In the postscript to his seminal "Causation", David Lewis devotes a section to the topic of killing.
Killing, so they say, is causing to die. I am sure that I -- and likewise you , and each of us -- have caused ever so many people to die, most of them people yet unborn. Acts of mine are connected to their deaths by long chains of causal dependence. But I have never killed anyone I hope.
For instance, suppose I write a strong recommendation that lands someone a job; so someone else misses out on that job and takes another; which displaces a third job-seeker; this third job-seeker goes elsewhere, and there meets and marries someone; their offspring and all their descendants forevermore would never have lived at all , and a fortiori would never have died, and so presumably their deaths would not have occurred but for my act... It would be strange to single out my act as the cause of all those deaths. But it is a cause of them.... And still I deny that I have ever killed anyone.
So killing must be a special kind of causing to die. But what distinguishes this special kind of causation? [pp184-185]
He proceeds to dismiss candidate answers: It's not that the killing must shorten the victims life, a back up killer might be lurking in the background to insure that the victim would not have lived a second longer. Its not that the killing must be an act which makes the death probable; you can kill with a randomizing device. It is not the killing must lead to the death by some simple process. You can kill with an elaborate Rube Goldberg machine. It is not the the killing act and the death must be temporally or spatially proximate. You can kill by lighting a bomb with a thousand mile/thousand year fuse. The death doesn't have to be foreseeable: its a killing even if you didn't know the gun was loaded. And its not that killing must be done in the first person. You give the hostess poisoned chocolates, she innocently feeds them to a guest. You have killed the guest.
I suggest a different way to distinguish the right kind of causing: by its insensitivity to circumstances. When an effect depends counter factually on a cause, in general it will depend on much else as well. If the cause had occurred but other circumstances had been different, the effect would not have occurred. To the extent that this is so, the dependence is sensitive. Likewise if a causal chain consists of several steps of causal dependence, we can say that the chain is sensitive to the extent that its steps are. ... It may be that the causation depends on a an exceptionally large and miscellaneous bundle of circumstances all being just right. If any little thing had been different, that cause would not have caused that effect. When my strong recommendations cause lives and then deaths, that is comparatively sensitive causation -- there are many differences that would have deflected the chain of events. But if you shoot your victim point-blank , only some very remarkable difference in circumstances would prevent his death. The same is true if you set a Rube Goldberg machine, or a delayed action bomb, working inexorably towards its lethal outcome. The case of the bomb with a randomizer also is comparatively insensitive: the bomb might very well have chanced not to go off, but it isn't the fine details of the circumstances that would make it different.[p 186]
I don't think this works.
Consider my Rube Goldberg machine. I push this button here and machine gathers three seed numbers. It performs a simple calculation using those numbers and if the result corresponds to the social security number of any living American a signal is sent to a satellite which locates that person and vaporizes him or her with a laser. Most of the details are classified but I can tell you where I get my three numbers. They are the previous day's 1) closing NASDAQ average 2) exact noontime temperature (to three decimal points Celsius) of downtown Butte. Montana and 3) average number of pigeons in Trafalgar square.
Would you like to push the button? I expect not. Because I think you would say that if the button leads to someone's being vaporized, you've killed them. But notice that this apparatus is exquisitely sensitive to small differences in circumstances: the smallest fluctuation in the market, a gust of breeze in Montana, a flutter of a wing on the other side of the earth would make a difference to who, if anyone. dies.
I think this shows that Lewis's account is wrong. I think it also shows that Lewis was wrong to treat his account as equivalent to one of Jonathan Bennett.
Jonathan Bennett restates my suggestion this way: killing requires "that the causal chain run through a stable a durable structure rather that depending on intervening coincidental events".
My killing machine illustrates that a causal chain can run through a stable and durable structure-- the machine-- even though the chain is extremely sensitive to intervening coincidental events.
Will Bennett's "stability" and "durability" by themselves explain why my machine kills? I don't think so. If I stipulate that my machine is flimsy, requires constant repair and is bound to soon fall apart, does that make the death it causes when you push the button less of a killing? If we have a competition in which readers must tell stores about how a pushing a button here leads by the coincidence of factors 1-3 to a death there, how do we tell if the linkages described constitute a "a stable and durable structure". And why should stability and durability matter?