The (in)famous ‘trolley problem’ began as a simple variation on an example given in passing by Philippa Foot (1967), involving a runaway trolley that cannot be stopped but can be steered to a path of lesser harm. By switching from the perspective of the driver to that of a bystander, Judith Jarvis Thomson (1976, 1985) showed how the case raises difficulties for the normative theory Foot meant to be defending, and Thomson (along with many others after her) compounded the challenge with further variations that created still more puzzles of broader interest. In recent years, her thought experiments have even been co-opted by psychologists engaged in the empirical study of moral judgment (Greene 2001, 2008; Hauser 2006). Yet more than thirty years after launching the trolley problem, Thomson (2008) has now strikingly reversed course, retracting the very claim she had originally used to raise puzzles for Foot. I shall argue that this reversal is a mistake, leading to a needlessly counterintuitive, contrarian position about damage-control cases. Instead of overturning her earlier position, her new variations merely uncover a surprising insight about the conditions under which one may permissibly sacrifice another for a good end.

1. The central case at issue is what Thomson now calls Bystander's Two Options: you are a bystander who sees a runaway trolley headed toward five innocent people who cannot move off the track; you cannot stop the trolley, but you have access to a switch that will divert it onto a side track where one innocent person is trapped. Is it permissible to throw the switch and divert the trolley toward lesser harm? The common answer, and Thomson's earlier one, is that it is: in cases like this, where there is a public threat and we can act on it so as to divert it toward lesser harm, and where all else is equal, we may do so (some would say we must). The problem this raised for Foot is that this is a case of killing one to save five, since we kill the one by diverting the trolley onto his path (and we are merely saving the five, rather than avoiding killing them, since we would merely have let the five die had we done nothing); yet Foot had argued, plausibly, that while we may refrain from saving one in order to save five, we may not typically kill one to save five. This is because the stringent negative duty not to kill the one typically overrides the weaker positive duty to save the others – as illustrated by the thought that we may not kill one in order to use his organs for transplants to save others. Bystander's Two Options thus seems to pose a puzzle for Foot's view, as Thomson once argued, at least as an exception that needs to be explained.

Thomson has now changed her mind about the central intuition here: she now thinks that it is impermissible for you to divert the trolley toward lesser harm. The argument has two parts. First, imagine a variant, Bystander's Three Options, in which you also have the option of turning the trolley onto yourself (you are stuck on a third track), saving the five and sparing the other one. Presumably we are not morally required to make such drastic sacrifices for strangers, and most of us wouldn’t if we could. Suppose you wouldn’t. Could it then really be permissible for you to turn the trolley onto the one to save the five, imposing on him a burden that you refused to take on yourself? Thomson believes not: if you’re unwilling to divert the trolley onto yourself, then you have no business diverting it onto the one, and so you must do nothing and allow the five to die. Now return to Bystander's Two Options: if you may not turn the trolley onto the one in Bystander's Three Options, could it really be permissible for you to turn the trolley onto the one in Bystander's Two Options, given that you wouldn’t turn it onto yourself even if you could? Again, Thomson believes not: we cannot ‘decently regard [ourselves] as entitled’ to impose such burdens on others that we would not be willing to take on ourselves (2008: 366).

What about people who would be willing to sacrifice themselves, rather than the one on the side track, to save the five strangers if this were possible? Such a thing is rare, Thomson thinks, and not really admirable in any case, but suppose we have such a case. This is the second part of the argument. Again, Thomson asks: could you now permissibly turn the trolley onto the one in Bystander's Two Options? She still believes not, but for a different reason: since ex hypothesi the one on the side-track does not consent to being run over, and is not morally required to give such consent, you may not turn the trolley onto him; for you have another option that is permissible, namely, doing nothing, and it is preferable to choose that permissible course than to do something as problematic as killing the one without his consent.

Thus, putting both parts of the argument together: whatever your own inclinations, you may not divert the trolley toward lesser harm in Bystander's Two Options, but must allow it to run its course, resulting in the deaths of the five. To divert the trolley turns out, on Thomson's current view, to be morally equivalent (at least along the dimension of permissibility/impermissibility) to pushing a large man off a bridge and into the trolley's path, using his crushed body to stop it before it reaches the five – a case she had earlier argued was morally very different.

2. Let us consider whether Thomson's argument succeeds. One crucial principle upon which it relies is: Thomson motivates P by considering what she takes to be a parallel case where you have a good end that can be achieved only through a cost that you could pay but would prefer not to pay, and you pursue the end by making someone else pay the cost. Suppose, for example, that you want to send $500 to Oxfam but don’t want to give up the money yourself, so you steal it from your neighbour and send it in. Surely, she points out, this would be wrong, and she takes this to support the above principle about killing; indeed, she takes killing B in Bystander's Three Options to be far worse than stealing money for charity, since the loss involves life rather than money (2008: 365).

  • P: ‘A must not kill B to save five if he can instead kill himself to save the five’ (2008: 365).

This analogy, however, is a poor one. Stealing from one's neighbour is obviously wrong to do, even in the pursuit of good ends, except in very special and extreme circumstances; and merely preferring to save some money obviously does not qualify as such an exceptional circumstance. By contrast, the kind of case we are interested in is a very special one – which is precisely why it has attracted so much attention. The killing is no ordinary killing: as Foot pointed out, it is the result of merely diverting a public threat toward lesser harm, rather than initiating a lethal sequence against someone; and the victim's being killed is a mere side-effect of one's life-saving action, rather than an intended means to one's end. Our whole question, then, is whether in this very special set of circumstances it might be permissible to kill another as part of saving five, even where one declines to sacrifice oneself for that same end. No light is shed on this question by considering a very different case where the action in question (stealing to save oneself money) is uncontroversially wrong for independent reasons. Such a case cannot support a general principle, such as P, that would settle our question about the trolley.

Is there a better reason to believe P? Perhaps it will strike some as just intuitively compelling: surely it is problematic, even in pursuit of a good end, to impose a major harm on an innocent third party in preference to imposing it on oneself. Again, however, there is a methodological danger here. While this intuition may be very plausible in certain kinds of case, we will just beg the present question if we generalize from those cases and then apply the abstract intuition to Bystander trolley cases, disregarding their special features. Suppose you are given the choice of shooting one innocent person to prevent a terrorist from shooting five others. Whatever else we might say about this case, it seems that if you were given the choice to make yourself the sacrificial victim, then you could not opt for shooting another person instead; perhaps you could legitimately refuse to do anything, but you could not in any case shoot another while sparing yourself. Even if this is right, however, it shows at most something about ordinary cases of killing, where one initiates a lethal sequence. It shows, for example, that if you are just as large as the man standing next to you on a bridge, you may not presume to shove him in front of the trolley in order to stop it before it runs over five others, while sparing yourself; you may jump in front of it yourself or do nothing. But none of this shows us anything about the Bystander trolley cases, where one is merely diverting a public threat toward lesser harm that is a side-effect of that diversion. Similarly, drivers of vehicles may have a special responsibility to avoid harming others, even by taking actions that harm themselves where necessary, which might again support a principle of sacrificing oneself before innocent others in some instances; but this intuition cannot be imported to lend support to P as applied to Bystander cases where there is no such special responsibility.1

Perhaps instead of looking to intuitions about other cases to support P we might find such support in a deeper principle. Thomson's discussion suggests an appeal to a principle such as: This has a ring of plausibility: to be prepared to violate Q might seem ‘indecent’ because it is disrespectful of others, treating them as less important than oneself. Yet it is far from clear that this is correct. Respecting others as moral equals may not require the strong condition in Q, but only a weaker condition: This principle says only that if A is plausibly to take it to be permissible for him to sacrifice B, then A must allow that if he were instead in B's position, he would have no legitimate complaint against B for choosing to sacrifice him. The relevant symmetry implied by A's respecting B as a moral equal has to do not with A's being willing to treat himself as he is willing to treat B, but with A's being willing to accept B's right to do the same to him were their positions reversed; that is the only sense in which A must obviously be willing to ‘take on the burden’ himself: he acknowledges that he must reasonably accept a similar burden in reversed circumstances. (To put it another way: the relevant principle, according to R, is closer to the Golden Rule than to a requirement of impartiality; it tells A, in effect, to do unto B as A would be willing for B to do unto him, rather than telling A that he must not favour himself over B.) If this properly captures the requirement of respect and consistency at issue in the kind of case we are considering, then we could reject Q in favour of R. And R does not support P: it is consistent with R that A may kill B to save the five despite being unwilling to kill himself to save the five – at least in this very special type of case.3

  • Q: A's respecting B as a moral equal requires that A not sacrifice B (without B's consent) for end E unless A would be willing to sacrifice himself for E if he could do so instead.2

  • R: A's respecting B as a moral equal requires that A not sacrifice B for end E unless A recognizes B's equal right to sacrifice A for E if their positions were reversed.

3. It is so far an open question, then, whether P applies to Bystander trolley cases, and we have seen, as yet, no good reason to accept it, either by drawing on intuitions about cases or by appealing to more basic principles. That said, it should be granted that there is at least some moral tension in the idea that it can be permissible to choose to sacrifice another over oneself, even in this kind of case. The question, however, is whether this tension is problematic enough to support Thomson's rejection of the common belief that it is permissible to turn the trolley in Bystander cases where one would not be willing to sacrifice oneself. The other possibility is that it is permissible to turn the trolley as most of us believe, despite this tension, and that this just shows us something interesting and surprising: there are some cases – namely, these special cases of diverting a public threat toward lesser foreseen harm – in which it is permissible to sacrifice another even where one would not be willing to sacrifice oneself.

If there is something paradoxical in this, there is in any case something far more paradoxical in Thomson's claim that morality requires us to allow a public threat to do maximal damage when we could minimize it simply by redirecting the threat. We might be stuck with such a result if the worries Thomson raises were more compelling, but we have seen that they plausibly stem from the misappropriation of intuitions from importantly different cases and a misconstrual of the kind of symmetry implied by respecting others as equals. If we have to bite a bullet, we do better to accept the permissibility of sometimes sacrificing another even where we would not sacrifice ourselves than to deny the powerful intuition – held by (most) non-consequentialists and consequentialists alike – that damage control is generally permissible in diversion cases.

It helps to dispel the sense of paradox surrounding the idea of permissibly sacrificing another before oneself to recall how we are led to it by two plausible considerations. First, there are strong consequential reasons in favour of turning the trolley in Bystander cases, as recognized both by consequentialists and by (most) non-consequentialists. These reasons obtain whether or not the innocent victim will be another person or the agent herself. Second, a plausible morality would not require suicide to save strangers, as by turning the trolley on oneself. This is something Thomson herself accepts (indeed, she goes further to deny that it would even be admirable to opt for such extreme altruism).4 Putting these two considerations together, we can see why the consequential reasons in favour of turning the trolley do not translate into a requirement for an agent to turn it onto herself, though for all that, they remain weighty – something Thomson repeatedly fails to take sufficiently seriously: there is a great cost, after all, in doing nothing. Add to this now the fact that there is no similar counterweight to the consequential considerations when the victim will be someone else, and we are very plausibly led to the conclusion that we may turn the trolley onto another person even while permissibly refusing to turn it upon ourselves. This result just follows from the interaction of the above factors if we give them their due weight. It is an interesting result to learn to live with, not a reason to deny the permissibility of damage control. Thomson has once again extracted a useful insight from the Trolley problem, but not the one she thinks she has.

4. I have focussed so far on the first part of the argument, which is what is new and most interesting. As noted earlier, however, Thomson also argues that even if one were willing to sacrifice oneself where possible, one may not turn the trolley onto another without his consent. This may be treated more briefly.

First, it is worth noting that Thomson's argument for this claim relies on her claim that it is in any case entirely permissible to do nothing, allowing the greater harm to occur: given the availability of that morally safe option, she thinks, it is better to go with that than to presume to kill an innocent person without his consent. But it is hardly obvious that doing nothing is permissible. Again, there appears to be intuitive support for this claim, but this appearance may arise from conflating permissibility and blamelessness in thinking about these cases. Perhaps we should not blame someone who does nothing here: any morally decent person will have cultivated a strong reluctance to bringing about the death of an innocent person, and it is unreasonable to expect that she will overcome that reluctance in a flash and turn the trolley toward an innocent person, even if there is in fact sufficient justification for so acting. This intuition may then lead one to suppose that it is thus permissible to do nothing, but that doesn’t follow. It is possible that doing nothing is simply wrong, though blameless, given the combination of the strong consequential considerations and the special circumstances of a diversion-of-threat case. And if so, then it won’t do to cite inaction as the clearly preferable alternative to diverting the trolley.

More importantly, Thomson has introduced nothing new in bringing in the notion of consent, and to insist upon its necessity in Bystander trolley cases is simply to beg the interesting questions. Consent is no doubt often relevant. Consider Foot's (1994: 282) example of Rescue II: it seems (to many) that we may not typically drive over one person stuck in the road, against his will, in order to go and rescue five others, but that perhaps we may if that person is instead the father of the five and consents to it or even demands it (though we’ll want to get that in writing). The whole question, however, is whether trolley cases, involving mere diversion of an existing, public threat, are different from other cases of killing, such as deliberately driving over someone. One manifestation of this question is precisely the question whether the consent that would be necessary in cases of the latter sort is still necessary in trolley cases. Those who believe that it is generally permissible to turn the trolley obviously think that this is a special case where consent isn’t necessary. So Thomson's quick appeal to consent won’t gain any traction with those who don’t already share her view.

It seems, then, that neither part of her new argument is persuasive. This is not to deny that there will continue to be disputes over the rationale for the common intuition that it is permissible to divert public threats toward lesser foreseen harm: people will continue to argue over the relevance of such factors as the difference between acting on the threat itself and acting on a person as a means to diminish the threat, or between diverting a lethal sequence and initiating one, or between causing foreseen but unintended harm and using someone's harm as a means. My purpose has been just to show that Thomson has not given us cause to abandon her original insight – the one that revealed the trolley problem to be a genuinely interesting one – but has simply uncovered one more insight, with characteristic ingenuity.

1 Thomson considers a variant of this idea, arguing that a driver must kill himself (e.g. by swerving into a stone wall) in preference to swerving and killing one innocent person, if these are his only options for avoiding killing five others (2008: 371–72). Her own point in raising this issue, however, is not explicitly to support P, but to show how this principle cannot further be used to derive the implausible result (with an argument parallel to her argument for the Bystander case) that a driver who has only two options – swerving and killing one or doing nothing and killing five – and is unwilling to kill himself must likewise refrain from killing the one. The difference, she notes, is that in refraining from turning the trolley onto the one, the driver will be killing the five, which is different from the bystander's merely failing to save the five if he refuses to turn the trolley onto the one (372). Thomson takes the latter to be permissible, whereas the former is not, which is why her argument against turning onto the one does not go through in the driver case, thus saving her from the objection that her argument would prove too much.
2 This states only a necessary condition. As mentioned earlier, in the second part of her argument Thomson denies that the hypothetical willingness to sacrifice oneself is sufficient to justify sacrificing another. I’ll return to this.
3 Again, it is important that R states only a necessary condition. We can endorse R and its application to this sort of case without claiming that meeting the condition in R in other cases would be sufficient to guarantee respect for a sacrificed victim, much less to justify the action. It is plausible, for example, that I would fail to respect someone properly if I were prepared to sacrifice her, against her will, in the Transplant-type case mentioned earlier, and that this would be wrong to do, even if I were willing to allow for that person to do the same to me were our positions reversed. The claim, then, is only that R’s being satisfied plausibly helps to ward off objections to sacrificing another (when unwilling to sacrifice oneself) in the special sort of case with which we are concerned.
4 It is an interesting matter how best to defend the claim about the limits of morality's demand for altruism, but for present purposes I will not digress to explore this, since Thomson already grants it.

References

Foot
P
The problem of abortion and the doctrine of the double effect
Oxford Review
 , 
1967
, vol. 
5
 (pg. 
5
-
15
)
Foot
P
Steinbock
B
Norcross
A
Killing and letting die. Repr
Killing and Letting Die
 , 
1994
2nd edn
New York
Fordham University Press
(pg. 
280
-
89
)
Greene
J
, et al.  . 
An fMRI investigation of emotional engagement in moral judgment
Science
 , 
2001
, vol. 
293
 (pg. 
2105
-
8
)
Greene
J
Sinnott-Armstrong
W
The secret joke of Kant's soul
Moral Psychology
 , 
2008
, vol. 
3
 
Cambridge, MA
MIT Press
(pg. 
35
-
80
)
Hauser
M
Moral Minds: How Nature Designed Our Universal Sense of Right and Wrong
 , 
2006
New York
HarperCollins
Thomson
JJ
Killing, letting die, and the trolley problem
The Monist
 , 
1976
, vol. 
59
 (pg. 
204
-
17
)
Thomson
JJ
The trolley problem
The Yale Law Journal
 , 
1985
, vol. 
94
 (pg. 
1395
-
415
)
Thomson
JJ
Turning the trolley
Philosophy and Public Affairs
 , 
2008
, vol. 
36
 (pg. 
359
-
74
)