I know my life is good because I’m the one living it. In the same spirit as Descartes’ “I think, therefore I am”, I have greater certainty about the goodness of my experiences than I do about any fact about the external world.

Some theories of population ethics deny this. Average utilitarianism might say:

Hey Michael, I know you’re happy, but other people are happier than you, so it would be better for you not to exist. The experiences you think are good are actually bad, because they’re not good enough.

To that, I say:

No, you’re wrong, Mister Average Utilitarianism. I’m the one living my life, I can tell how good my experiences are, and you don’t get a say in the matter. My good experiences don’t turn bad just because other people exist. I am the ultimate judge of the goodness of my experiences.1

In Why Care About the Moral Law?, Richard Y. Chappell wrote: “Morality should serve the interests of moral subjects, not the other way around.”2 If a theory of ethics says that your good life isn’t good after all, then how is that theory serving your interests? And you’re the one who experiences that goodness, not anyone else, so how can that theory be serving anyone else’s interests?

Cross-posted to the EA Forum.

Contents

The separability principle

If I tell you “I’m torturing an animal in my apartment,” do you go “well, if there are no other animals being tortured anywhere in the world, then that’s really terrible! But there are some, so it’s probably not as terrible. Let me go check how many animals are being tortured.”

(a minute later)

“Oh, like ten billion. In that case you’re not doing anything morally bad, carry on.”

I can’t see why a person’s suffering would be less morally significant depending on how many other people are suffering.

—Kelsey Piper3

(Philosophers sometimes refer to “the Egyptology thought experiment”,4 which conveys the same message, but I like Kelsey’s version better.)

This is the separability principle: the goodness or badness of a particular person’s5 experience is evaluated on its own; it does not matter what experiences other people are having (except insofar as people’s experiences affect each other).

Most theories of population ethics are incompatible with separability

If I am the ultimate judge of my own subjective experience, and you are the ultimate judge of your experience, and Sally is the ultimate judge of her experience, then only two forms of interpersonal aggregation are permitted:

  1. Aggregation is impossible in general. Outcomes cannot be ranked when ranking them would require trading off one person’s welfare against another’s.
  2. Aggregation is linear.

Views that are ruled out

If the separability principle is true, then many theories of population ethics must be false. Here are some notable ones:

The average view says that the value of an outcome equals the average of everyone’s welfare. On this view, the value of my welfare depends on whether other people are better or worse off than me. If I’m happy but less happy than average, then the average view says the world would be better without me.

Critical-level views say that positive welfare only counts if it exceeds a certain “critical level”. If you say your life is good, but your welfare falls below the critical level, then this view says: no, you’re wrong, your life is not good.

Suffering-focused views give more weight to suffering than to a comparable amount of happiness. If experience A makes me happy and experience B makes me suffer, and I judge them to exactly cancel each other out, then the separability principle says that “the universe” cannot put more weight on B than on A. Note: Separability does not rule out the (probably-true) factual claim that it’s easier to cause suffering than to cause an equal amount of happiness, or that the maximum suffering a human can experience outweighs their maximum happiness.

Interpersonal lexical views say that some greater form of suffering counts for more than any amount of a lesser form—an hour of torture is worse than any number of stubbed toes. Lexical orderings are not fully ruled out by separability, but for them to survive, the lexicality must live within a single person. If you, personally, would prefer a mountain of smaller sufferings over an hour of torture, then lexicality holds. But if you would accept an hour of torture in order to relieve a lifetime of mild chronic pain, then the universe must accept your judgment.6

Views that innately value variety of experience (as a matter of ethics, not metaphysics) give less weight to an experience if an identical experience happens elsewhere. If two identical but physically separate minds go through identical experiences, does that count once, or twice? There is a metaphysical question of whether the two minds are having one or two phenomenal experiences, and either answer is compatible with separability. However, if the answer to the metaphysical question is two, then the answer to the ethical question must also be two.

Views that are permitted (other than the total view)

At first glance, separability seems to entail the total view of population ethics. But it doesn’t necessarily. Some other views are compatible with the separability principle:

Certain person-affecting views. A person-affecting view (PAV) is potentially compatible with separability, but only if they accept some strange principles.

The first step to reconciling PAV with separability is to say that two outcomes are incomparable if they do not contain the same people—therefore, we cannot say that creating a new life is a good thing. However, incomparability alone is not sufficient. PAV encounters another problem: if I say it’s better that I was born, then how can PAV say that my existence is incomparable to non-existence? PAV’s way out is to say that my judgment only counts after I’m born; if I don’t exist yet, then there is no “me” to make the judgment, so it’s not bad if I’m never created. This requires biting another bullet: the judgment changes depending on what time it is. PAV declines to say whether my birth would be a good thing—until after I’m born, at which point it is forced to agree that my life is good. (This problem arises for PAV even without accepting separability.)

Views that accept “Pareto plus” improvements, but nothing else. Separability entails the “Pareto plus” principle:7 if a person is made better off or a happy person is created, and no one is made worse off, then the resulting outcome is better. However, separability still allows for the possibility that two outcomes are incomparable if neither one is a Pareto plus improvement over the other. If a change makes Alice better off and Bob worse off, such that the aggregate welfare is higher, then separability does not require that we call the change an improvement.

Total incomparability is not permitted by the separability principle, because total incomparability robs me of the right to judge my own welfare. If my happiness improves and nothing else changes in the world, then the end result must be better.

Rejection of objective axiology.8 You could say: if there is such a thing as “goodness from the point of view of the universe”, then the goodness of the universe goes up when one individual’s welfare improves. However, there is no such thing as goodness from the point of view of the universe.

The only alternative to totalism is incomparability

Separability is consistent with some non-totalist views of population ethics, as listed above. What they have in common is that they all treat some pairs of outcomes as incomparable: neither is better than the other, nor are they equal.

The total view follows from separability plus one more premise:9

Completeness: For any two outcomes A and B, either A is better than B, or B is better than A, or A and B are equal.

A caveat: To get the standard version of the total view, we need an additional assumption that each person’s welfare can be described by a utility function—or, equivalently, that each person’s preferences satisfy the four Von Neumann-Morgenstern axioms. This assumption may be false.10 One way it could be false is if people have lexical preferences: they would choose any number of stubbed toes over an hour of torture. (In technical terms, that means their preferences violate the VNM continuity axiom.) If people have lexical preferences, then their preferences cannot be aggregated as total view requires, even if we accept separability and completeness. But we can get something close: add up everyone’s torture, add up everyone’s stubbed toes, and rank outcomes by torture first and stubbed toes second. McCarthy, Mikkola & Thomas (2020)11 show that utility aggregation still works under lexical preferences, although aggregate utility must be represented as a list of numbers rather than a single number.

In other words, if individuals don’t have VNM utility functions, then we don’t get the standard total view, but we do get something we might call a quasi-total view12—the value of an outcome is the aggregation of every person’s well-being, but the result of that aggregation isn’t a single “utility” number.

The separability principle is difficult to deny: you are the arbiter of your own experiences, and no force in the universe can tell you otherwise. Separability alone rules out most possible theories of population ethics. The only remaining theories are totalism, quasi-totalism, and those that refuse to compare most outcomes.

Notes

  1. I previously wrote about how sometimes I don’t know whether I’m having a good experience. That observation raises an important puzzle, but it does not contradict my point. I can’t always introspect on my own welfare—because I’m introspecting on a memory of a recent experience, not the experience itself; or because introspecting changes my experience—but I am the only person who can directly experience my welfare. ↩

  2. Chappell was paraphrasing an older quote by William K. Frankena: “Morality is made for man, not man for morality.” ↩

  3. The source is a Tumblr post that I can no longer find. ↩

  4. The following statement of the Egyptology thought experiment comes from Teruji Thomas in The Oxford Handbook of Population Ethics (2022), chapter 12:

    Belinda wonders whether it would be morally permissible to have a child. She considers many factors: her own feelings and desires; whether the child is likely to have a good life; whether the child will have a positive or negative impact on other people who now exist, or who might exist in the future. But one thing she need not investigate is how well off the ancient Egyptians were. It would be one thing if Belinda’s decision now could change their welfare, by enacting some kind of posthumous benefit or harm. But suppose this is not the case: the welfare of the ancient Egyptians will be unaffected, and Belinda knows it. Then their welfare is simply irrelevant to Belinda’s choice. She thinks so, anyway.

    ↩

  5. I use the term “person” broadly, to include all moral patients to whom population ethics applies, which may include non-humans, or individuals who are not persons by some other definition of “person”. Some people believe that population ethics operates differently for persons and non-person moral patients; for example, Robert Nozick suggested that utilitarianism may be true for non-human animals but not for humans. My arguments are agnostic with respect to that belief. However, for simplicity of language, I will sometimes refer to “people” when what I really mean is “entities for whom the person-relevant theory of population ethics is relevant”; strictly speaking, those entities might not be persons. Similarly, I will prefer the term “people” over “persons” for linguistic simplicity, and because “persons” implies precision at an incorrect target (the entities I am referring to are not necessarily persons).

    I realize that this footnote was not linguistically simple, but I am hoping that I can reduce analytic-philosopher-speak in the essay’s body by cramming it into this footnote. ↩

  6. Claude offers examples of painful operations people used to undergo in the pre-anesthetic days in exchange for long-term benefits, including cosmetic nose reconstruction, cataract couching to restore sight, and anal fistula removal. Citations are included in the linked Claude chat. ↩

  7. The term comes from:

    Sikora, R. I. (1978). Is It Wrong to Prevent the Existence of Future Generations? In R. I. Sikora & B. Barry (Eds.), Obligations to Future Generations (pp. 112–166). Temple University Press. ↩

  8. Axiology is the study of goodness, just as morality is the study of right and wrong. I’m being precise because one might reject objective axiology while still accepting objective morality: it could be morally wrong to take an action, even if there is no objective way to say whether that action makes things better or worse. ↩

  9. I’m speaking informally when I say that one more premise is required. Exactly what premises you need depends on how you formally define the separability principle. For the purposes of this essay, I prefer to talk about separability as a strong intuition, not an exact formal definition. For a formal analysis, see:

    Thomas, T. (2022). The Oxford Handbook of Population Ethics. Chapter 12: Separability and Population Ethics. ↩

  10. For the record, I expect that humans in reflective equilibrium (if there is such a thing) do have VNM utility functions. ↩

  11. McCarthy, D., Mikkola, K., & Thomas, T. (2020). Utilitarianism with and without expected utility. doi: 10.1016/j.jmateco.2020.01.001 ↩

  12. That’s what I chose to call it; I don’t know if there is a standard term for this. ↩