
Introducing the Distinction
We are all going to die. But some of us will die sooner than others. And the actions of other agents make a difference to when we die. Let us say that x, in doing something, is a factor in y’s death at t just in case y would have died later than t had they not done that. (We are using “factor” as a term of art here—it has obviously been chosen because it bears some resemblance to ordinary usage, but it is not important to us that it matches ordinary usage.) And to ensure that our current topic does not get too muddled with doing/allowing discussions, let us not be too expansive about what counts as doing something. In particular we shall generally not count omissions, failures to do something, as a kind of doing something.
Someone can kill someone else and not be a factor (in the defined sense) in that person’s death. Suppose Jones pushes a button that releases a trapdoor that sends Smith plummeting into a canyon to his death. If Jones had not pushed Smith, Smith would have been killed by a sniper right after the time of the actual trapdoor release. It is a deep canyon: Smith dies later than he would have if Jones had not released the trapdoor. Jones doesn’t know that—they just want Smith dead. Obviously, Jones killed Smith, but Jones wasn’t, in the sense defined, a factor in Smith’s death.
Our focus here will be on failures of entailment in the other direction. It is clear that one can be a factor in someone’s death without killing them. Suppose Jones takes Smith to the beach and a lunatic sniper, whose presence Jones did not foresee, kills various people on the beach, including Smith. Jones did not kill Smith (the sniper did), and this is true even if Jones coerced Smith into going to the beach. But it is easy to fill in the details so that Jones was a factor in the de-fined sense—in particular, it is easy to fill in the details so that if Jones hadn’t taken Smith to the beach, Smith would have died later. For pretty much anything we do, it is easy to imagine ways in which, by doing that thing, we are a factor in someone’s death. (Indeed, given butterfly effects, it is plausible that we can be pretty sure, and perhaps even know, for any action we perform, that there is someone who would have lived longer had we not performed that action.)
When x is a factor in y’s death, y would have lived longer had x not acted the way they did. How much longer? This varies from case to case, but it bears emphasis that the time differential provides no clean guide to whether x killed y or not. If x shoots person y in the head, x kills y. But the narrative can be filled in so that y would have lived only a tiny bit longer had they not been shot (as when one shoots a person who is close to death anyway); and of course it can also be filled in so that y would have lived a great deal longer had they not been shot (as when a person is fatally shot in their youth and would have otherwise lived a long life). The relevant time differential is not going to be the key to the contrast between cases where one is a mere factor in someone’s death and cases where one kills someone. What does explain that contrast?
Investigating the Distinction
“To kill” is an example of what linguists and grammarians call a causative verb, a kind of verb that implies causality. (Where V is a causative verb, “x V’d y but x had no effect on y,” makes no sense.) We wish to draw attention to four plausible ideas concerning how (at least a large class of) causative verbs work, all of which straightforwardly apply to “kill.” The first two draw upon prominent themes in Hart and Honoré’s classic Causation in the Law, an important resource for any study of the interplay of ethics and causation. The focus of those authors, of course, is on applications of causal ideology in a legal setting, but they emphasize that the courts’ uses of causal language “have their roots in certain features of a variety of concepts which permeate the daily non-legal discourse of ordinary men”. And while their focus is not on causative verbs per se, their discussion can easily be adapted into some promising heuristics concerning how such verbs are put to work.
Mediating Agents
The first idea is this. If x, by doing F, is a factor in y’s death, but y’s death also crucially depends on some later intervention by a voluntary agent that exploits x’s doing F, then x does not kill y. This applies a general idea about causation in the law that Hart and Honoré put as follows:
[F]ree, deliberate and informed act or omission of a human being, intended to exploit the situation created by defendant, negatives causal connection.
(For the purposes of this essay, we can rewrite “causal connection” as “caus-ative connection.”)
And here is one of their vivid examples:
A throws a lighted cigarette into the bracken which catches fire. Just as the flames are about to flicker out, B, who is not acting in concert with A, de-liberately pours petrol on them. The fire spreads and burns down the forest. A’s action, whether or not he intended the forest fire, was not the cause of the fire: B’s was.
The phrase “intends to exploit” in their formulation is crucial. As Hart and Honoré note, “The act of a consumer who uses a product without knowing that it is defective will not bar an action against a negligent manufacturer”. Similarly, if x plants a car bomb and y—ignorant of the danger—voluntarily turns the key, resulting in the death of y’s family, that does not stop us saying that x killed y’s family.
We find something like this idea in some more recent work on causatives. For example, Richmond Thomason distinguishes “efficient agents” from “automatic agents” (where paradigms of the latter include the wind or a fire) and offers the generalization that “an efficient agent is precluded [from being in the relevant causal relation to the outcome of a process] by another efficient agent between it and the body [of a process]”. (Note, though, that for reasons just given, Thomason’s generalization is too incautious.)
Hart and Honoré explain what they have in mind by a “free, deliberate and informed act”: Intermediate agents that are acting out of “self-preservation” to prevent significant harm coming to them, or who are acting under some legal or social obligation, are not intended to count as engaging in such an act. In this way historians often give some number in the millions when asked how many people Stalin killed. As Thomason would put it, when we use “kill” in this kind of way, various intermediate animate agents are “automatized.”
Abnormal Interventions
Here is a rough first pass at the second idea: If x is a factor in y’s death by doing F, but the death crucially depends on an abnormal intervention between x’s F-ing and y’s death, then x does not kill y. Hart and Honoré offer a version of this idea, generalized to causal responsibility in general:
The basic principle is that normal physical events, even subsequent to the wrongful act, do not relieve a wrongdoer of responsibility but that an ab-normal conjunction of events (in this case the wrongful act and the third factor) negatives causal connection, provided that the conjunction is not designed by human agency.
Excerpted from Killing and Other Dastardly Deeds: Hard Problems for Moral Theory Copyright © 2026 by John Hawthorne. Available from Princeton University Press.
