Limits of Kindness. Oxford: Oxford University Press.

Harman, Elizabeth. 2015. “Review: Caspar Hare, The Limits of Kindness.” Oxford: Oxford

University Press. Ethics 125 (3): 868-872.

Harsanyi, John. 1955. “Cardinal Welfare, Individualistic Ethics, and Interpersonal

Comparisons of Utility.” Journal of Political Economy 63: 309-321.

Harsanyi, John. “1977. “Morality and the Theory of Rational Behavior.” Social Research 44:

623-656.

Jackson, Frank, and Robert Pargetter. 1986. “Oughts, Options and Actualism.” Philosophical

Review 95 (3): 233-255.

Jackson, Frank, and Michael Smith. 2006. “Absolutist Moral Theories and Uncertainty.”

Journal of Philosophy 103 (6): 267-283.

Kamm, Frances. 2001. Morality, Mortality Volume II, Oxford: Oxford University Press.

Otsuka, Michael. 2015. “Prioritarianism and the Measure of Utility.” The Journal of Political

Philosophy 23 (1): 1-22.

Otsuka, Michael. 2012. “Prioritarianism and the Separateness of Persons.” Utilitas 24 (3):

365-380.

Otsuka, Michael, and Alex Voorhoeve. 2009. “Why it Matters that Some Are Worse Off

Than Others: An Argument Against the Priority View.” Philosophy and Public Affairs 37

(2): 171-199.

Rabinowicz, Wlodek. 2002. “Prioritarianism for Prospects.” Utilitas 14 (1): 2-21.

Parfit, Derek. 1997. ‘Equality and Priority.” Ratio 10 (3): 202-221.

Parfit, Derek. 2011. On What Matters, Volume One. Oxford: Oxford University Press.

Parfit, Derek. 2012. “Another Defense of the Priority View.” Utilitas 24 (3): 399-440.

Thomson, Judith. 1985. “The Trolley Problem.” The Yale Law Journal 94 (6): 1395-1415.

Thomson, Judith. 2008. “Turning the Trolley.” Philosophy and Public Affairs 36 (4): 359-374.