Setiya argues that moral side-constraints (e.g. against killing as a means) are best understood as agent-neutral: “In general, when you should not cause harm to one in a way that will beneﬁt others, you should not want others to do so either.” (97) So, if we should not kill one even to prevent five killings, we should also prefer, as a mere bystander in either case, that the event of Five Killings occurs than that Killing One to Prevent Five (K1P5) does. This verdict is awkward on its face. It only gets worse when we add further cases for comparison.
Consider Killing One but Failing to Prevent Five (K1FP5). This is just like K1P5, except that — disastrously — killing the one fails to achieve the desired goal of preventing the five other killings. So six killings happen in total. K1FP5 is clearly vastly worse than K1P5. It’s presumably comparable in undesirability to Six Killings (where all the same killings occur, but none were intended to prevent others), maybe slightly better since at least one of the killings was more well-intentioned.
Now, if K1P5 is vastly preferable to K1FP5, which in turn is no worse than Six Killings, then presumably K1P5 is likewise vastly preferable to Six Killings. Five Killings is also preferable to Six Killings, of course, but presumably not quite so vastly so. In which case, we should conclude that K1P5 is preferable to Five Killings after all, contrary to the assumption of a(n agent-neutral) moral side-constraint.
Can we say more to establish that the preferability of K1P5 doesn’t lie between Five Killings and Six? Well, the general worry here is that this understates how much worse K1FP5 would be. The implication would be that it doesn’t matter all that much, once the one has been killed in an attempt to save the five, whether or not the five are actually saved. Setiya appears to say as much: “Ethically speaking, the damage has been done” (p. 104) once the five lives have been threatened, rather than when they are actually killed.
This seems like an odd claim to make about what we really ought to prefer. Surely the difference (of five killings prevented) between K1P5 and K1FP5 should, for example, outweigh the difference (of just one fewer killing) between Five Killings and Six, given that “all else is equal” between the respective members of each pair — whether the lives are saved or not makes no difference to what other harms occur in the paired scenarios.
Consider how much you should care about one of the lives saved in K1P5. I would have thought that this should be more or less constant. But on the view under consideration, you must instead care much less about preventing existing lethal threats than you do about the introduction of a new lethal threat (at least if the prevention attempt was itself wrongful), even holding all else equal (such that the same harms result from the prevention attempt whether it is successful in saving this life or not). And that seems wrong.
I can understand a deontologist thinking that the importance of saving this life is outweighed by the importance of not killing a new innocent victim. But it’s something else entirely to downgrade the importance of saving this life, such that once the new victim has been killed, it now matters less whether or not our target life is saved as a result. But that seems to be what the agent-neutral deontologist is committed to.
Or am I missing something?