Judith Jarvis Thomson paper, “Self-Defense,” offers a general criterion for when it is permissible to kill in self-defense. She argues that somebody who violates “your rights that they not kill you, therefore lack rights that you not kill them.” (302) She further offers three types of cases where killing in self-defense is impermissible – the common factor being the innocent bystander. These impermissibility cases preclude agent-relative permissions, which, this paper argues, is done on faulty grounds. Thomson’s model cannot overlook the explanatory power of agent-relative permissions, and her explanation does not give a satisfactory reason to jettison the idea.

Thomson finds three cases where one is permitted to kill in self-defense; notably, they all involve an offender’s violation to one’s right to life. First, there is the case of the villainous aggressor (283) – for instance, somebody who would like to kill you by running you over with a truck. Second, there is the case of the innocent aggressor (284) – here, some drug made an innocent person go mad to the point of running you over with a truck. Finally, there is the case of an innocent threat (287) – the illustration here is a villain rolling a fat man, who will crush you upon impact, down a hill. All these cases threaten our right to life, but there are some stipulations noted by Thomson.

Thomson refines her view further in a number of ways. First, self-defense also captures cases of “grave bodily harm” (286). In such cases, the action taken may only be minimal – for instance, a punch on the shoulder would not merit killing the attacker. However, action do not have to be strictly proportionate; in other words, I am permitted use more force than is threatened against me, say, in cases of rape or grievous bodily harm. (286) Second, there is an asymmetry with violent aggressor cases in that violent aggressors relinquish their self-defense rights with respect to their victim; in other words, violent aggressors may not reciprocate their victims defensive action. Third, a third party may intervene on behalf of the victim but not on behalf of the aggressor or threat. There is something different about villainous aggressor cases verses the innocent aggressor cases and the innocent threat cases[1], but the status of the aggressor (i.e. the fault) does no “moral work” (286). Rather, it is in virtue of the fact that there is a threat to the right to life that self-defense is permitted.

She then outlines three cases where it is “impermissible for you to kill a person in defense of your life.” (289) First, the “Substitution-of-a-Bystander cases” (289) are situations where an innocent bystander is put in harm’s way in place of you. Second, the “Use-of-a-Bystander cases” (290), similar to Kant’s second maxim, are situations where an innocent bystander is used harmfully for the price of your safety. Third, the “Riding-Roughshod-over-a-Bystander cases” (290) are situations where the harm of an innocent bystander has a close causal connection with your safety. She is strict insofar as the impermissibility of killing in these cases and further rejects “agent-relative permissions.” (307)

The general thought of agent-relative permission is that subjects are permitted to care more about their own good over the good of others. Accordingly, if I were to care more about my life than some bystander’s life, then I am permitted to kill an innocent bystander in each of Thomson’s three cases. Agent-relative permission seems more convincing and intuitive when combined with self-referential altruism[2]. If we could sacrifice the life of an innocent person to save our child, surely we would and think it permissible to do so. Thomson pushes back on the general claim by arguing that we lack an “across-the-board agent-relative permission,” (307) which she supports with the scenario of choosing either starving to death or eating a baby (agent-relative permissions would permit us to eat the baby). The idea is that agent-relative permissions permit too much. Maybe agent-relative permissions work with talk of trolleys and killing faceless bystanders, but it might be committed to too much if it permits cannibalism and infanticide.

However, our moral intuitions can be clouded by biases, and it might be the case that agent-relative permission accurately reflects our moral intuitions but is met with cases which appeal to our bias nature. For a moment, let us grant that agent-relative permission allows us to kill an innocent bystander to save ourselves. If this is true, should the method of death matter to us? How about the characteristics of the bystander? Perhaps a slow, painful death of the bystander puts more pain and evil in the world; perhaps if the bystander was the Archbishop of Cambrai, agent-relative permission may not be enough save us[3]. With respect to eating babies so we do not starve to death, I do not any inherent reason to object to cannibalism in a life or death scenario; however, one might object to infanticide in virtue of some maximizing principle insofar as picking the younger life over the older life. Indeed, if one were committed to this type of consequentialism, agent-relative permission would not permit us to eat the baby to save ourselves. Yet, one does not have to be committed to valuing babies over anybody else, or find infanticide inherently more wrong than any other killing. So Thomson’s argument against agent-relative permissions seems to be an emotional appeal.

Perhaps Thomson was instead appealing to our basic intuition that the mere fact of intending the death of a bystander, as agent-relative permission seems to imply, is unacceptable.  However, on Thomson’s picture, “It is irrelevant to the question whether X may do alpha what intentions X would do alpha with if he she did it.” (294) Still, there might be good reason to reject agent-relative permission by simply stating that Thomson’s model is more powerful and has no room for it. This is feasible. It might be said that agent-relative permission subsumes the crux of her theory – namely, what is relevant for self-defense is only that somebody violating your right to life loses their right that you not kill them. Moreover, her distinctive cases on impermissible self-defense seem untenable given agent-relative permission. Is Thomson’s model mutually exclusive with agent-relative permission?

There may be a way to salvage Thomson’s theory while accepting agent-relative permission. Let us make a distinction between what is permissible and what is “merely excusable.” (285) Permissible actions are morally neutral, whereas a merely excusable action implies some moral failing but does not carry blame.[4] All things considered, Thomson’s idea of permissible killing in self-defense by virtue of one’s right to life still carries the same weight. What are weakened are Thomson’s cases of impermissible killing in self-defense, where impermissibility, which entails moral failing and blame, becomes merely excusable by virtue of agent-relative permission. Still, the force of her argument stands, along with agent-relative permissions.

Agent-relative permissions seem too close to our general moral intuitions to be overlooked. Moreover, although Thomson’s model of self-defense seems convincing, her reasons for rejecting agent-relative permissions seem superficial and unsatisfactory. This paper suggests that agent-relative permissions must be held, and that they can be mapped onto Thomson’s model given distinctions of what are permissible and what are merely excusable.

---

[1] Thomson leaves open the asymmetrical proportionality with respect to a threat being innocent versus a threat being villainous.

[2] I do not see any obvious problems with extending agent-relative permissions to self-referential altruism, but the argument does not depend on this point.

[3] There is a certain tacit threshold to agent-relative permission where we are permitted to care about our own good only slightly more than the good of others.

[4] For instance, crimes committed under duress are still blameworthy yet they are excused because they were committed while unlawful coerced.