A New Paradox of Deontology
Why only consequentialism combines normative authority, guidance, and concern for rescuable victims
Scheffler’s classic “paradox of deontology” asks how it can be wrong to minimize morally objectionable actions: If killing is so bad, shouldn’t we endorse one killing to prevent five? But deontologists can naturally respond that they don’t see killing as a bad to be minimized, but as a wrong to be prohibited in each instance.
My latest draft paper, ‘Preference and Prevention: A New Paradox of Deontology’, argues that deontologists face a deeper problem: once an agent has gone ahead and killed one in an attempt to prevent five other killings, deontologists cannot accommodate how strongly we should hope that her attempt succeeds.
For clarity, here are the four possible outcomes to compare (against a common backdrop of Protagonist choosing whether to kill one to prevent five other killings):
Five Killings: Protagonist does nothing, so the five other murders proceed as expected.
One Killing to Prevent Five: Protagonist kills one as a means, thereby preventing the five other murders.
Failed Prevention: As above, Protagonist kills one as a means, but in this case fails to achieve her end of preventing the five other murders. So all six victims are killed.
Six Killings: Instead of attempting to save the five, Protagonist simply murders her victim for the sheer hell of it, just like the other five murderers. So all six victims are killed.
Here’s the argument in a nutshell (using ‘≻’ to indicate ideal preferability, and ‘≻≻’ to indicate vast preferability, where this is stipulated to mean preferability of a magnitude strictly greater than the extent to which we should prefer one less generic killing):
(1) Deontic constraint (for reductio): Protagonist acts wrongly in One Killing to Prevent Five, and ought instead to bring about the world of Five Killings.
(2) If an agent can bring about W1 or W2, and it would be wrong for them to bring about W1 (but not W2), then W2 ≻ W1. (key premise)
(3) Five Killings ≻ One Killing to Prevent Five. (from 1, 2)
(4) One Killing to Prevent Five ≻≻ Failed Prevention. (premise)
(5) Failed Prevention ⪰ Six Killings. (premise)
(6) Five Killings ≻ One Killing to Prevent Five ≻≻ Failed Prevention ⪰ Six Killings. (3 - 5, transitivity)
(7) It is not the case that Five Killings ≻≻ Six Killings. (definition of ‘≻≻’)
# Contradiction (6, 7, transitivity).
As I argue in the full paper, “we should regard (4) as an unassailable moral datum, the rejection of which would entail severe moral disrespect to the five extra murder victims.” After all, Failed Prevention contains everything that’s morally objectionable about One Killing to Prevent Five, plus five additional, completely gratuitous killings. There is no respect whatsoever in which Failed Prevention is morally preferable. So the only way to lack the preference demanded by (4) is to fail to much care about those five additional killings. I take that to be plainly morally unacceptable. So everyone must accept (4).
But once you prefer Five Killings ≻ One Killing to Prevent Five, there’s not enough “room” left between Five Killings and Six to fit in the stronger preference that (4) demands. The preference chain indicated in (6) is incoherent, entailing a contradiction. It turns out that we can only accommodate (4) if we instead accept the consequentialist’s preference ordering, on which One Killing to Prevent Five ≻ Five Killings.
I then argue that this is effectively to embrace consequentialism (in substance if not in name). In particular, I argue that:
(I) Denying (2)—e.g. by claiming that we should prefer wrong actions to be performed— would rob deontic verdicts of their normative authority.
(II) Insofar as one can distinguish narrowly “act-directed” from broader “state-directed” motivations, the latter have greater normative authority. (Compare ‘Constraints and Candy.’)
(III) Deontologists can’t escape my argument by withdrawing to a narrower/moralized conception of “preference”, or by refraining from making any claims about preferability at all, because (i) any true view must be coherently completable, and (ii) there are clearly moral truths about broad preferability, e.g. that a decent moral agent should prefer that a child not be struck by lightning rather than be struck (all else equal).
The only way out for deontologists, that I can see, would be to invoke rampant incommensurability: perhaps the deontic reasons can’t be meaningfully weighed against the reasons of beneficence, leaving us without any basis for forming all-things-considered preferences in cases where the two conflict. But, as I conclude:
Rather than affirming deontic constraints, this view transforms them into (indeterminate) moral dilemmas. One might say that we “deontologically ought” to respect the constraint, but it would be equally true to say we “consequentially ought” to violate it, and quietism rules out the claim that we definitively ought to care more about the constraint than about the consequent benefits of violating it. All we can say is that we ought to feel (irreparably) torn, which leaves us entirely lacking in practical normative guidance.
If we are to combine normative authority, normative guidance, and adequate respect and concern for the five rescuable victims (after the other one has already been killed as a means), then we need consequentialism.
Any objections?
Update: see ‘Deontology and Preferability’ for essential background that wards off the most common objections. The paper is now published in Free & Equal. My post ‘The Curse of Deontology’ provides a broader overview of the paper, including the distinction I introduce between “quiet” and “robust” deontology, and more on the costs of rejecting the “key premise” (2).
Matthew was taking me through this the other evening - clever argument! Going through it with him, I wanted to say something like this.
Distinguish between two senses of ‘preferability’: axiological preferability, let’s say, is the normative preferability of worlds (centred or uncentered). Deontic preferability, let’s say, is the normative preferability of a particular centre - an agent at a time (you might be that agent, or you might not be.) In your case, let that agent be the person whose act we’re evaluating, and let that time be prior to the evaluated action.
If we stipulate that we’re using deontic preferability throughout the argument, I want to reject (4): Five Killings = One Killing to Prevent Five, because we’re only evaluating the centre, and the evaluation is the same for either outcome.
If we stipulate that we’re using axiological preferability throughout, then I want to reject (2): ‘being wrong’ is a deontic property - but axiologically speaking, there might be cases in which W1 and W2 are equally preferable, or else W1 is preferable to W2.
To motivate the coherence of this idea: suppose someone’s hand is stuck in a properly functioning toaster turned up to six, and someone points in the direction of the toaster as asks, “is that good?” Contextually, the fact that their hand is burning (bad) is more salient than the fact that their hand toaster is functioning properly (good qua toaster), and so I would be tempted to answer ‘no’. But if I were genuinely confused about the referent of the ‘that’, I wouldn’t be able to answer; if they rolled their eyes and said, ‘but overall, is that bad?’, that wouldn’t help me. I’d be stuck until I knew which thing they were talking about.
In like fashion: I’m thinking that because preferability talk is, for contingent reasons, more axiologically coded, I feel pulled to read the argument in that light - but it might still be that there are two irreducibly distinct senses of ‘preferability’ (and ‘vast preferability’) at play, and that if we disambiguate, the argument can’t go through with either to the deontologist’s liking.
The positive motivation for the idea that ‘preferability’ is ambiguous between these two senses: (1) if it saves deontology, that’s to its credit; (2) by the look of the comments, it looks like a lot of people are wanting to say something like this - that on the one hand there’s the preferability of acts, and on the other hand, there’s the preferability of worlds. You point out that worlds (centred and uncentered) include both acts and outcomes (and, with centred worlds, acts by agents at times): but centres alone only include agents at times!
Apologies in advance- this is very jaggedy and probably super confused.
[no idea how to avoid incommensurability though; does any of this help with that?]
I have three questions:
(1) Are there actually any (secular) deontologists? Or is deontology just something that a subset of professional philosophers continue to pretend to believe in because that was the team they picked when taking Ethics 101, it would be embarrassing to change their mind in public, and anyway they need to continue to pretend to believe in deontology in order to keep their jobs as bioethicists? (This thought brought to you courtesy of the "Treating Persons as Means" page on SEP, which gives the strong impression that the author can't find a way to make the concept intelligible but doesn't want to admit it.)
(2) >Insofar as one can distinguish narrowly “act-directed” from broader “state-directed” motivations, the latter have greater normative authority.
("More of a comment than a question...") This is the entire point of your disagreement with deontologists, so simply stating it like this appears to be begging the question. (This criticism applies more to this post than to your draft, which considers act-directed motivations at somewhat greater length, even if deontologists may still feel the draft doesn't provide adequate support for this claim.)
(More substantially, and with an actual question...) Nye, Plunkett & Ku seem to be arguing that deontology can be grounded in fittingness considerations — the words "fitting" and "fittingness" appear on practically every page of their paper. By contrast, you seem to use the language of fittingness here only very reluctantly: you don't use it at all in these posts, and in your draft you use it only when replying to them directly and quickly move on from talking about "what is fitting" to talking about "what matters" or "what one ought to do." To what extent do you think your divergence from Nye, Plunkett & Ku should be characterized in terms of divergent views on what is fitting, vs. divergent views on the normative force of fittingness considerations?
(3) >All we can say is that we ought to feel (irreparably) torn, which leaves us entirely lacking in practical normative guidance.
"We ought to feel irreparably torn, and lacking in practical normative guidance" seems to me an entirely accurate description of the human condition, quite independently of our views on deontology. Is there any reason to believe it is *wrong*, as opposed to just an unappealing conclusion for a certain variety of moral philosopher?
Incidentally, I think "ipso fact" on p. 22 of your draft should be "ipso facto." (An ipso facto typo?)