March 29, 2024

Deriving an ‘is’ from an ‘ought’


Philosophers sometimes draw descriptive conclusions from normative premises. In other words, they sometimes derive an ‘is’ from an ‘ought’. I think this is mistaken.

Note: I’m not talking about the other direction, deriving an ‘ought’ from an ‘is’, which is more often discussed.


  1. Heller (2004) against modal realism: If modal realism is true, then we are never morally required to save drowning children. After all, if modal realism is true, then there will be the same total amount of good and bad in the universe no matter what we do. At most, we can change which of the infinitely many equally-real worlds is actual, which is merely an indexical fact that has no moral significance. But we are sometimes morally required to save drowning children. So modal realism is false.

  2. ??? against panpsychism: If panpsychism is true, then we are morally required to look out for the welfare of electrons. After all, if panpsychism true, then electrons are sentient, and sentience is sufficient for moral patienthood. But we are not morally required to look out for the welfare of electrons. So panpsychism is false.

    Note: I don’t know of anyone who defends this argument in print. Gottlieb & Fischer (forthcoming) object to the argument, but they don’t object to its form. They just deny that panpsychism has consequences for how we are morally required to act because we don’t know enough about what affects the welfare of electrons.

  3. Prior (1959) against eternalism: If eternalism is true, then you shouldn’t be relieved when bad things are in the past. After all, if eternalism is true, then the past, with all the good and bad things in it, is just as real as the present. But you should be relieved when bad things are in the past. So eternalism is false.

  4. Bostrom (2006) against the unification thesis: If the unification thesis is true, then whether it is wrong to torture someone depends on whether there is an exact copy of that person’s brain somewhere else in the universe being tortured. After all, if there is such a copy, then according to unification, there will be no additional suffering if you torture the person. But it is wrong to torture someone whether or not there is an exact copy of their brain anywhere else in the universe being tortured. So the unification thesis is false.

  5. Simon (2017) against manyism: If manyism is true, then the amount of concern you should have for an “experience cloud” (the set of many different conscious entities that, according to manyism, are associated with what would normally be regarded as a single person) increases exponentially with the corresponding brain’s mass. This is because, according to manyism, the number of conscious entities in an experience cloud increases exponentially with the brain’s mass. But all experience clouds should be given equal consideration, regardless of the corresponding brains' masses. So manyism is false.

  6. Baggett and Walls (2016) for the existence of God: If God does not exist, then there are no objective moral truths. But there are objective moral truths. So God exists.1

  7. Hare (2009) for first-person realism: If first-person realism is false, then I’m not metaphysically special, in which case what’s best for me is not always what’s best for the world, in which case I don’t always have most reason to do what’s best for me. But I always have most reason to do what’s best for me. So first-person realism is true.2

The structure of the arguments

All of these arguments have normative premises and descriptive conclusions. Superficially, they are just instances of modus tollens, which is of course a valid form of argument:

  1. D → N
  2. ~N
  3. ∴ ~D

Here D is a descriptive claim and N is a normative claim.

Note that D → N is itself a normative claim, so this form of argument not only rests on a normative premise but rests entirely on normative premises. (Though I think something fishy would be going on even if it had descriptive premises in additional to normative ones.)

Why the arguments fail

If this form of argument is valid, then why is it mistaken? Basically, because there’s no plausible way to defend both premises at the same time: (1) can be defended only if knowledge of D is prior to knowledge of N, but (2) can be defended only if N can be known independently of D.

I’ll work through one of the above arguments to make this more concrete, but I think parallel comments could be made about the other arguments.

Premise (1) of Heller’s argument says that if modal realism is true, then we are never morally required to save drowning children. His defense of this premise rests on the consequentialist assumption that our moral obligations depend solely on what state our actions leave the entire world. Crucially, this is inconsistent with the moral status of a particular action being something you can know without prior knowledge of its consequences.

Premise (2) of his argument says that we are sometimes morally required to save drowning children. But Heller can’t, and doesn’t, appeal to the consequences of saving downing children to defend (2), because these consequences are one of the things that is in question in the debate over modal realism. Instead, he simply takes (2) to be obvious. But given what he himself says in defending (1), to the extent to which we are uncertain about modal realism, we should also be uncertain about (2).

We can frame this as a dilemma. The truth of N either depends on D or not.

  • Horn 1: The truth of N depends on D. Then there is no non-question begging way to defend ~N, since any defense must take a stance on the truth of D.
  • Horn 2: The truth of N is independent of D. Then D → N is indefensible, since the only plausible defense of it rests on the assumption that the truth of N depends on the truth of D.

Heller’s argument is undermined by this dilemma, and I think the other example arguments are, too.3

To be clear, this dilemma applies only to arguments that fit the above pattern, including the details about how the premises can be plausibly defended. This is not something that follows simply from the logical form of the arguments. I’m not saying that every argument which uses modus tollens fails, which obviously isn’t true.

Further comments

  • The proponents of these arguments make much of the counterintuitive normative consequences of the given descriptive claims. But apart from examples 6 and 7, the descriptive claims from which the normative consequences follow are themselves counterintuitive. So there’s a perfectly good explanation of why, say, it’s counterintuitive that we are morally required to look out for the welfare of electrons if panpsychism is true: it’s because it’s also counterintuitive that electrons have welfare.

    Maybe there are cases when counterintuitiveness counts against a normative view, but it doesn’t if there’s a perfectly good explanation of why it is counterintuitive. In most or all of these cases, there is such an explanation.

  • This form of argument is easy to parody. For example, suppose that I believe both that we have a moral obligation to help those in need if we are able to do so, and that I in fact have no moral obligation to give away money because there are no effective giving opportunities. Then someone gives me evidence that, actually, there are many effective giving opportunities. I then point to my moral beliefs to “prove” that this can’t be right: if there were such opportunities, then I would be morally required to do more than I believe I am morally required to do. Clearly something is going wrong here.

  • Intuitively, I’m motivated to reject these arguments simply because it seems obvious to me that normative beliefs aren’t a guide to physical reality. So I accept something like the following principle:

    If you have a set of inconsistent descriptive and normative beliefs, you should always revise the normative beliefs to fit the descriptive beliefs, never the other way around.

    Though, strangely, I didn’t need to appeal to any such principle in section 4 when arguing that the arguments are mistaken. This makes me suspect I didn’t identify the root problem with the arguments.

  • You might be tempted to object to this form of argument on normative anti-realist grounds. The idea is something like: normative claims aren’t really about the world, but are instead about us and our attitudes, so they aren’t a guide to objective features of the world.

    But I don’t think that this is a good objection. Even if normative anti-realism is true, some combinations of normative claims are about objective features of world and not about us. For example, in saying (1) that utilitarianism is true and (2) that φ-ing is right, I’m committing myself to the descriptive claim (3) that φ-ing maximizes happiness. It makes no difference whether utilitarianism is objectively true.

  • Yalcin (2012), inspired by Forrester (1984), presents another argument that attempts to derive an ‘is’ from an ‘ought’: (1) if you are going to kill me, you should kill me gently, (2) you should not kill me gently (because you should not kill me at all), so (3) you are not going to kill me. The argument is clearly mistaken. Yalcin takes this as a reason to doubt the general validity of modus tollens (!). Anyway, I think that whatever is going wrong with this argument (probably a scope ambiguity) is different from what is going wrong with the above arguments.

  • Dorr (2002) also presents an argument which attempts to derive an ‘is’ from an ‘ought’: (1) if lying is wrong, the souls of liars will be punished in the afterlife, (2) lying is wrong, so (3) the souls of liars will be punished in the afterlife. He doesn’t claim that this argument is sound. He just claims that you could have reason to believe (1) and (2) (for example, on the basis of testimony) even if you don’t already accept (3), and that this would give you reason to accept (3). I think Dorr is right about this case, and what I say in section 4 would have to be complicated to accommodate this.

    Note that testimony could give you good reason to accept the premises of all of the above arguments, too. My best guess at what’s going on here is that the person who is giving you the testimony needs to already have the relevant descriptive knowledge in order to be reliable, and their testimony tacitly includes this descriptive information. Since none of the proponents of the above arguments attempts to defend their premises on the basis of testimony, this won’t save any of them.4

  1. I’m following the presentation of the SEP article on moral arguments for the existence of God. I haven’t read the book. ↩︎

  2. Yes, this is as crazy as it sounds. Hare acknowledges without responding to the concern that this is wishful thinking (p. 57). In an endnote he refers to Heller’s argument against modal realism as an example of “is” from “ought” arguments in the philosophical literature (p. 102). ↩︎

  3. There is a further argument that helps reinforce this dilemma. Suppose, against what the dilemma suggests, that both premises of the argument are a priori. So, in particular, ~N is a priori, along with D → N. Then the conclusion of the argument is also a priori, since it’s entailed by a priori premises. But the conclusion is a descriptive claim and so is a posteriori, not a priori. So at least one premise of the argument is a posteriori. Here it is most natural to say that it is ~N that is a posteriori, as suggested by horn 1 of the dilemma.

    This argument isn’t completely decisive. The relevant descriptive claims at issue in the example arguments are the sort of claim that philosophers argue about from the armchair, so some philosophers might claim they’re a priori despite being descriptive. Still, this plausible for at most some of the descriptive claims (modal realism?), not all of them. For example, whether panpsychism is true depends on the true theory of consciousness, which had better be posteriori. ↩︎

  4. I know of only two explicit discussions of ‘is’ from ‘ought’ arguments in the philosophical literature. The first is from Parfit in a Cogito interview (1995). Parfit considers and rejects the following argument: (1) some people deserve punishment, (2) if determinism is true, then no one deserves punishment, so (3) determinism is not true (p. 158). He doesn’t attribute this argument to anyone, but mentions it in response to a question about whether “our views about metaphysics are likely to depend on our views about ethics, rather than vice versa” (p. 157). Parfit goes on to say that, like me, he doubts that this sort of argument could ever work. The other brief discussion is from Richard Chappell in a blog post on moral cluelessness. Chappell mentions the same argument against determinism and, like Parfit, says that it doesn’t work. ↩︎