II. THOMSON’S ARGUMENT

Thomson has offered the following argument in favour of the conclusion that diverting in Trolley, like harvesting in Transplant, is wrong. In rough outline, she first considers:

Second, Thomson hopes we will agree that option 2 is morally wrong in this case. Third, she observes that it is directly implausible that the wrongness of option 2 depends on the presence of option 1. So if we judge it wrong to divert the trolley into Bugsy when we have the option of diverting it into ourselves, we must judge it wrong to divert it into Bugsy in the original Trolley case where we lack this option.Footnote ⁹

At first sight it is unclear why Thomson's reader would agree with her that option 2 is impermissible in the three-option case. When I first read Thomson's paper I was convinced that diverting in Trolley was permissible, and her three-option case seemed to me equivalent to:

Here it seems clearly permissible to take option 2, even though this results in Bugsy's death. It would certainly be nice of you to take option 1, but, since there are no deontological constraints against simply saving one group rather than another, option 1 would be supererogatory. In the same way, if we think that there are no deontological constraints against diverting a trolley into one group rather than another, why should we think it wrong to divert the trolley into Bugsy rather than yourself?Footnote ¹⁰ Why not say that, while diverting the trolley into yourself would certainly be nicer than diverting it into Bugsy, this second option is still permissible?

Against this way of thinking about the case, Thomson suggests that saving the five would be a good deed, and appeals to the directly plausible idea that it is wrong to make others pay the costs of your good deeds instead of paying them yourself. By way of analogy she suggests that it would be wrong to donate to Oxfam by stealing money from others rather than sending one's own money.Footnote ¹¹ In categorizing the diversion of the trolley away from the five as ‘a good deed’, Thomson is emphasizing the plausible idea that neither you nor Bugsy are morally required to divert the trolley into yourselves in order to save the five. It is indeed plausible that it is wrong to force a cost on someone if she isn't morally obligated to assume it. There are, however, cases in which we do seem permitted to do things that result in someone's bearing a cost that she would not be obligated to assume herself. We seem permitted, in cases like Bazooka Holder's Three Options, to save ourselves rather than Bugsy even though Bugsy would not be required to save us rather than himself.

What really lends force to Thomson's claim is its invocation of deontological constraints against inflicting harm. The idea isn't that it's wrong for us to act in any way that results in someone bearing a cost she isn't obligated to assume voluntarily. The idea is that it's wrong for us to inflict a cost on someone that she isn't obligated to assume. This is a plausible idea, and it supports Thomson's conclusion. But the problem is that proponents of the idea that it is permissible to divert the trolley in the standard two-option case are committed to the view that, if there are any deontological constraints against inflicting harm, they do not apply to cases of diverting it or inflicting it as a ‘mere side-effect’. These proponents will not be impressed by the analogy to stealing to give to Oxfam (or more plausibly: harvesting someone else's organs rather than giving up your own) because, if they think this wrong, they will see it as an instance of the ‘problematic’ kind of infliction of harm that goes on in Transplant, rather than the ‘unproblematic’ kind of infliction that goes on in Trolley.Footnote ¹²

Thus, while Thomson's argument brings out some intuitive ideas that conflict with the intuition that it's permissible to divert in Trolley, philosophical defenders of a moral difference between Trolley and Transplant seem to have an escape-route. They can deny her intuitions about the three-option case and insist, as Thomson herself once did, that the plausible idea that we can't inflict costs on people who aren't obligated to assume them doesn't really apply to cases like Trolley. I believe that the only way to make the argument conclusive is to show that there is no place between Trolley and Transplant for these friends of a moral difference to hide.

IV. AN ELEVEN-STEP PROGRAMME FOR TROLLEY INTUITIONS

Recall the original Trolley case, in which the only way to save five children tied to a track is to divert the trolley about to hit them into Bugsy. Our first step is to ask whether it would matter if diverting the trolley caused a different trolley to hit Bugsy, as in:

It seems obvious that there is no moral difference between diverting the trolley in Trolley and Shifted Track. Surely it cannot matter whether diverting it kills Bugsy with the same trolley or a different one.

Some authors have, however, suggested that it is easier to justify inflicting harm by ‘diverting an existing threat’ than by ‘creating a new threat’, and it might seem that pulling the switch in Trolley merely ‘diverts an existing threat’ while doing so in Shifted Track ‘creates a new one’. But if a ‘threat’ is anything that will participate in an event or interaction that will cause harm, the threats to the five include not only the trolley, but the section of track that will guide the trolley into them. Pulling the switch in Shifted Track transfers the track-section from threatening the five with a trolley to threatening Bugsy with a trolley, and in this sense ‘diverts an existing threat’.

The only way to restrict the threats to the trolleys would be to construe ‘threats’ as something like things that will ‘directly’ cause harm, say by harmfully interacting with someone's body. But on this reading, you wouldn't count as diverting an existing threat in either Shifted Track or Trolley if there were stationary trolleys parked in front of Bugsy and the five, and the oncoming trolleys will kill by hitting these stationary trolleys and causing them to crush whoever is on the other side. Since it obviously cannot matter morally whether the trolleys themselves run over victims or smash into other trolleys that do this, it cannot matter morally whether you kill someone by ‘diverting a “direct” threat’ or ‘creating a new “direct” threat’.Footnote ²⁰ Proponents of a moral distinction between diverting and creating threats find it plausible that it is relatively easy to justify ‘arranging that something that will do harm anyway . . . shall do harm to fewer rather than more’.Footnote ²¹ But whether the thing in question does harm by interacting with victims’ bodies or causing other things to do so seems as obviously devoid of intrinsic moral significance as skin colour.

To confirm the lack of moral difference between diverting the trolley in Trolley and Shifted Track, suppose that you had three options: (1) divert the trolley headed towards the five onto a track with both Freddy and Suzy, (2) pull a switch that moves a track-section from connecting the trolley to the five to connecting a second trolley to Bugsy, or (3) do neither 1 nor 2. It seems indefensible to choose option 1 over option 2. How could you explain to Freddy and Suzy's parents why you killed both of them rather than only Bugsy, or no one at all? Or suppose that Suzy wasn't squarely centred on the track, so that in addition to killing Freddy, option 1 merely ran a 1/11 risk of killing Suzy, or injured her so badly that she dies seven years earlier than she otherwise would. Some fumbling suggestion about diversions of direct versus indirect threats is clearly an insufficient justification for killing an additional child, exposing her to a 1/11 risk of death, or depriving her of seven years of life.

Our second step asks whether it would matter if you connected and disconnected the tracks by means of different intermediaries, as in:

It seems obvious that there is no moral difference between disconnecting the first track and connecting the second in Shifted Track and Different Sections. Surely it cannot matter whether you do this with a single track-section or with relays that move different track-sections.

Pushing the button in Different Sections, unlike shifting the single track-section, does not seem to ‘divert’ anything from causing harm to the five to causing harm to Bugsy. But why should it matter whether, in benefiting the five and harming Bugsy, you shift some of the very same things from causing the deaths of the five to causing Bugsy's death instead? Some authors have characterized things that can be shifted from participating in interactions that will harm the five to participating in interactions that will harm Bugsy as ‘generalized perils’ or ‘public threats’.Footnote ²² But this falsely suggests that absent your intervention these things ‘threaten’ or will cause harm to Bugsy as well as the five. Thomson once suggested that we should view transferable causes of death as ‘bad [things], up for distribution’.Footnote ²³ But if we are permitted to ‘redistribute’ the harm of death from the five to Bugsy by making a track-section connecting a trolley to the five instead connect a trolley to Bugsy, why should we not be equally permitted to redistribute this harm by pressing a button that removes this track-section and connects a trolley to Bugsy with a different track-section? If, alternatively, we are not permitted to ‘redistribute’ death by pressing a button that connects a trolley to Bugsy and disconnects a trolley from the five by moving different track-sections, why should we view a single track-section as ‘up for distribution’ in a sense that permits us to shift it in order to achieve this very same result?Footnote ²⁴

On reflection, it seems directly implausible that transference of the selfsame physical object should make an intrinsic moral difference. Suppose we initially thought that flipping a switch would shift a single track-section from connecting a trolley to the five to connecting a trolley to Bugsy. Suppose we were about to flip the switch, when a proponent of the diverting–creating distinction burst onto the scene with allegedly urgent news: ‘Wait! Stop! If you flip the switch it won't transfer the same track-section from connecting a trolley to the five to connecting a trolley to Bugsy! Flipping the switch will cause a totally different track-section to connect a trolley to Bugsy!’ Surely it would be absurd for us to change our minds about whether to flip the switch in response to this information.

Frances Kamm has suggested a second moral difference between Shifted Track and Different Sections. According to Kamm, when you divert in Trolley, the events by which you harm Bugsy (like the trolley changing direction) don't actually cause the greater good of the five being saved; they instead have this greater good as their ‘non-causal flip side’. Of course, there are plenty of events that harm Bugsy and cause the five's salvation, like your arm moving, the switch moving, and electric current flowing. But these explain the salvation of the five only by explaining why the trolley moves. Kamm's idea seems to be that in Trolley (and Shifted Track), the event that most proximately explains both the good and the harm – the movement of the trolley (or track-section) – itself constitutes the five being free from threats. But in Different Sections the most-proximate-joint-explainer of harm and benefit (the depression of the button) does not constitute the freedom of the five from threats; it merely causes such freedom by causing the distinct event of the first track-section moving.Footnote ²⁵

On its face, the fact that an act inflicts harm on Bugsy and saves five by means of most-proximate-joint-explainers that constitute rather than cause the freedom of the five from threats looks as obviously devoid of intrinsic moral significance as skin colour. Kamm suggests, however, that the movement of the trolley in Trolley constitutes not only the freedom of the five from threats, but the greater good of the five being saved, and that this is morally important. It is, however, simply false that the movement of the trolley itself constitutes the benefits that count in favour of diverting it. The trolley's turning away from the five at time t benefits them by making it the case that rather than dying at t + 1, they live from t + 1 to t + n. But the benefit of their living from t + 1 to t + n is by no means identical to or constituted by the trolley turning away from them at t.Footnote ²⁶

More importantly, the fact that a most-proximate-joint-explainer of a harm and a benefit constitutes rather than causes the benefit itself also seems about as morally relevant as skin colour. Kamm claims that, when you harm B and benefit F by causing a most-proximate-joint-explainer that constitutes F's benefit, you merely substitute B for F as the party who is harmed, while if you harm B by causing a most-proximate-joint-explainer that causes F's benefit, you subordinate B to F (164–6). ‘Subordination’ sounds morally worse than ‘substitution’, but Kamm appears to be affixing relevant-sounding labels to manifestly irrelevant metaphysical distinctions. In the same way that the initial plausibility of the racist maxim ‘take care of your own kind!’ vanishes when we see that ‘own kind’ amounts to nothing more than shared skin colour, the plausibility of a Kammian distinction between ‘substitution and subordination’ seems to vanish when we see that it amounts to nothing more than a distinction between most proximate-joint-explainers that constitute rather than cause someone's being free from threats.

Kamm's attempt to connect her talk of substitution and subordination to directly plausible, non-stipulative ideas seems half-hearted and confused. Some of her remarks seem to conflate her distinction with others, like that between diverting and creating threats (discussed above) and that between harmful means that do and do not involve victims (discussed below).Footnote ²⁷ The rest hinges on her claim that we must not divert a trolley from hitting a tool that can be used to save many people to hitting one person, or use such a tool when this will harm someone, because this would imply:

The idea seems to be that if your most proximate harmful means of saving some people cause rather constitute their freedom from threats, the means act as a tool to which those harmed are subordinated, and given this they are subordinated to those saved by the tool. Why, however, should it matter whether your most proximate harmful means cause harm? Why not say that in moving your arm or pulling the switch in Trolley you subordinate Bugsy to your arm or the switch, and in turn subordinate him to the five you save with these things?

While talk of tools may lend superficial plausibility to the idea that worrisome subordination is present, suppose our choice was to save either Bugsy or a tool that we knew would save five who would otherwise die (say a needle with enough life-saving serum). Surely it would be permissible to save the tool, and preposterous that this involves some objectionable elevation of the tool or the five over Bugsy.Footnote ²⁸ In choosing to save the tool we are not fetishizing it as intrinsically valuable; the only sense in which the five ‘transmit their importance’ to the tool is that we should care instrumentally for its salvation out of our intrinsic concern for them. To say that an instrumental preference for pulling the tool rather than Bugsy's body out of the water ‘subordinates’ Bugsy to the tool and thence the five seems equivalent to the manifestly absurd claim that choosing to row right towards five drowning swimmers instead of left towards one subordinates the one to rowing right, which subordinates him to the five. Exactly similar remarks go for diverting the trolley from the tool into Bugsy.Footnote ²⁹

To confirm the irrelevance of these diverting-creating and Kammian distinctions, suppose that you actually had three options: (1) pull a switch that transfers a single track-section from connecting the first trolley to the five to connecting a second trolley to Freddy and Suzy, (2) push a button that activates two relay switches, one of which removes the track-section connecting the first trolley to the five, but the other of which connects a third trolley to Bugsy with a different track-section, or (3) do neither 1 nor 2. Surely it would be indefensible to choose option 1 over option 2. Could you really with a straight face try to justify your act to Freddy and Suzy's parents on the grounds that ‘had I killed Bugsy instead of both Freddy and Suzy, it wouldn't have been with any of the things that would have killed the five’, or ‘the most proximate means by which I would have done both good and harm would have caused rather than constituted the five's being free from threats’? These would be preposterous justifications for killing an additional child, or, if Suzy wasn't centred on the track, for exposing an additional child to a 1/11 risk of death or depriving her of seven years of life. Making baseless noises like ‘Bugsy would have been subordinated while Freddy and Suzy were merely substituted’ would not detract from the preposterousness.

Our third step asks whether it would matter if, instead of causing a moving trolley to hit Bugsy, the button caused an initially stationary trolley to hit him, as in:

It seems obvious that there is no moral difference between pressing the button in Different Sections and Start-Up. Surely it cannot matter whether the button causes an initially moving or an initially stationary trolley to hit Bugsy.

Michael Moore has, however, suggested that it is permissible to save five by doing something that diverts a moving trolley into Bugsy because ‘there is a big natural cause already in play’ and ‘your act makes a small [causal] contribution [to Bugsy's death] relative to [it]’, while it would be wrong to save the five by doing something that sets a stationary trolley moving into Bugsy, because your act would make ‘a larger causal contribution’ to Bugsy's death. This is surprising, because as Unger pointed out, this sort of ‘protophysical’ distinction between altering a moving object's direction and setting an object at rest in motion seems ‘not just ethically, but even physically . . . absolutely empty’.Footnote ³⁰

While relativization to the actor's frame of reference gives physical content to the distinction between moving and stationary trolleys, nothing will, I believe, make sense of Moore's attempt to distinguish morally between such cases in terms of whether your causal contribution to Bugsy's death is ‘small’. First, suppose that out of sheer indifference to Bugsy's interests you diverted a trolley from harmlessly moving down an unoccupied main track onto the side-track on which Bugsy was standing. It would be absurd to claim that your causal contribution to Bugsy's death was ‘small’ and that ‘you didn't do much’. But whatever the size of your causal contribution to Bugsy's death is supposed to be, it cannot be altered by the presence of the five on the main track. Second, suppose that to harvest Bugsy's organs and transplant them into the five, all you had to do was press a button that would divert an organ-harvesting robot from harmlessly moving down the hallway to entering Bugsy's room, where it will automatically remove his organs. It would be ridiculous to say that this – unlike wielding the scalpel yourself – is permissible because it ‘wouldn't do much’ and ‘would make a small contribution’ to Bugsy's death. Third, ‘the size of your causal contribution to Bugsy's death’ would presumably have to be something like the amount of conserved-quantity transfer or number of events you must cause to make something kill him. But suppose that in the original Trolley case there are massive forces holding the trolley on its course, and you must create enormous forces or Rube-Goldberg-like series of events to divert it. Suppose, however, that in Start-Up the second trolley is on frictionless, downward-sloping tracks, and the button starts its motion by giving it a tiny nudge. Now your causal contribution to Bugsy's death is greater if you divert the moving trolley than if you move the stationary trolley, but I doubt Moore wants to say this makes it wrong to divert the moving trolley but permissible to move the stationary one.

Whatever Moore wants, it is obvious that the amount of conserved-quantity transfer or number of events involved in killing someone is as devoid of intrinsic moral significance as skin colour. His baffling suggestion to the contrary seems to be a confused extension of the already dubious idea that it is easier to justify doing what you know will cause other agents to harm victims than it is to justify harming victims by means that bypass the decisions of others. When you cause others to harm victims, it can be tempting to ‘put the blame on’ the others, pleading that, after you acted, they could have chosen not to do harm. But it is just crazy to ‘put the blame on’ the trolley for killing Bugsy if you divert it into him; after you pulled the switch, the trolley most certainly couldn't have chosen not to run him over! Blaming the trolley instead of yourself for Bugsy's death clearly requires you to ignore the fact that absent your interference it would not have harmed him.Footnote ³¹

To confirm the irrelevance of Moore's distinction, suppose you had three options: (1) push one button that disconnects the first trolley from the five and connects a second moving trolley to Freddy and Suzy, (2) push another button that disconnects the first trolley from the five and sends an initially stationary third trolley into Bugsy, or (3) do neither 1 nor 2. It would be indefensible to choose option 1 over option 2. You could not seriously try to justify this to Freddy and Suzy's parents by saying ‘my causal contribution to Bugsy's death would have been big, but my causal contribution to Freddy and Suzy's deaths was small’. This sounds false, and would in any event be an absurd justification for killing an additional child or (if Suzy wasn't centred on the track) for exposing an additional child to a 1/11 risk of death or depriving her of seven years of life. Adding ‘I would have been to blame for Bugsy's death, but the second trolley is to blame for Freddy and Suzy's death’ would be outrageously absurd.

Our fourth step asks whether it would matter if the button that saves the five caused bombs to fall on Bugsy, as in:

It seems obvious that there is no moral difference between pressing the button in Start-Up and Dual-Action Bomber. It cannot matter that the button saves five and kills Bugsy through explosions instead of trolley movements.

One might suggest that because the bomber and the bombs, but not the trolleys, were constructed in order to destroy things, they have this as ‘their function’, so it is somehow ‘less of an accident’ if the rear bombs kill Bugsy in this case than if the second trolley kills him in Start-Up.Footnote ³² But surely it makes no intrinsic moral difference what anyone's intentions were in making the trolleys and bombs – it wouldn't change anything if we discovered that the trolleys were constructed for the purpose of running people over. One might also suggest that saving the five in Dual-Action Bomber kills Bugsy by ‘introducing a threat (your bombs) into his environment’, while doing so in Start-Up kills him with ‘a threat (the second trolley) already present in his environment’.Footnote ³³ But whether we count a threat as ‘present’ in someone's environment is simply a matter of something like how far it must move in order to kill him, and this clearly has no intrinsic moral significance. If we were sure it would kill Bugsy (and only Bugsy), it wouldn't matter if pushing the button in Start-Up caused the second trolley to travel from the south-west tip of Portugal to the north-east tip of Siberia before running him over.

To confirm the irrelevance of such factors, suppose you had three options: (1) push one button that sends a trolley into the missile and causes a second trolley to start moving and hit Freddy and Suzy, (2) push another button that causes the frontal bombs of a drone to fall on the missile and causes its rear bombs to fall on Bugsy's house, or (3) do neither 1 nor 2. Clearly it would be indefensible to choose option 1 over option 2. Something about its being the function of the second button to kill or its introducing a threat into Bugsy's environment is a ridiculous justification for killing an additional child or (if Suzy wasn't centred on the track) for exposing an additional child to a 1/11 risk of death or depriving her of seven years of life.

Our fifth step asks whether it would matter if the bombs you drop on Bugsy explode chemicals or kill him more ‘directly’, as in:

Surely there is no moral difference between pressing the button in Dual-Action Bomber and Dual-Action Bomber 2. It seems obvious that it cannot matter whether your bombs kill Bugsy by exploding chemicals or crushing him.

Frances Kamm has, however, proposed a view according to which it would appear to be permissible to drop your bombs in Dual-Action Bomber but wrong to drop them in Dual-Action Bomber 2. According to Kamm, it is wrong to save five when ‘something our means brings along with it causes [Bugsy's death] . . . directly’, but permissible to save five when ‘what our means bring along with them indirectly causes’ his death. She contrasts a case where to save five from dying you must drive them to the hospital, but driving to the hospital will eject lethal acid from a tank on your car onto Bugsy (Acid), with a case where driving the five to the hospital will produce vibrations that will cause rocks at the roadside to fall and crush Bugsy (Vibrations). Kamm claims that in Acid, something you need to save the five (your car's moving to the hospital) ‘introduces something (acid) in a context where it directly harms someone’, while in Vibrations what you need to save the five (your car's moving) merely causes things ‘already present in the context’ (rocks) to ‘directly’ cause death. In the same way, it would seem that in Dual-Action Bomber 2, something you need to save the five (flying and pushing the button where you do) ‘introduces something’ (the rear bombs) ‘into a context where they directly harm’ Bugsy, while in Dual-Action Bomber, what you need to save the five merely causes things ‘already present in the context’ (chemicals in the basement) to ‘directly harm’ him.Footnote ³⁴

As I mentioned, whether a cause of death is ‘already present’ or ‘introduced into’ a context is simply a matter of something like the distance it must travel to harm someone, and whether something causes harm ‘directly’ or ‘indirectly’ is simply a matter of something like whether it harmfully interacts with a victim's body or causes something else to do so. As I also mentioned, these factors seem as obviously devoid of intrinsic moral significance as skin colour. Kamm fails to make even a half-hearted attempt to shoulder the extreme burden of proof against the relevance of such manifestly irrelevant factors. After claiming that the important moral distinction is between ‘subordinating’ Bugsy to the five and merely ‘substituting’ Bugsy for them, Kamm admits that she can see no reason why there is any difference in terms of ‘subordinating’ and ‘substituting’ between using means that cause harm ‘directly’ as opposed to ‘indirectly’. Indeed, she says ‘the fact that we should not allow ourselves to do even the latter if a greater good were not at stake suggests that the weight of the greater good . . . is being transmitted so as to elevate the use of the means relative to the importance of not harming a person [thus subordinating this person]’.Footnote ³⁵

To confirm the irrelevance of whether your means of saving the five causes something that more ‘directly’ or more ‘indirectly’ kills Bugsy, suppose you had three options: (1) push one button that causes your frontal bombs to fall on the missile and causes your rockets to launch into Freddy and Suzy's house, which will cause vibrations that detonate chemicals in the protected basement in which they are hiding, (2) push another button that causes your frontal bombs to fall on the missile and causes your rear bombs to fall on and crush Bugsy, or (3) do neither 1 nor 2. It seems obviously indefensible to choose option 1 over option 2. That your bombs would kill Bugsy directly without the mediation of something like a chemical explosion is an unconscionable justification for killing an additional child or (if Suzy was partly shielded from the explosion) for exposing an additional child to a 1/11 risk of death or depriving her of seven years of life.

Our sixth step asks whether it would matter if the button released a single bomb that both destroys the missile and kills Bugsy, as in:

It seems clear on reflection that there is no moral difference between pressing the button in Dual-Action Bomber 2 and The Big Bomb. Surely it cannot matter whether the button destroys Bugsy and the missile with different bombs or the same one.

One attempt to distinguish between Dual-Action Bomber 2 and The Big Bomb is suggested by Matthew Hanser. Hanser claims that ‘you do B intentionally’ if you act with the intention of doing A and you know that A will ‘non-accidentally’ cause B. According to Hanser, A ‘non-accidentally’ causes B if A's happening is ‘sufficient to explain’ B's happening, and he suggests that this is true if A's happening ‘formally explains’ B, or A-type events have the function of causing B-type events. Hanser suggests that if you drop a nuclear bomb on a factory and it destroys neighbouring houses, the nuclear bomb has the function of destroying everything in its blast-radius, so its detonation suffices to explain the destruction of the houses, and you destroy them intentionally. On the other hand, he suggests that even if you can predict that dropping small conventional bombs on a factory will cause some of them to fall on neighbouring houses, they do not have the function of doing this, so your dropping them does not suffice to explain the destruction of the houses, and you do not destroy them intentionally.Footnote ³⁶ Following Hanser, one might try saying that in dropping the big bomb you kill Bugsy intentionally, but in pressing the button that drops your rear and frontal bombs you do not.

But this would be a poor application of Hanser's analysis. There is just as much reason to say that pushings of the button that releases both the frontal and rear bombs have the function of destroying everything in the blast radii of both sets of bombs as there is to say that droppings of the big bomb have the function of destroying everything in its blast-radius. Just as engineers designed the mechanism for releasing the big bomb to deliver its blast to the point on which it lands, they designed the mechanism for releasing the frontal and rear bombs to deliver their blasts to the points on which they land. More importantly, whether you do something ‘intentionally’ in Hanser's sense cannot have intrinsic moral importance. On his analysis whether we do B intentionally depends on whether what we intend in doing B is ‘sufficient to explain B’, but whether it is depends on which parts of the full causal history of B we find salient.Footnote ³⁷ Obviously, which parts of B's causal history we find salient is as devoid of intrinsic moral importance as skin colour. Moreover, Hanser's favoured generator of salience – facts about the ‘functions’ of artefacts – are historical facts about the intentions of the people who constructed them, which, as noted above, are obviously devoid of intrinsic moral importance.

A second attempt to distinguish between Dual-Action Bomber 2 and The Big Bomb is suggested by William Fitzpatrick. According to Fitzpatrick, if you intend to bring about state of affairs S, and S constitutes another state of affairs S′, then you intend to bring about S′. Fitzpatrick claims that if you drop munitions that destroy an area containing both your target and civilians, your intended means of destroying the target include the entire area being destroyed, and this area being destroyed constitutes the destruction of the civilians, so you intend the death of the civilians. By contrast, he claims that if you drop munitions on a factory knowing that ‘fallout’ will kill nearby civilians, your intended means need not include the destruction of the area the civilians occupy, so you need not intend their deaths.Footnote ³⁸ Following Fitzpatrick, one might claim that in dropping the big bomb, your intended means of destroying the missile include the destruction of an area containing Bugsy, so you intend Bugsy's death, but in pushing the dual-purpose bomb release button you need not intend the destruction of the area around Bugsy, so you need not intend his death. Given a version of the Doctrine of Double Effect [DDE] according to which there are stronger intrinsic moral reasons against inflicting harms we intend than harms we merely foresee, this might make the first act wrong but the second permissible.

But this cannot be right. First, there is just as much reason to say that you intend the destruction of the area destroyed by your rear bombs as there is to say that you intend the destruction of the entire blast-radius of the big bomb. The destruction of the area destroyed by your rear bombs does nothing to destroy the missile, but this is equally true of the destruction of those parts of the big-bomb's blast radius that aren't actually occupied by the missile. Since in both cases the only thing that serves your purpose is the destruction of the area occupied by the missile, this is all you intend to destroy.Footnote ³⁹ Second, while Fitzpatrick's criteria for when one state of affairs constitutes another are unclear, if a state of affairs you intend constitutes a distinct state of affairs, you need not intend the second state. If Fitzpatrick is right that your skull's being flattened constitutes your dying, then driving your car or wearing your clothes each day constitutes their being worn out. But as R. G. Frey observes, while we intend to wear our clothes and drive our cars each day, we merely foresee and do not intend their being worn out.Footnote ⁴⁰

Finally, Fitzpatrick is wrong to suggest that it matters intrinsically whether what we intend ‘constitutes’ or merely causes harm. Surely it is no more permissible to engage in area bombing if you know with certainty that, while the destruction of the intended area will not kill civilians outright, it will cause them to die of their wounds some seconds, minutes or days after you drop your bombs. Fitzpatrick might claim that the destruction of the area would still constitute the civilians’ ‘being seriously injured’,Footnote ⁴¹ but there seems to be no moral difference between intending something that constitutes serious injury and intending something equally sure to cause death that doesn't constitute serious injury. Suppose we could save the five by destroying a missile in the vicinity of Bugsy's house by either (a) dropping the big bomb, which we know with certainty will kill Bugsy by causing him to die of his wounds in two days, or (b) bribing a cannibalistic giant to smash the missile by giving him both Freddy and Suzy, whom we know with certainty he will eat in two days. Should we really take option (b) because we would be intending a mere cause of death rather than a ‘constituent’ of serious injury?

It is true that dropping the big bomb saves the five in virtue of its causing something (the bomb's explosion) that causes Bugsy's death rather ‘directly’, while the explosion that kills Bugsy in Dual-Action Bomber 2 does not help save the five. So in pressing the bomb-release in The Big Bomb you must intend something that kills Bugsy more ‘directly’ than anything you must intend in Dual-Action Bomber 2. But as we have seen, the mere number of causal links between what saves the five and what kills Bugsy is as obviously devoid of intrinsic moral importance as skin colour. To confirm the irrelevance of this distinction, as well as those inspired by Hanser and Fitzpatrick, suppose you had three options: (1) push a button that will both cause your frontal bombs to fall on the missile and cause your rear bombs to fall on Freddy and Suzy, (2) drop the big bomb on the missile, the explosion of which will envelop Bugsy, or (3) do neither 1 nor 2. It seems manifestly indefensible to choose option 1 over option 2. You could not seriously try to justify your conduct by claiming that you merely ‘killed Freddy and Suzy accidentally’ or ‘foresaw causing their lethal injuries’ while pressing the second button would have ‘been sufficient to explain Bugsy's death’ or ‘involved intending Bugsy's lethal injury’. This sounds false, and would in any event be an absurd justification for killing an additional child or (if Suzy was partly shielded) for exposing an additional child to a 1/11 risk of death or depriving her of seven years of life.

Our seventh step asks whether it would matter if you killed Bugsy in the course of eliminating a threat to the five by more ‘indirect’ means, as in:

In real life, Gettier terror bombing would be much less likely to save the five than dropping the big bomb. You could not be too sure that bombing the hospital would actually demoralize the crew into surrendering, as opposed, say, to strengthening their resolve to launch the missile in revenge. But to determine whether there is any intrinsic difference between these kinds of bombing, suppose that in a cool hour the missile crew decided that if Timmy died, the costs of continued resistance would outweigh its expected benefits. Fearing that they might be moved to fight on by an unreasonable desire for revenge, they constructed a reliable device that would detect, via photo-sensor, any explosions over the hospital, and subsequently destroy the missile. If you knew all this, it seems obvious that there would be no moral difference between dropping the big bomb on the missile and exploding the fragmentation bombs over the hospital. Both acts cause explosions that destroy the missile and kill Bugsy; it's just that the second act destroys the missile by activating the photo-sensor.Footnote ⁴³

The fact that you drop the fragmentation bombs directly over Bugsy might make dropping them feel more like ‘directly targeting’ him than dropping the big bomb on the missile. But in both cases the destruction of Bugsy's actual location is nothing to your purpose. Dropping the big bomb saves the five by causing an explosion in the area occupied by the missile; that the explosion spreads horizontally to Bugsy is simply an unfortunate side-effect of creating it. In the exact same way, dropping the fragmentation bombs saves the five by causing an explosion over the hospital that causes the photo-sensors to destroy the missile; that the explosion spreads vertically to Bugsy is simply an unfortunate side-effect of creating it. Obviously, whether an explosion harms someone by spreading horizontally or vertically is as devoid of intrinsic moral relevance as skin colour.

To confirm the lack of intrinsic difference between doing something that kills Bugsy and ‘directly’ eliminates a threat to the five (dropping the big bomb) and doing something that kills Bugsy and ‘indirectly’ eliminates such a threat (Gettier terror bombing), suppose you had three options: (1) destroy the missile with the big bomb, knowing that both Freddy and Suzy are in its blast-radius, (2) destroy the missile by activating the photo-sensor by exploding fragmentation bombs over the hospital, knowing that Bugsy is inside, or (3) do neither 1 nor 2. The bare directness with which causes of Bugsy's death save the five – distinct from their having a greater chance of saving them – would be an unconscionable justification for killing an additional child or (if Suzy was partly shielded) for exposing an additional child to a 1/11 risk of death or depriving her of seven years of life.

Our eighth step asks whether it would matter if the crew was reliably tracking Bugsy, as in:

It seems obvious to me that there is no moral difference between Gettier and Lewissian terror bombing. I cannot believe that it makes an intrinsic moral difference whether your killing Bugsy saves the five by getting the crew to believe falsely that you have killed Timmy or by getting them to believe truly that you have killed Bugsy.

However, Warren Quinn defended a version of the DDE according to which there is an important moral difference between Gettier and Lewissian terror bombing. Quinn observed that, because what causes the crew to surrender and saves the five is simply the flash of the explosion of the fragmentation bombs (rather than their fatal effects below), the Lewissian terror bomber need not intend any physical effects on Bugsy. But, Quinn noted, since the crew surrenders only because the bombs explode in Bugsy's vicinity, the Lewissian terror bomber ‘does strictly aim at exploding bombs in [Bugsy’s] vicinity . . . the bomber strictly intends to involve [Bugsy] in something (to make his bombs explode over [Bugsy]) in order to further his purpose precisely by way of [Bugsy] being involved’. Quinn's idea seemed to be that, because the crew is reliably tracking Bugsy, the five are saved only because Bugsy instantiates the property of bombs exploding over him, so the Lewissian terror bomber must intend that Bugsy instantiate this property. Moreover, Quinn suggested that this feature of Lewissian terror bombing is morally important. He urged us to accept a version of the DDE according to which it is harder to justify acting with the intention of causing a victim to instantiate a property in virtue of which she will be harmed than it is to act with the mere foresight that you will cause a victim to instantiate such a property.Footnote ⁴⁵

Quinn proposed a ‘Kantian’ rationale for his version of the DDE, suggesting that if an agent acts with the intention of causing victims to instantiate harmful properties, he treats them as means to his ends in a distinctively problematic way: ‘He sees them as material to be strategically shaped or framed by his agency’ and ‘treat[s] them as if they were then and there for his purposes’ (348). There are, however, deep problems with this rationale, many of which have been excellently articulated by Jonathan Bennett.Footnote ⁴⁶ It is quite unclear what Quinn means by ‘seeing someone as material to be shaped’ or ‘treating someone as if she existed for your purposes’. A Lewissian terror bomber who loves all the children equally and wishes only to save as many as possible certainly need not hold any silly metaphysical or teleological beliefs about ‘what Bugsy is’ or ‘what Bugsy exists for’. The Lewissian terror bomber does intend to make a fact involving Bugsy (that bombs explode over him) obtain, which he knows, under these circumstances, will cause Bugsy's death.Footnote ⁴⁷ But in what sense does this embody a more problematic attitude towards Bugsy than the Gettier terror bomber's intention to make a non-Bugsy-involving fact (that bombs explode over the hospital) obtain, which he equally well knows will kill Bugsy?

Quinn seemed closer to articulating a relevant difference between Lewissian and Gettier terror bombing when he suggested that a standard (or Lewissian) terror bomber's ‘victims are not only harmed but, in some sense, used’ (350). The Lewissian (unlike the Gettier) terror bomber does make use of Bugsy's presence, and the claim that ‘You were just using me!’ does seem to be a distinctive and plausible moral objection. But the injunction against ‘using’ others does not really support Quinn's version of the DDE. First, Quinn's actual formulation of the DDE makes it difficult to justify using, not victims, but (at most) facts about victims. For instance, Quinn's formulation makes it difficult to justify eliminating individuals who pose obstacles (like fatally blasting a fat man from a cave's mouth to save others from drowning inside), but this does not use victims as means of saving others.Footnote ⁴⁸

Second, the plausible injunction against ‘using’ others is quite distinct from – and much less important than – the morality of inflicting and preventing serious harm. To complain plausibly that ‘he was just using me’ is to complain that you were wrongfully deceived by a manipulator into serving his purposes. The wrong is essentially one of dishonesty or infidelity to an implicit undertaking to tell the truth or provide important information.Footnote ⁴⁹ We can justifiably complain of someone's being manipulated in this way even if it leaves her no worse off, and the complaint actually becomes infelicitous if the manipulation leads to serious harm. If someone tricks your friend into going on a date with him in order to make his wife jealous, and all this does is waste your friend's time and disappoint her expectations, it sounds right to complain that he ‘used’ her. But if he tricks your friend into going on a date in order to kill her – or even to non-fatally harvest one of her kidneys to sell on the black market – it would be a bizarre moral understatement to complain that he ‘used’ her. The manipulation would not be worth mentioning given the much larger issue of his seriously harming your friend.

Finally, to the extent that Bugsy is actually a subject of ethical concern, the Lewissian terror bomber (who does not deceive or manipulate Bugsy) need not intend that Bugsy instantiate any properties. Suppose that Bugsy and his twin were in an accident that destroyed Bugsy's body and his twins’ cerebrum, and surgeons transplanted Bugsy's cerebrum into his twins’ body, resulting in an individual who awakens with the same mental states as Bugsy. It seems clear that this would be just as good for Bugsy as survival: he should expect to have the experiences of the individual in his twin's body, and those who care for Bugsy should, out of this care, strongly prefer the transplant and want the individual in his twin's body to be as well-off as possible. So to the extent that Bugsy is a subject of ethical concern, he (who survives) is distinct from his body (which does not).Footnote ⁵⁰ But the Lewissian terror bomber need not intend instantiations of properties by Bugsy, the subject of welfare who goes where his cerebrum goes. Since the missile crew is only able to track Bugsy by tracking his body, they would not notice if your agents moved Bugsy out of the hospital in a glass jar before you dropped your bombs. Indeed, we can imagine that, in Lewissian Terror Bombing, you had arranged for this to happen, and for Bugsy to be safely transplanted into a back-up body that you have waiting for him. But, at the last minute, you are informed that the doctor was killed before he could extract Bugsy from his current body. Now it should be clear that, if you proceed with the raid, you must intend that Bugsy's current body (which the crew is tracking) instantiates the property of bombs exploding over it, but you merely foresee and do not intend that Bugsy, whose being trapped in that body is nothing to your purpose, instantiates the property of bombs exploding over him.

Should it really matter whether, in doing what you know will harm Bugsy, you intend that his body instantiates some property? Suppose your initial evidence was that you would be Gettier terror bombing – that the crew thought Timmy was in the hospital, but unbeknownst to them, Timmy had been evacuated and Bugsy had been taken to the hospital, so that the bombs exploding over Bugsy's body would play no role in saving the five (or a bazillionFootnote ⁵¹) children. Suppose you were about to drop your bombs, when you received news that the crew has discovered that Timmy is gone and Bugsy is in the hospital. Should you really change your mind about whether to drop your bombs on the grounds that, because you now know that the location of Bugsy's body matters to the salvation of the five (or bazillion), you would be intending not only that the bombs explode over the hospital – which you already knew would kill him – but intending that the bombs explode over his body? Such reasoning seems manifestly absurd.Footnote ⁵²

Or suppose that you had three options: (1) drop bombs over one hospital, in which the crew falsely believes Timmy and Lucy are trapped, while in reality they have been evacuated but unbeknownst to the crew Freddy and Suzy have become trapped in that hospital, (2) drop bombs over another hospital, in which Bugsy – whose body the crew is reliably tracking – is trapped, or (3) do neither 1 nor 2. On reflection it seems indefensible to choose option 1 over option 2. That killing Bugsy would have involved intending that bombs explode over his body, while killing Freddy and Suzy involved intending only that bombs explode over a hospital in which you knew them to be trapped, seems on its face to be an absurd justification for killing an additional child or (if Suzy was partly shielded) for exposing an additional child to a 1/11 risk of death or depriving her of seven years of life. Quinn's attempt to overcome this appearance of absurdity by claiming that killing Bugsy, unlike killing Freddy and Suzy, would have involved problematically ‘using’ someone fails because it mistakes a minor injunction against deceptively manipulating others for a major injunction against particular ways of harming others.Footnote ⁵³

Our ninth step asks whether it would matter if what saved the five was a physical effect on Bugsy's body, as in:

It seems obvious to me that there is no moral difference between Lewissian and standard terror bombing. I cannot believe that it makes an intrinsic moral difference whether your killing Bugsy saves the five by getting the crew to see the explosion that kills him or by getting them to see his body.

Since standard terror bombing saves the five only in virtue of the crew's seeing Bugsy dead, it might seem that the standard (unlike the Lewissian) terror bomber must actually intend Bugsy's death as a means of saving the five. If, as a traditional reading of the DDE claims, it is harder to justify inflicting harms that we intend than harms that we merely foresee, the standard terror bomber's intending Bugsy's death would make his act harder to justify than the Lewissian terror bomber’s. But an argument of Bennett's shows that that the standard terror bomber need not intend Bugsy's death either. Bugsy's death consists in the permanent inoperativeness of his bodily or mental activity, but all the standard terror bomber needs to save the five is for Bugsy's body to remain inoperative long enough to be seen by the crew and cause them to surrender. Of course, under the circumstances this temporary inoperativeness cannot be caused without also causing permanent inoperativeness, but this seems no different from the Lewissian terror bomber's inability to cause an explosion over Bugsy's head without also causing fragments to rain down and kill him.Footnote ⁵⁴

Still, because the standard terror bomber needs Bugsy's body to suffer damage that he knows will in fact prove fatal, it would seem that the standard (unlike the Lewissian) terror bomber must intend for Bugsy to suffer foreseeably fatal injuries. It might seem plausible to interpret the DDE as holding that it is harder to justify inflicting these kinds of injuries, or committing foreseeably lethal ‘assault’ or ‘battery’, with the intention than with the mere foresight of doing so, which would give us grounds for saying that standard terror bombing is harder to justify than Lewissian terror bombing.Footnote ⁵⁵ There are, however, ways of varying the case so the lethal physical effect you intend on Bugsy does not seem to constitute an ‘injury’ or ‘battering’. Suppose that the only way to demoralize the missile crew was for them to see you gently strap a bomb-vest on Bugsy and set the timer to explode sometime after they surrender, which you know with certainty will kill Bugsy (because it cannot be removed or defused without exploding). Here you need only intend that the timer be set and Bugsy have the vest strapped on him, which is not itself his being injured or battered in any intuitive sense. But much as we saw with the cannibalistic giant, it seems obvious that it makes no intrinsic moral difference whether in knowingly killing someone the lethal effects you intend to have on her body constitute or merely cause injuries or batteries. If you could demoralize the crew and save the five by (1) gently strapping vests on Freddy and Suzy which you knew with certainty would kill them or (2) bombing Bugsy, which you knew with certainty would kill him, it would be preposterous to take option 1 on the grounds that you merely intended to set the vests and strap them in, foreseeing but not intending that this would injure or batter them.

There is, however, a legal sense of ‘battery’ in which your causing something simply to touch or move someone's body (without her consent) is itself your battering her if this moving or touching in fact causes her to be harmed. John Mikhail considers a version of the DDE according to which it is harder to justify committing ‘battery’ in this sense with the intention than with the mere foresight of doing so, which would entail that standard terror bombing and deliberately strapping a lethal bomb-vest on Bugsy are harder to justify than Lewissian Terror Bombing.

While Mikhail's primary interests are in descriptive psychology, he asserts that this version of the DDE ‘[makes] no mention of what I take to be morally irrelevant information, such as race, gender, religious affiliation, and so forth’ (259). But this claim seems to reflect an inadequate appreciation of what the principle is actually saying. ‘Intending battery’ may sound morally relevant if it calls to mind actions that are intended to harm others seriously, or affect their bodies in other ways that are more likely to harm them seriously than actions that risk collateral damage. But what Mikhail's DDE is actually saying is that it's intrinsically harder to justify deliberately causing an outcome in virtue of which you know someone will be harmed if that outcome involves something's touching or moving her body. It seems obvious, however, that it is morally irrelevant whether you deliberately cause something to touch or move someone or deliberately cause it to do something else to her (like hover around her) that you know with equal certainty will harm her in the exact same way.

Suppose you could save the five by setting the timer on and strapping a hovering bomb vest around Bugsy, which would not touch or move him but follow him by detecting his movements and altering its location. Clearly, it is no easier to justify this than it is to justify strapping a bomb vest on Bugsy that follows him by gently touching his body. Suppose you could save the five by pressing a button that would cause a robot to put a bomb vest around Bugsy, and your initial evidence was that it would be a hover-vest. Suppose you were about to push the button, when you received information that the vest is actually an ordinary one. Should you really change your mind about whether to push the button on the grounds that, because you now know that the vest will stay lethally close to Bugsy by touching his body instead of hovering a few millimetres above it, you would be intending not simply that it hover around him but that it touch him? To do so would be ludicrous; the information on which Mikhail fixates is at least as obviously irrelevant as information about Bugsy's race, gender or religious affiliation.Footnote ⁵⁶

To confirm the lack of intrinsic difference between Lewissian and standard terror bombing, suppose you had three options: (1) explode fragmentation bombs, which the crew will see from afar, over Freddy and Suzy whom they are reliably tracking, (2) bomb Bugsy, whose lifeless body will be seen by the crew, or (3) do neither 1 nor 2. It seems indefensible to choose option 1 over option 2. That killing Bugsy would have involved intending that explosions impact his body, while killing Freddy and Suzy involved intending only that explosions occur so close to them that you knew with equal certainty they would be killed, is clearly an absurd justification for killing an additional child or (if Suzy was partly shielded) for exposing an additional child to a 1/11 risk of death or depriving her of seven years of life.

Our tenth step asks whether it would matter if saving the five required deliberately destroying Bugsy's organs, as in:

It seems obvious that there is no moral difference between lethally injuring Bugsy in Standard Terror Bombing and Organ Targeting. It cannot matter whether you must fatally immobilize Bugsy or fatally eliminate his organs.

It might seem that Bugsy has a special claim on his organs, and that obliterating them does not ‘[allow] a person to keep things to which he has a claim’.Footnote ⁵⁷ But could this really make obliterating his organs any harder to justify than dropping bombs that you know with equal certainty will kill him by disrupting his vital processes? Doesn't Bugsy (or his representatives) have just as much of a claim on our allowing his vital processes to continue as he does on our leaving his vital organs in place, because doing otherwise would in both cases kill him?Footnote ⁵⁸ To confirm the equal weightiness of these claims, suppose a second mechanism will disable the launch sequence if it detects the cessation of Freddy's and Suzy's bodily processes, giving you three options: (1) bomb Freddy and Suzy, which will lethally stop their bodily processes, (2) obliterate Bugsy's organs, or (3) do neither 1 nor 2. Clearly, it would be indefensible to choose option 1 over option 2. That killing Bugsy would have destroyed his organs, on which he has a special claim, while killing Freddy and Suzy simply interrupted their vital processes, would be a preposterous justification for killing an additional child or (if the sensor was connected only to Freddy, and Suzy was partly shielded) for exposing an additional child to a 1/11 risk of death or depriving her of seven years of life.

Our eleventh and final step asks whether it would matter if, instead of saving the five by destroying Bugsy's organs to disable the launch sequence in Organ Targeting, you could save the five by harvesting Bugsy's organs and transplanting them into the five in Transplant. It seems obvious that there is no moral difference between these acts. Surely it cannot matter whether you lethally cause Bugsy's organs to be absent from his body in order to prevent a missile from killing the five or in order to transplant the organs into the five to prevent their dying of organ failure.

The fact that you save the five by transplanting into them the same organs you remove from Bugsy might give Transplant a more saliently ‘redistributive’ character, calling on the thought that you are wrongfully giving what belongs to Bugsy to someone else. Tooley suggests that removing Bugsy's organs in Transplant may be wrong because ‘[Bugsy] has more of a claim to his organs – simply because they are his – than the other five individuals’ and ‘the choice [in Transplant] is . . . between allowing a person to keep things to which he has a claim, and giving them to someone else who does not have a comparable claim to them’ (215–16). But clearly it cannot matter whether, in benefiting others by depriving Bugsy of that to which he has a special claim, you give that very thing to others or do something else with it that is useful to them. Surely stealing someone's property is no more objectionable than wantonly destroying her property in exchange for a payment from a third party equal to the value of her property. Stealing from the poor to give their meagre belongings directly to the rich is no less wrong than destroying their belongings as a sacrifice to Poseidon, in exchange for which he rains down on the rich manna of equal value to those belongings.

To confirm the moral equivalence of Organ Targeting and Transplant, suppose you had three options: (1) obliterate the organs of Freddy and Suzy with a large-calibre rifle to please Poseidon, who will in exchange for this obliteration heal the five, (2) remove Bugsy's organs and transplant them into the five, or (3) do neither 1 nor 2. It seems indefensible to choose option 1 over option 2. That in killing Bugsy you would have given his organs to the five, while in killing Freddy and Suzy you merely destroyed their organs to save the five, would be an absurd justification for killing an additional child or (if Freddy's death and a risk or lesser harm to Suzy would be sufficient to please Poseidon) for exposing an additional child to a 1/11 risk of death or depriving her of seven years of life.

V. WHERE DO WE GO FROM HERE?

I have thus argued that it is morally irrelevant whether, in doing what saves five and kills Bugsy, you: (1) kill Bugsy with something that would not have ‘directly’ killed the five, (2) kill Bugsy with something that would not have ‘indirectly’ helped kill the five, (3) kill Bugsy with things that are initially stationary, (4) kill Bugsy with things designed to have destructive effects, (5) save the five with means that cause something ‘directly’ to kill Bugsy, (6) save the five in virtue (or with the intention) of causing something that itself ‘directly’ kills Bugsy, (7) save the five in virtue of something that kills Bugsy and ‘indirectly’ eliminates a threat to the five, (8) save the five in virtue of Bugsy's body instantiating some property in virtue of which he dies, (9) save the five in virtue of Bugsy's body being lethally touched or moved, (10) save the five in virtue of Bugsy's organs being lethally removed, or (11) save the five in virtue of Bugsy's organs being lethally removed and transplanted into the five.

None of these differences could make the difference between permissibly and impermissibly killing Bugsy to save five. Moreover, because none of these differences can justify even exposing an additional individual to a 1/11 risk of death or depriving her of seven out of seventy-seven years of life, they cannot make small moral differences that cumulatively make the difference between permissibly and impermissibly killing one to save five.Footnote ⁵⁹ Thus, because these differences are all that separate saving the five by diverting in Trolley and saving them by harvesting Bugsy's organs in Transplant, and they cannot make the difference between permissibly and impermissibly killing one to save five, it cannot be permissible to divert in Trolley but impermissible to harvest Bugsy's organs in Transplant.Footnote ⁶⁰

If this is right, then either the intuition that it's wrong to harvest Bugsy's organs or the intuition that it's permissible to divert the trolley is mistaken. Well, which is it? While a full resolution of this question is beyond the scope of this article, I wish to conclude by explaining what I take to be the important considerations that bear on answering it, and why I think they favour the view that it is our intuition that we may permissibly divert the trolley that has to go.

The first important consideration is that the Doctrine of Doing and Allowing (DDA), according to which it is intrinsically harder to justify inflicting harm on others than it is to justify failing to help others, seems directly plausible. As Philippa Foot put it, it seems that ‘it takes more to justify an interference than to justify the withholding of goods or services’, and as Bernard Williams put it, it seems that ‘each of us is specially responsible for what he does’, including the harms our actions produce.Footnote ⁶¹ As we saw in section III, the fact that an act would kill someone seems on its face to be a perfectly good justification for taking an alternative course of action that kills no one, even if it saves fewer individuals.

An absolutist version of the DDA, on which no amount of benefit could justify inflicting any amount of harm, seems (to me at least) quite implausible. If you could prevent serious harm (like death or significant suffering) to someone by inflicting a relatively trivial harm (like being jostled or pricked by a pin) on someone else, it would seem absurd to refuse to prevent the harm on the grounds that it would involve harming someone else. Similarly, if we knew with certainty that billions of innocents would survive if and only if we killed one other innocent, it seems difficult to deny that it would be permissible to kill the one. But it does seem plausible that it is a great deal more difficult to justify inflicting harm than it is to justify failing to prevent harm. Perhaps the most plausible version of the DDA would be one according to which the infliction of harm can be justified only by the fact that it is needed to prevent an enormously, catastrophically or mind-bogglingly greater amount of harm. What exactly constitutes a mind-bogglingly greater amount of harm will be vague, but I think that vague comparisons like this are unavoidable in any plausible form of ethical thought.

If the direct plausibility of such a version of the DDA can survive further critical scrutiny,Footnote ⁶² we would seem to have the same kind of reasons to accept it as we have to accept other plausible principles, like the Principle of Beneficence, which states that there are moral reasons to promote the welfare of others. We would then have the very same kind of evidence for the view that moral reasons against inflicting harm are much stronger than moral reasons against allowing harm as we have for the view that there are moral reasons against allowing harm in the first place. But if we must accept such a version of the DDA along with the Principle of Beneficence, the conclusion seems inescapable that, because diverting the trolley inflicts the harm of death on Bugsy and does not prevent a mind-bogglingly greater amount of harm to others, it is impermissible to divert it.

Our intuitions about Trolley may rebel at this conclusion, but as we have seen, our intuitions about cases are unsustainable. Still, there seems to be a more principled source of discomfort with the conclusion that we may not divert the trolley. This is that, according to this application of the DDA, whether we are permitted to do what results in the most individuals surviving depends on who happened to be threatened. If the trolley had been headed towards Bugsy rather than the five, we would have been permitted to do what would have resulted in five surviving by allowing it to hit Bugsy. But now, simply because the trolley happened to be headed towards the five instead of Bugsy, the DDA requires us to do what results in only one surviving by allowing it to hit the five. This sort of dependence of what we may do on where the trolley was originally headed can, in Alastair Norcross's words, seem to implausibly ‘invest . . . the course of nature with moral significance’, and can seem inconsistent with Thomson's observation that ‘there is no principle of moral inertia . . . no prima facie duty to refrain from interfering with existing states of affairs just because they are existing states of affairs’.Footnote ⁶³ This second consideration suggests that we must be permitted to minimize harm even if this involves causing harm to befall those who are not already threatened in cases like Trolley. But since as we have seen there is no relevant difference between Trolley and Transplant, this supports the conclusion that our intuition that we may not harvest Bugsy's organs is mistaken, and that we are permitted do that as well.

I think, however, that that this second consideration is mistaken, and that the DDA does not implausibly invest the course of nature with moral significance. The DDA does not hold that you should not divert in Trolley simply because the trolley will hit the five in the course of nature. The DDA permits all sorts of alterations of the course of nature, including most beneficent alterations that make others better off than they would have been. The only alterations it condemns are those that make others worse off, which it condemns, not because they are alterations of the course of nature or existing state of affairs, but because they are inflictions of harm. The DDA does not hold that which set of individuals is threatened by the trolley has basic moral significance; it merely holds that it has significance as a determinant of whether your act inflicts harm, or makes someone worse off than she would have been absent your interference.

Still, as Bennett has observed, it can seem unclear why our reasons to avoid acting in ways that result in someone's suffering harm should be strengthened by the mere fact that the harm would not have occurred absent our interference.Footnote ⁶⁴ I confess that this does not seem so unclear to me; I find Williams's suggestion that we have special obligations not to make others worse off plausible. But I think there is an even more plausible way of understanding the DDA.

When you prevent harm to the five by inflicting harm on Bugsy, you benefit the five at Bugsy's expense, in that you benefit them by making things worse for Bugsy. This is very different from simply choosing to save five swimmers rather than one swimmer, where you do not benefit the five by making anything happen that makes the one worse off. I think that it is highly plausible that it is harder to justify benefiting some individuals at the expense of others than it is to justify simply benefiting some individuals rather than others. The idea here is to reject the Machiavellian dictum that ‘the ends justify the means’ by holding that, if doing something will benefit some individuals at the expense of others, the benefit loses (some of) its status as a good reason to do that thing. If this is right, it is typically harder to justify inflicting harm than it is to justify allowing harm, not because reasons against inflicting harm are in themselves stronger than reasons against allowing harm, but because the benefits that would justify inflicting harm would come at the expense of others, and this weakens their justifying force.Footnote ⁶⁵

It thus seems to me that the benefits of killing, even as a side-effect in cases like Trolley, come at the expense of those killed and cannot be justified by such benefits as the salvation of five others. McMahan has recently objected to this sort of extension of deontological constraints against inflicting harm to cases of foreseen but unintended harm on the grounds that, ‘Since almost all contemporary warfare involves the killing of innocent bystanders in large numbers, this view entails pacifism as a matter of practical policy.’Footnote ⁶⁶ Even if McMahan were right about the entailment, I am sceptical of the force of his objection. If the most plausible set of moral principles we can find entails that we should be pacifists, I think we should be pacifists. The mere fact that plausible principles tell us to revise our intuitions about cases is not a good reason to reject those principles – else racists could justifiably conclude that skin colour per se must be morally relevant.

But McMahan's claim that deontological constraints against foreseeably inflicting serious harm would entail de facto pacifism seems to presuppose that the constraints would be absolute. As I indicated, however, the most plausible versions of these constraints hold that we are justified in inflicting serious harm on innocents if (but only if) it is the only way to prevent a catastrophically or mind-bogglingly greater amount of serious harm. In real life, neither the benefits nor the harms of military operations are certain, but the stronger the reasons we have to avoid doing something, the less willing we should be to risk doing it, and the greater the expectation of benefit we need to justify this risk.Footnote ⁶⁷ Thus, under conditions of risk, a plausible non-absolutist version of the DDA will require that the possible benefits of our acts, weighted by their probabilities, be greater than the possible harms inflicted by our acts, weighted by their probabilities, by a mind-boggling margin.

I do not think that it is impossible for contemporary military operations to have this property. Carefully targeted strikes against genuine terrorists who really are about to attack large numbers of innocents are probably capable of having it. Given the murderous ambitions of the Nazis in the Second World War, daylight raids on factories and fighting in populated areas in critical battles probably had this property as well.Footnote ⁶⁸

But it probably is not easy for contemporary military operations to have the property of being reasonably expected to prevent mind-bogglingly greater amounts of serious harm than they are likely to inflict. A strike that is certain to prevent a terrorist from killing one innocent but has a 10 per cent chance (or even perhaps a 1 per cent chance) of killing another innocent does not seem to have this property, any more than doing what is certain to kill one innocent to save ten (or even perhaps a hundred) others. If we can guarantee no less than a 10 per cent or 1 per cent chance of seriously harming civilians who are near a target we wish to strike,Footnote ⁶⁹ we should presumably be extremely hesitant to strike it, and strike only if the consequences of not striking are expected to be much worse than the certain death of an innocent individual. If despite our best efforts to minimize collateral damage the expected amount of harm we will inflict on innocents by fighting a war cannot be expected to exceed by a mind-boggling margin the harm we will prevent, we should on this view not fight that war. This will probably rule out many wars, but as the example of the Second World War illustrates, it will not rule out all.

If I am correct that there is no intrinsic moral difference between killing innocents as a ‘side-effect’ and killing them by directly targeting them, the greater ease of justifying acts that risk collateral damage must lie entirely with their greater ratio of expected benefits to expected harms. Because terror bombing is almost never likely to prevent mind-bogglingly (or even marginally) greater harms than it inflicts, it makes sense to think of non-combatants as, for almost all practical purposes, morally immune from direct attack. Because operations that risk some collateral harm to civilians are far more likely to prevent mind-bogglingly greater harms, it makes sense to think of them as morally easier to justify. But we make a grave mistake if we categorize unintended harmful effects as in themselves easier to justify, and a graver one if this categorization leads us to accept collateral damage as justified by a ‘proportional’ good any time a salient military objective cannot be achieved without it. If there is, as I have argued, no intrinsic moral difference between directly attacking someone and harming her as a side-effect, then the expected benefits of acts that risk collaterally harming innocents are proportional to their expected harms only if a direct attack on those innocents would be justified if (per impossibile) its expected benefits exceeded its expected harms by the same margin.Footnote ⁷⁰