There's something that bothers me with bracketing, that I'd be interested to hear a reply to. It seems to affect bottom up bracketing as you have presented it here. I am unsure if it affects top down bracketing as well, but it sounds like it might.
Suppose I have an action available to me that will have the following 3 effects, impossible to disentangle:
I get a tasty snack right away.
Some broken glass is left somewhere. I am confident it will remain buried for 100 years, but after that I think it is highly likely to resurface and cause someone a nasty injury, due to where it is located.
Some effect on the long term future which I think will be significant in expectation, but which I am clueless about the sign of.
My understanding is that bracketing would say not only is this act permissable, but that I should do it. This is because it benefits me, and I am clueless about its effects on every other beneficiary (the person who is injured by the glass in 100 years may well be compensated by a benefit from 3).
But this feels wrong! If I just consider the effects that I am not clueless about, the bad seems to clearly outweigh the good. The same intuition that motivated bracketing in the first place would seem to say that this act is wrong.
I can buy the plausibility of a theory that says it is indeterminate whether I should take this act or not, due to the uncertainty bundled up in 3. And I can buy a theory which says the act is wrong (since that is what my intuition strongly points towards). But a theory that says it is wrong not to take this act feels like it has gone wrong somewhere.
Apologies if this is naive and has already been addressed at length somewhere. I would be interested to know how proponents of bracketing respond to this.
Hi Toby––thanks, it's a good question. If you expected your broken glass to badly hurt a person tomorrow, I suppose you wouldn't feel clueless about the net effect on the person's lifetime wellbeing. If so, why feel clueless if the nasty injury occurs later? If injuries from broken glass in the ground are still happening in that future society, it can't have changed that much. (Otherwise, why assume that the broken glass would cause harm, rather than nothing or a benefit, in the first place?)
But suppose we are clueless about the net effect on the injured person's lifetime wellbeing. In that case, if we "ask" the person (from our ex ante perspective, given our imprecise credences) whether she prefers that we remove the glass rather than leaving it in the ground, her welfare-based answer is "no." Consequently, it would seem adequate to bracket her out and bracket you in (from a "consequentialist" point of view). The person supplies no contrastive reason favoring either option; you do.
I should also say, though, that not leaving the glass in the ground seems quite overdetermined by the reasons we have. E.g., deontological reasons not to actively cause serious foreseeable harm to a person (or violate their right to bodily integrity), and reasons not to risk doing so, provide a pretty clear verdict here. (Such reasons are relevant when it comes to actively causing the existence of future people, too.) You can't cut a person without their consent and then (over)compensate them for it.
But suppose we are clueless about the net effect on the injured person's lifetime wellbeing. In that case, if we "ask" the person (from our ex ante perspective, given our imprecise credences) whether she prefers that we remove the glass rather than leaving it in the ground, her welfare-based answer is "no." Consequently, it would seem adequate to bracket her out and bracket you in (from a "consequentialist" point of view). The person supplies no contrastive reason favoring either option; you do.
Just to check: then we should also bracket me out in the snack case, right? I presumably don’t know that eating the tasty snack determinately improves my lifetime welfare overall. So your version of bracketing wouldn't actually recommend #1, as Toby assumed. It would remain silent in the decision situation he depicted.
To be action-guiding, person-based bracketing requires "knowing" the net effect on the lifetime well-being of at least one (lifetime) person, right?
I’m one of the judges of the competition. My comments shouldn't be taken as a full review of a post. And, unfortunately, I won’t have capacity to comment on every post or engage with all replies. Thanks so much to everyone who entered!
I find the case for bottom-up bracketing very well-argued and compelling.
I’d like to see more engagement with non-identity issues that threaten the action-guidingness of bottom-up bracketing.
Suppose a not-yet-existent Alice will exist with a happy life in worlds where I do A and does not exist in worlds where I do B. If Alice’s welfare is zero in B-worlds, then she prefers A over B, and so enters the aggregation. But aggregating over all such future beings presumably leaves the aggregate comparison indeterminate (e.g., maybe these are all the beings that will live in a utopia if x-risk is averted, and we’re clueless about how A and B affect x-risk).
Instead of setting Alice’s welfare to zero in worlds where she doesn’t exist, we could use some kind of counterpart relation that ends up making Alice’s comparisons between worlds in which she does and doesn’t exist indeterminate (e.g., Michael’s comment here). Since the contingent persons would then all be bracketed out, you might actually end up with a determinate comparison that way. But such views seem in tension with the post’s person-centred perspective.
Hi Jesse and Anthony, thank you––regarding your second point, I think it's indeed a nonstarter, from a person-centered point of view, to make up a "counterpart" for Alice. One might think the same regarding your first point: The assumption that Alice's welfare is zero in B-worlds, where she does not exist, is made up. As Mogensen puts it (p. 1027): "We may believe, for example, that a person’s welfare level is defined in outcomes in which she does not exist, and that it takes a value of zero (Greaves & Cusbert, 2022; Holtug, 2001; Roberts, 2003). This would indeed undermine my argument. However, it is also a position that I find incredible, since I find it incredible that non-existent people have well-defined levels of any measurable quantity, be it height, weight, or welfare, any more than logic has a well-defined temperature."
If this is right, it immediately gets us out of the worry expressed in your first point: Alice is indeterminate between the options, so that she is bracketed out (i.e., she provides us with no contrastive reason to favor an option rather than another).
However, it seems to me that this cannot be right. Suppose Alice net-suffers terribly in A-worlds. Surely, this gives us a contrastive reason to choose B rather than A, as far as Alice is concerned. Pace Mogensen, I believe we should be able to get to this verdict without assuming that Alice's nonexistent welfare is literally zero in B-worlds: In the scenario where Alice exists (we choose A and x-risk is averted), she has a welfare-based perspective on how worlds where she doesn't exist compare/rank relative to the world where she exists and suffers, and the comparison prefers nonexistence. (In other words, Alice supplies a welfare-based objection against our choosing A, grounded in the welfare-based take she has on our choice in the scenario where she comes into being.)
So, we have an Alice-indexed contrastive reason not to choose A of the following two-dimensional magnitude: (p ∈ [0, 1]; her negative welfare, say -10, or perhaps [-20, 0)). This reason has a determinate direction (favoring B rather than A) but an indeterminate strength. Now suppose that, in choosing A, we also determinately save Mel from malaria, providing her with a +5 welfare benefit, say. Mel thus supplies a contrastive reason of determinate strength favoring A rather than B. The sum of contrastive reasons is now indeterminate, so that either choice is permissible.
This is the correct verdict, I think. Both intuitively and from person-centered "first principles," impartial altruists should not favor either option in this case. What we have here is one moral patient favoring A, another one opposing A, and the balance of their contrastive reasons being unclear. It makes sense, intuitively and theoretically, that either option should be permissible. (On scope-sensitive forms of altruism, this verdict may change depending on the number of people like Alice and the number of people like Mel involved in the case, as one would expect.)
It is important not to lose sight of the fact that overall indeterminacy is not to be avoided tout court. We should avoid it where, and only where, it does not actually obtain from an impartial altruistic point of view. Consider the following variation on the above case. Suppose our uncertainty regarding Alice is deeper: She might also have a net-happy life in A-worlds, who knows. In that case, Alice provides us with no contrastive reason to favor either option; if we "ask her," she's fine with either choice (given her welfare stakes and our credences). Assuming that, in choosing A, we determinately save Mel from malaria, impartial altruists should not reach a verdict of overall indeterminacy: Mel is not fine with either choice and should thus determine our overall verdict.
The empirical situation (regarding x-risk, various ways in which human survival may be realized, and s-risk) is, of course, complicated. My post does not address the question of what kinds of cases we happen to be facing in the actual world today. It aims to make a contribution to the prior conceptual, moral, and decision-theoretic questions.
I'd be curious about how you think about transworld identity and personal identity and to get your take on the following proposal for counterpart relations. I expect that bottom-up bracketing would rarely be action-guiding in practice without counterpart relations, because of huge numbers of contingent moral patients pointing in each direction, but counterpart relations could make some version action-guiding in practice by allowing us to bracket out far more of them.
This is a pretty rough sketch.
One way I'm thinking of grounding counterpart relations is by looking for measures that naturally generalize our concepts of transworld and personal identity. For example, Alice in A is the same person as Alice in B and her past self because of common causes of the features important to Alice's identity, like her memories, psychological dispositions, etc.. Maybe we'd say specifically due to the preservation or inheritance of the physical structures (her brain) responsible for her psychological features from her past selves.
All of this is a matter of degree, because Alice changes over time and could diverge between worlds A and B. Almost all of the particles in her neurons and elsewhere in her body are replaced over time, and which particles make her up will diverge across worlds. In cases of amnesia, brain damage and normal human aging and experience, people can change fairly dramatically in their psychological features. We might say that transworld and personal identity matching are themselves graded. Alice is kind of both the same person and kind of a different person, and more a different person the larger the divergence and the less is inherited.
Then, Alice and Bob, if both human, share more psychological dispositions than Alice and a chicken, and this can be traced back to more recent common causes of their psychological dispositions, i.e. more recent common ancestry between any two humans than between a human and a chicken, passed through genes. This could let us preferentially match counterparts that are more closely related/trace more of their genes that account for psychological features back to the same more recent ancestors.
Transworld identity is just a special case of this kind of causal counterpart relation where the common causes of psychological features are especially potent or of a particular kind, e.g. the "same" brain, which we can trace through spacetime. The gap between distinct persons is admittedly generally much larger than between the same person across worlds or at different points in time, but if it's more a difference in degree than kind since both are graded, then this would give us reason to accept or reject both transworld/personal identity and causal counterpart relations together, rather than accept transworld/personal identiy and reject causal counterpart relations.
One potential supporting argument for the generalization from direct inheritance of psychological features to common causes of psychological features for counterpart relations:
While adult Alice inherits some psychological features from her infant self and infant brain, many of her cognitive capacities as an adult — which are plausibly important to her identity — were not directly inherited from her infant self or infant brain (or fetal brain), but were caused in part by general and often region-specific DNA-driven processes to restructure her brain. Furthermore, the vast majority of the particles in her neurons and brain get replaced even if most neurons are preserved, and it's (I'd guess) the DNA in those neurons are necessary for ensuring their function even as the particles in them turn over. So, Alice's psychological features and identity over time and across worlds come in large part from her DNA. The same has been typically true for conscious biological beings on Earth so far.
The persistent DNA particles over time cause Alice's psychological features over time and across worlds. Copies of them from her ancestors cause many of her relative's similar psychological features over time and across worlds. What grounds her identity is common causes of her psychological features, but going back up the causal chain, we find common causes of psychological features between moral patients, too, not just within, through copies of the same DNA.
If common causes, like someone's DNA, of shared psychological features across worlds are what ground transworld identity, then we should be able to use common causes, like a common ancestor's DNA from which copies were inherited, of shared psychological features to ground counterpart relations across worlds to generalize transworld identity, too.
We can use other common causes of shared psychological features to ground counterparts more generally. Or even common causes of individuals' existences across worlds, e.g. if someone is deciding between dog breeders to breed a puppy for them (or buying a puppy from them and creating demand for further puppies), they are causally downstream of each of the possible resulting puppies, and this further common cause supports matching the possible resulting puppies across worlds as counterparts.
Interesting, thanks––more later, just one (meta-)comment: Maybe, if huge numbers of individuals are pointing in each direction (i.e., if we are clueful enough to be able to pick out these individuals and assign them a contrastive reason of determinate direction but vague magnitude), it would be a mistake for impartial altruists to reach a determinate overall verdict. In cases where the balance of contrastive person-attributable reasons is vague, it seems inadequate to favor either option. In other words, indeterminacy may not be a problem to "solve" in these cases (whether hypothetical or actual); instead, it would be a problem if our theory told us that the situation, in these particular cases, is determinate.
I might be able to share a couple of thoughts on identity and counterpart relations later––it's very tricky business, of course. For now, just one more thought on your apparent intuition (?) that, in the cases we are actually/empirically facing today, we should be able to set far future individuals aside. There might be another, principled way of doing this that may align quite well with the defense of person-centered bottom-up bracketing offered in the post (which justifies its verdicts in terms of each person's ex ante perspective on our choices). Start with the plausible, widely shared idea that no number of mild, barely noticeable headaches can outweigh unbearable, torture-level suffering, i.e., draw a distinction between "lower" and "higher" kinds of suffering (and perhaps an analogous distinction between lower and higher kinds of happiness, à la John Stuart Mill). Then argue that sufficiently tiny probabilities of higher suffering still fall into the lower prospect category, i.e., can be traded off with headaches. (In other words, come up with a way to avoid the kind of fanaticism at issue here.) Finally, argue that our empirical situation happens to be such that, with our actions today, we can only make sufficiently tiny probability-dents in the fates of individual far future people (as represented by us).
Hi Impartial, thanks for this reply! Can't speak for Jesse, but I definitely agree with this:
It is important not to lose sight of the fact that overall indeterminacy is not to be avoided tout court. We should avoid it where, and only where, it does not actually obtain from an impartial altruistic point of view.
So, if bracketing (either TD or BU) is entirely non-action-guiding for the reasons Jesse mentions, that doesn't bear on whether bracketing is a good normative theory. But it does bear on your claim at the start of your post: "Bottom-up, person-centered bracketing can thus preserve principled impartial action guidance".
Hi again––thanks! Yes, I should have been more precise about what the claim is I'm trying to argue: that person-centered bottom-up bracketing provides action guiding (determinate) verdicts in just those cases, whether hypothetical or actual, where altruistic action guidance plausibly obtains, and outputs indeterminacy in the other cases. In other words, the claim is that there are no normatively problematic failures of action guidance.
Is this a version of (weak) actualism as a person-affecting theory? In worlds where Alice doesn't exist, she has no properties like welfare to provide us any reasons or welfare-based objections.
Weak actualism, as in the literature (Hare, 2007, Spencer, 2021 section 6), would say to pick among the options that maximize "self-conditional value". An option is permissible only if it's no worse than any other for the people who (ever) exist in the resulting outcome. In other words, assuming we choose the option A, we compare all the options pairwise on the basis of the reasons provided by the people who would exist given our choice of A,[1] and if A is no worse overall than any other option on these comparisons, then it maximizes self-conditional value.[2]
It sounds similar to what you wrote here, too:
Perhaps, on a person-centered approach to ethics, that is simply true (cf. Boonin 2014 and Pummer 2024): If we choose B, we will have made it the case that (i) no actual person has negative welfare, which they could object to on personal welfare grounds, and (ii) all actual persons have the maximum amount of positive welfare they could have had. (Person p2, once she exists, will have no welfare-based objection to our choosing B: We could not have made her better off. Nor will p1, who will not exist.) On a person-centered approach, we may therefore have maximized people’s welfare in the sense that matters.
I also think maximizing self-conditional value as formalized in the literature is the wrong approach for an actualist when deciding between at least 3 options, because it's myopic, and we should use backward induction on behalf of the people who would exist given an option. This can avoid the repugnant conclusion. I have an unpublished draft based on the idea of backward induction for weak actualism I'd be happy to share, and I discuss the problem a bit here.
Thanks, Michael—briefly for now (I'm short on time today and this week): I think some actualist views deserve to be taken very seriously, but for the purpose of defending person-centered bottom-up bracketing, I meant to leave open what procreative reasons are supplied by the persons we can bring into existence. E.g., while the framing of welfare-based objections/regrets suggests an asymmetric view (focused on preventing miserable lives), antiobjections/gratitudes might give us reasons too (to bring happy lives into existence, or to grant them permitting/justifying/canceling force). I agree myopia is an issue but think avoiding it is doable. I'd be interested in reading your draft.
Bottom-Up Bracketing: A Person-Centered Response to Cluelessness — EA Forum
Bottom-Up Bracketing: A Person-Centered Response to Cluelessness
TL;DR
NOVAH (No Violence At Home) was incubated by Charity Entrepreneurship (now Ambitious Impact) in 2024 to test a promising idea: preventing intimate partner violence through edutainment, in our case a serialised radio drama. Over the past two years we have produced and aired two seasons in Rwanda.
We are currently evaluating our second season through a randomized controlled trial with 2,400 couples in Rwanda in partnership wi...
I am writing up two charity ideas I’ve wanted to investigate, but haven’t been able to, in the hopes someone will investigate them further. If you end up investigating either, I'd be most interested to see what conclusions you come to!
Vulture preservation as a global health intervention
Since the mid 1990's the vulture population in India has been rapidly declining. The reason is due to a few chemicals which are fed to livestock, which is highly poisonous to vultures.
...
This summer, 27 residents came to Berkeley for the first cohort of the Generator Residency, a three-month program for AI safety generalists run by Kairos and Constellation.
Residents receive mentorship from experienced generalists and researchers and are working on projects like running workshops for professionals entering AI safety, building a cross-org...
This competition entry has been selected for publication by the Forum team.
Summary
Kollin, Clifton, DiGiovanni, and Macé (2025) propose two “bracketing” theories for recovering altruistic action guidance under cluelessness, formalizing a response that DiGiovanni (2025) had earlier viewed sympathetically but raised doubts about. On bottom-up bracketing, we set aside beneficiaries for whom it is indeterminate which option is better, and then aggregate over the remaining beneficiaries. On top-down bracketing, we instead evaluate the largest subsets of beneficiaries for which the comparison is determinate. Although bottom-up bracketing yields Kollin et al.’s intended verdict in the motivating donation case (Against Malaria Foundation > Make-A-Wish Foundation, despite long-term cluelessness) and satisfies Super-Strong Pareto, the authors reject it because it fails to extend their baseline consequentialist relation and, in particular, violates Statewise Dominance.
I argue that this rejection is ill-motivated. In their central counterexample to bottom-up bracketing, one person is determinately better off under option A than option B, while neither A nor B is determinately better for the other people. Bottom-up bracketing therefore chooses A. B wins under top-down bracketing only because the other people’s individually indeterminate welfare comparisons are pooled before determinacy is assessed. But if individual persons are the basic sources of altruistic reasons, the bottom-up order is more plausible: Assess each person’s complete represented prospect first, and only then aggregate the resulting person-attributable contrastive reasons (i.e., reasons to prefer A rather than B). Infinite cases reinforce this by showing that anonymous, non-contrastive aggregation can erase morally important facts about how the individuals fare. The resulting view rejects unrestricted Statewise Dominance. This does not undermine its status as an impartial, altruistic, welfarist, and entirely consequence-focused theory; whether it counts as “consequentialist” depends on how narrowly that label is defined.
This defense of bottom-up bracketing also answers DiGiovanni’s (2025) worries about bracketing: Persons’ complete prospects provide a non-arbitrary, morally motivated decomposition of consequences, and indeterminate person-attributable comparisons do not supply any contrastive reason in either direction. Bottom-up, person-centered bracketing can thus preserve principled impartial action guidance where total expected welfare remains indeterminate.
In a final bonus section, I draw out an implication suggested by the infinite cases considered: that DiGiovanni’s epistemological approach—set intuitions and heuristics aside, reason from “first principles”—should lead him to question whether the notion of impartial altruism is as determinate as he assumes it is. In particular, there are scope-insensitive views that can lay equal claim to being impartial, altruistic, welfarist, and entirely consequence-focused. This leads to a troubling kind of moral uncertainty whose analysis and resolution I leave for another day.
1. Bottom-up versus top-down bracketing
Even if we cannot determine whether A or B produces greater total expected welfare, A may be determinately better with respect to some consequences, while the remaining consequences are too unclear to favor either option. Perhaps those consequences can be “bracketed,” leaving us with a reason for A and no reason for B.
Kollin, Clifton, DiGiovanni, and Macé (2025) distinguish two bracketing procedures. Bottom-up bracketing first identifies the beneficiaries for whom the comparison between A and B is determinate, sets aside the other beneficiaries, and then aggregates over those who remain. Top-down bracketing instead searches for the inclusion-maximal subsets of beneficiaries relative to which the comparison is determinate, and ranks an option above another only if all inclusion-maximal bracket sets weakly favor it and at least one strictly favors it. The authors describe bottom-up bracketing as prima facie compelling and show that it yields their intended verdict in the motivating Against Malaria Foundation vs. Make-A-Wish Foundation case: Against Malaria Foundation is more choiceworthy, even if we are clueless about either option’s long-run effects.
The authors nevertheless favor top-down bracketing because only top-down extends their baseline expected-total-welfare relation. This relation respects the principle of Statewise Dominance (which we will define momentarily). Bottom-up bracketing violates Statewise Dominance, and the authors reject it for this reason (pp. 15–20). Their discussion turns centrally on the following case:
e
¬e
A
(1, 10, 0)
(1, 0, 10)
B
(0, 0, 20)
(0, 20, 0)
The options are A and B, and the agent has maximally imprecise beliefs, i.e., their credal set permits every probability p ∈ [0,1] for the state e, where p = P(e). The state is assumed to be independent of the choice between A and B. The three entries in each vector represent the welfare of persons 1, 2, and 3, respectively.
For person 1, the comparison is simple: EU_1(A, p) = 1, EU_1(B, p) = 0. A is therefore determinately better for person 1 than B, regardless of which admissible probability is used. For person 2, A is better when p > 2/3, while B is better when p < 2/3. Similarly, for person 3, A is better when p < 1/3, while B is better when p > 1/3. Therefore, for persons 2 and 3, neither option is determinately better than the other, nor are they determinately equal.
The Super-Strong Pareto principle says that if A is determinately better for at least one person and B is determinately better for no one, then A is better overall (or more choiceworthy). That condition holds here: A is determinately better for person 1, while B is neither determinately better for person 2 nor for person 3. Bottom-up bracketing reaches the same verdict. It brackets the persons for whom the comparison is indeterminate, retains person 1, and chooses A.
This procedure violates Statewise Dominance, which says: If the aggregate value of B’s outcome is at least as great as that of A in every state, and greater in at least one state, then B is better overall (or more choiceworthy). In the above case, aggregate welfare ranks B above A under both e and ¬e: B yields total welfare 0 + 0 + 20 = 20, while A yields 1 + 10 + 0 = 11, in either state. B therefore statewise dominates A. It does not personwise pareto-dominate A, however, since person 1 is worse off under B in both states.
Bottom-up bracketing respects Super-Strong Pareto but violates Statewise Dominance. By contrast, top-down bracketing violates Super-Strong Pareto but respects Statewise Dominance. The top-down procedure bases its overall verdict on what is best for the inclusion-maximal subsets of individuals for which the agent is not clueless, i.e., for which the comparison is determinate. In the above case, the full set {person 1, person 2, person 3} is the unique inclusion-maximal set relative to which the comparison is determinate: B yields expected total welfare 20 under every admissible probability, while A yields 11. Top-down therefore chooses B.
Kollin et al. (2025) want their bracketing theory to be an extension of consequentialism (p. 18). Since consequentialism entails Statewise Dominance, they favor top-down over bottom-up bracketing. (Cf. Mogensen 2024 for a related view on consequentialist versus nonconsequentialist theories.)
2. A person-centered case for bottom-up bracketing
The bottom-up procedure arguably takes persons (or individuals) more seriously as moral patients. Consider the above case again. From person 1’s represented ex ante welfare standpoint, A is determinately better than B. Person 1 therefore would be “begging” us, as it were, to choose A (given our beliefs about how our decision will affect her), and would object to our choosing B on personal welfare grounds. For persons 2 and 3, by contrast, neither option is determinately better. They may have important conditional interests pointing in both directions, but neither has a contrastive reason to prefer A rather than B or B rather than A, nor would they object to either choice. (If the choice were theirs, they had only their individual welfare interests in view, and our credences, they could rationally choose either option.) If one person provides us with a determinate reason to prefer A over B, while no person provides us with a determinate reason to prefer B over A, then impartial altruists should arguably choose A. This is the thought behind Super-Strong Pareto.
If this is sound, Statewise Dominance is not a universal requirement of impartial altruistic choice. In cases like the above, it fails. And its failure is no mystery: It is perfectly intelligible based on person-centered contrastive reasons. (Cf. Clifton 2025 on reasons-based choice.) The top-down approach—which upholds Statewise Dominance—requires us to ignore these reasons and pool the people first, before making any comparisons. But individual persons are the fundamental welfare-bearers whose individual comparative perspectives determine whether things are better, worse, or neither. Our theory should therefore determine what each person’s represented prospect supports before pooling the individual stakes across persons.
What we have here, in effect, is a clash between different dominance principles: Personwise Dominance (Super-Strong Pareto) and Statewise Dominance. Kollin et al. (2025) side with Statewise Dominance by default (p. 18), but this is underargued. While it is plausible that consequence-focused, welfarist impartial altruism must respect suitable dominance principles, the choice between them depends on the basic object of altruistic concern, i.e., the basic source of altruistic reasons (see section 3.5 below). If individual persons are the basic object of altruistic concern, Personwise Dominance is the more plausible default.
Kagan (2015, p. 471ff., cf. Van Liedekerke 1995) discusses an infinite shift case that highlights the problem with pooling people before making any comparisons. (Recall that Statewise Dominance directs us to do just that.) Consider an infinite line of people at the following welfare levels:
..., -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, ...
The choice is between world W1 and world W2. In W1, exactly one person occupies each integer welfare level. In world W2, the very same people are each 100 units (or arbitrarily many units) better off. The anonymous set of welfare levels is unchanged. Both worlds contain exactly one person at every integer level. An evaluation sensitive only to that anonymous distribution is therefore hard-pressed to avoid the conclusion that neither world is better than the other, or more choiceworthy. (See Nebel 2026 and Muñoz 2026 for discussion of anonymity principles.) Yet every individual is much better off in W2. Each individual therefore provides us with a strong contrastive reason to prefer W2 over W1, and each is begging us to choose W2, as it were. Personwise Dominance principles (Pareto) rank W2 above W1, and Kagan says he “cannot take seriously” (p. 473) the contrary verdict. The lesson is not merely that infinite totals misbehave. Person-attributable comparisons preserve morally decisive information that anonymous distributions erase. Each person, given their comparative perspective on the choice, provides us with a strong contrastive reason to prefer W2 over W1.
Similarly, consider a choice between (1) saving persons p1, p2, p3, p4, ... and (2) saving persons p0, p1, p2, p3, p4, ... (i.e., two countably infinite sets of people, where the former is a proper subset of the latter), with all else being equal. If we pool the people first, we face pressure to conclude that neither option is better than the other: A straightforward total welfare comparison tells us that each option saves countably infinitely many lives. If, instead, we ask about each person’s contrastive welfare stakes first, we can distinguish the options immediately: Only p0 has a contrastive stake, and it favors (2). Hence, we choose (2).
Such cases do not by themselves establish bottom-up bracketing under cluelessness. They support the prior claim on which bottom-up bracketing rests: Individual identities and contrastive individual welfare stakes provide us with morally relevant information that anonymous welfare distributions can erase. Once that is granted, the natural procedure is to determine what each individual’s complete prospect supports before aggregating across persons.
3. Specific replies to and questions for DiGiovanni, Clifton, and others
3.1. An unexplained tension in DiGiovanni’s metanormative bracketing?
“... we might worry that it’s ad hoc to infer ‘the clueless view gives us no reason to choose’ from ‘the clueless view says the relative choiceworthiness of a and b is indeterminate’. Should we rather say the clueless view gives us an indeterminate reason to choose, and so we still can’t resolve infectious incomparability? I think the reasoning we just went through supports the ‘no reason’ perspective: Metanormativity is a different level of analysis than first-order normativity. If view n doesn’t consider a more or less choiceworthy than b, it seems natural to say that n doesn’t ‘have a complaint’ against choosing either a or b. And thus, n’s verdict has no weight in the choice between a and b.”
I must admit that this passage’s dialectical status was somewhat unclear to me. But if I’m not misinterpreting it, it may provide internal support for the person-centered rationale behind bottom-up bracketing defended above. DiGiovanni seems to argue that separate normative views are the appropriate units of metanormative deliberation. This suggests that we should ask whether and how each view compares the options, and base our overall verdict on the view-level verdicts. When a view regards A and B as incomparable, it has no objection to (or complaint against) our choosing either option, and we can therefore bracket it out.
This recommends the bottom-up procedure. First determine, for each separate normative view, whether and how it ranks A and B; then bracket out the individually incomparable views and aggregate over the others. This is structurally parallel to person-centered bottom-up bracketing: First determine, for each separate person, whether and how their complete represented prospect ranks A and B; then bracket out the individuals with incomparable verdicts and aggregate over the others.
DiGiovanni’s metanormative choice rule, however, is top-down rather than bottom-up. Under it, a metanormative bracket-set is any subset of normative views for which the options become comparable after aggregating the views in that subset. The rule then considers the inclusion-maximal such subsets. Consequently, an individually incomparable view need not be bracketed out: It may remain inside a maximal bracket-set if combining it with other views produces a determinate aggregate comparison. In this way, a normative view can be made to weigh in on a choice even though it is itself silent (indeterminate) on the choice. This seems inadequate.
Last but not least, the idea and language of “complaints” applies even more naturally at the first-order level, namely: to separate persons. If a person’s welfare determinately favors A over B, given our credences, she could object to/complain about our choosing B, on personal welfare grounds. Why not, therefore, hold that when an individual person has no contrastive stake in a choice, she provides us with no reason either way, and can thus be bracketed out? This is what bottom-up bracketing does.
Top-down bracketing, by contrast, can “force” an individual person to weigh in on a choice even though she is herself silent on (indeterminate about) the choice at hand, given our credences. (The person may remain inside a maximal bracket-set if combining her with other persons produces a determinate aggregate comparison.) If and to the extent that persons and their welfare-based individual perspectives are the basic objects of altruistic concern, this seems inadequate.
3.2. Does bracketing carve consequences arbitrarily?
“There are many different ways of carving up the set of ‘effects’ according to the reasoning above, which favor different strategies. For example: I might say that I’m confident that an AMF donation saves lives, and I’m clueless about its long-term effects overall. Yet I could just as well say I’m confident that there’s some nontrivially likely possible world containing an astronomical number of happy lives, which the donation makes less likely via potentially increasing x-risk, and I’m clueless about all the other effects overall. So, at least without an argument that some decomposition of the effects is normatively privileged over others, [we won’t have] much action guidance.”
Person-centered bottom-up bracketing avoids this objection. The decomposition is fixed independently of the desired verdict: The basic units are individual moral patients, and we assess each person’s complete represented welfare prospect under A and B—including all represented direct, indirect, near-term, and long-term effects—before asking whether either option is determinately better for them. We do not bracket “the long-run effects”; we bracket a person only when their complete represented prospect supplies no contrastive reason either way. An astronomical happy future therefore favors B only insofar as it makes some individuals’ prospects determinately favor B. (See section 3.6 for a comment on population ethics and nonidentity.)
Top-down bracketing also avoids DiGiovanni’s cherry-picking worry: Rather than selecting a convenient subset, it considers all inclusion-maximal determinate subsets. But top-down bracketing introduces new procedural choices that invite arbitrariness worries again: Why inclusion-maximality, rather than maximum-cardinality, say? And why unanimity among maximal sets? Why should a singleton and a set containing almost everyone have equal veto power?
3.3. Does bracketed cluelessness weaken “moral urgency”?
“Should I interpret [the bracketing verdict] as, ‘I ‘expect’ A to have systematically better effects than B, impartially speaking’? In particular, should I consider the impartial altruistic argument for choosing A over B to be as strong as if ‘all the other effects’ canceled out in expectation? Arguably not. I still care about those other effects, whether or not I can determine which strategy they recommend. I think that, properly disentangled, these effects give me some reasons in favor of A, some in favor of B, and it’s indeterminate how to weigh these reasons up. So I’m not sure I see the moral urgency [of the bracketing verdict].”
This reasoning threatens to prove too much by generalizing from cluelessness to the case of ordinary uncertainty. Suppose A directly saves a life, while its long-run effect is either an enormous benefit or an enormous harm to an astronomical number of people, each with a well-justified probability of exactly ½. Subjective expected-value reasoning says to choose A: The long-run effects cancel out in expectation, even though in actuality A may deterministically lead to the enormous harm and B to the enormous benefit. “Properly disentangled,” the effects would give us reason to favor B (though we don’t know this). So why, we might ask, consider the situation as morally urgent as if the effects canceled out (in actuality, not merely in expectation)?
But if that does not trouble us, it is unclear why indeterminacy should. While exact cancellation and bracketing differ epistemically—the former establishes an expected difference of zero, while the latter establishes no determinate direction—both leave us without a long-run contrastive reason for either option. Under both exact cancellation and bracketing, the moral importance of A derives from the determinate reason provided by the person saved.
3.4. Can a weighing model of reasons support Super-Strong Pareto?
The proposed person-centered rationale for bottom-up bracketing involves Super-Strong Pareto. Hedden (2024) argues (p. 589) that a weighing model of reasons fails to support Super-Strong Pareto (while it does support Weak and Strong Pareto). To the extent that such a model is independently plausible, Super-Strong Pareto may remain unjustified.
Kollin et al. (2025) have replied (p. 15) that this argument is question-begging, given that there is “no single ‘weighing model’ of reasons to fall back on.” I believe this reply to be on the right track but incomplete. (I also directionally agree with Kollin et al.’s replies to the various objections against Super-Strong Pareto from cyclicity, p. 13–16.) A weighing model of reasons is highly attractive, especially under a welfarist view of the content of our reasons. As argued above, the reasons provided to us by the welfare stakes of moral patients are plausibly contrastive: The better off an individual is in option A relative to option B, the stronger the reason to choose A rather than B (and the stronger the individual’s objection to our choosing B). The proposed model says to weigh up all such person-attributable contrastive reasons. This entails Super-Strong Pareto: If A is determinately better for at least one person, that person supplies a contrastive reason to choose A. If B is determinately better for no one, the respective people supply no contrastive reasons. Hence, weighing the contrastive reasons yields the verdict that A is better overall (or more choiceworthy).
On this person-centered model, unresolved considerations within a person’s prospect are not independently exported into an interpersonal aggregate. They must first resolve into a settled person-attributable comparison. If neither option is determinately better for a person, her prospect supplies no directional reason at the interpersonal stage. The justification, again, is that persons and their welfare-based perspectives on the choice at hand (as represented by the decision-maker), rather than subpersonal considerations or state-contingent fragments of personal prospects, are the basic objects of moral concern and the sources of welfare-based reasons.
3.5. Is person-centered bottom-up bracketing a “consequentialist” theory?
Kollin et al. (2025) reject Personwise Dominance (Super-Strong Pareto) and side with Statewise Dominance on the grounds that “consequentialism” requires Statewise Dominance (p. 18). As mentioned, while it is true that consequence-focused, welfarist impartial altruism must respect suitable dominance principles, the choice between them depends on the basic object of altruistic concern, i.e., the basic source of altruistic reasons. If individual persons are the basic objects of concern, Personwise Dominance (Super-Strong Pareto) is more plausible.
Nonconsequentialists have often complained that consequentialism, and utilitarianism in particular, fails to respect persons, their “separateness,” and their distinct comparative perspectives on the world by treating them merely as intrinsically unimportant “receptacles of value.” If I thought this objection were sound, it would trouble me enough that I would hardly be attracted to consequentialism. (But it is unsound, or at least avoidable—cf. Chappell 2015, who argues that “to avoid treating people as replaceable, utilitarians should be token-pluralists who see each person’s welfare as a distinct final good, rather than (separateness-violating) token-monists who see the aggregate welfare as the only final good.”) What attracts me to impartial altruism is the idea of equal concern for every single individual, and the aim of making their life go better rather than worse, by their own lights. Consequentialism, I would hope, is consistent with this. In any case, however, a theory that sides with Personwise Dominance over Statewise Dominance can be impartial, altruistic, and entirely focused on welfare consequences (for people). Whether this counts as “consequentialist” ultimately depends on how narrowly that label is defined.
3.6. Do population ethics or infinite ethics argue against a person-centered approach?
“It looks to me like bracketing with spacetime locations has a much better shot at being action-guiding than bracketing with persons. Despite the importance of this point, a proper discussion is beyond the scope of this post and our paper, though I’ve tried to put some intuition in a footnote.[7: Personal identity is fragile, such that many persons exist only in worlds where we take a particular action. For example, personal identity depends sensitively on the timing of conception, and through many chaotic pathways our actions affect the times at which future persons are conceived. (The point may be familiar from the nonidentity problem for person-affecting ethics.) This creates a problem: we can construct a maximal bracket-set based on ‘all the happy people who would exist if we do A’ and another based on ‘all the happy people who would exist if we do B,’ where these sets favor different actions. Since bracketing only favors A over B if A is better than B for all maximal bracket-sets, this would leave us still indecisive between A and B. ... <how spacetime locations as value-bearers may avoid such difficulties> ... ]”
I must admit I did not entirely follow the reasoning here. Clifton seems to suggest that top-down bracketing generates distinctive nonidentity-related problems if persons (instead of spacetime locations) are chosen as value-bearers. Suppose he is right that top-down bracketing must entail that A: create p1 at welfare level 10 and B: create p2 at welfare level 5 are equally choiceworthy. Perhaps, on a person-centered approach to ethics, that is simply true (cf. Boonin 2014 and Pummer 2024): If we choose B, we will have made it the case that (i) no actual person has negative welfare, which they could object to on personal welfare grounds, and (ii) all actual persons have the maximum amount of positive welfare they could have had. (Person p2, once she exists, will have no welfare-based objection to our choosing B: We could not have made her better off. Nor will p1, who will not exist.) On a person-centered approach, we may therefore have maximized people’s welfare in the sense that matters.
However, person-centered views have other theoretical options in population ethics, too. Let me illustrate this for person-centered bottom-up bracketing. In variable-population contexts, it must be based on some moral account of what person-indexed contrastive reasons, if any, option-dependent people supply across existence and nonexistence under uncertainty. Once such an account classifies each person’s prospect as favoring option A rather than B, B rather than A, tying, or having no direction, bottom-up bracketing proceeds just as in fixed-population cases: It retains persons whose comparison is determinate, brackets those whose comparison is indeterminate, and aggregates over the retained persons (i.e., weighs up the contrastive reasons supplied by them).
Clifton goes on:
“If it’s true that only spacetime location-based bracketing is action-guiding, that’s unfortunate, because spacetime locations don’t seem morally relevant! (You might or might not consider persons to be morally relevant -- maybe they, too, are just containers of experience-moments -- but spacetime locations definitely don’t seem to be.) We would be left with a tradeoff: Morally well-motivated but doesn’t get the intuitive answers in terms of action guidance (persons) vs. morally poorly-motivated but gets the intuitive answers (spacetime locations).”
But what intuitive answers, in terms of action guidance, does person-centered bottom-up bracketing fail to get? Also, even if it were true that person-centered views cannot avoid counterintuitive implications about action guidance, shouldn’t Clifton’s reasons-based approach side with what’s morally well-motivated (persons as value-bearers)?
Finally, Clifton writes:
“Interestingly, this is closely analogous to the situation in infinite ethics. Prominent theories of infinite ethics also depend on locations of value. (For example, we might want our theory to satisfy a ‘Pareto principle’ which says that an infinite world World 1 is better than another World 2 if it is better than World 2 at each location of value.) And it turns out that theories of infinite ethics based on persons also lead to much more widespread incomparability than those based on spacetime locations (Askell (2018), Wilkinson (2021)).[8] The implausibility of incomparability in the cases he considers leads Wilkinson to develop an approach to infinite ethics based on spacetime locations. We may have in infinite ethics a surprising source of support for spacetime bracketing?”
It seems to me that we do not—quite the contrary, as we have already seen in section 2. While Askell supports Clifton’s descriptive claim that person-based Pareto principles generate widespread incomparability, she rejects his intended inference (like many other authors in the infinite ethics literature). She defends person-based Pareto and accepts the resulting incomparability rather than adopting a more complete spacetime ordering without independent moral justification: “Ethics is concerned with the wellbeing of people. It is not particularly concerned with the pattern of utility across spacetime.” (Askell 2018, p. 84)
Furthermore, it is unclear how the mere fact of widespread incomparability renders the situation in infinite ethics “closely analogous.” This depends on the kind of incomparability at issue. Consider, for instance, the following comparisons involving countably infinite sets of people:
Proposition (i) is hard to deny; it seems to follow from impartiality (whatever impartiality is, precisely). (ii) follows from person-based Pareto. Now, add a transitivity premise:
(iii) If X = Y and Y < Z, then X < Z
The conjunction of (i), (ii), and (iii) entails the conclusion
But this is false. Instead of (i), we could have started with the proposition “saving persons a1, a2, a3, ... = saving persons b0, b1, b2, b3, ...”, on grounds of impartiality.
What gives? If, like many authors in the literature, we wish to preserve transitivity, the most plausible way out is
In other words, we introduce widespread incomparability (between any sufficiently disjoint infinite sets of people). But this sort of incomparability seems to me unrelated to problematic failures of action guidance.
Finally, in footnote 8, Clifton encourages the reader to consider Wilkinson’s (2021) Firing Line and Saving-a-Child-from-Traffic cases. As far as I can tell, person-centered bottom-up bracketing handles the latter without any noteworthy difficulties. The former is as follows: Infinitely many persons are lined up in front of us. We can either let every second person in the line be killed, or let every fifth person be killed. Wilkinson thinks the former option is worse, and uses this in an argument for spacetime locations over persons (as value-bearers). The option that kills every second person, he claims, kills more people in a sense that matters: It kills people that are more densely positioned over the line.
I disagree. Between the option of every second person dying and every fifth person dying, there is an overlap containing people who are sadly doomed anyway: {p10, p20, p30, p40, ...}. These people have no contrastive stake in the choice, so that they provide us with no contrastive reasons. Similarly, some people are safe either way and have no contrastive stake in the choice for this reason: {p1, p3, p7, p9, p11, ...}. We can affect the fates of {p2, p4, p6, p8, p12, ...} and {p5, p15, p25, p35, p45, ...}. Let us relabel the people in the former set as follows: {a1, a2, a3, a4, a5, ...}, and relabel the people in the latter set as follows: {b1, b2, b3, b4, b5, ...}. We have no net reason to prefer saving either set over the other.
4. Bonus [optional]: Is impartial altruism indeterminate?
In conclusion, let me draw out an implication suggested by the infinite cases just considered: that DiGiovanni’s epistemological approach—set intuitions and heuristics aside, reason from “first principles”—should lead him to question whether the notion of impartial altruism is as determinate as he assumes it is. In particular, I will briefly argue that some scope-insensitive views can lay equal claim to being impartial, altruistic, welfarist, and entirely consequence-focused. This may induce a particularly troubling kind of moral uncertainty concerning impartial altruism itself.
Consider, again, the above argument for widespread incomparability:
But (iv) is false. As discussed, one way out is to introduce incomparability and maintain that “saving a1, a2, a3, ... || saving b1, b2, b3, ...”. Another way out is to deny the transitivity premise (iii). Given the arguments Kollin et al. (2025) offer in defense of cyclicity (p. 13ff. and p. 22ff.), denying (iii) should be viable a fortiori; it commits us to a milder form of intransitivity, not to outright cyclicity.
The problem is that this may leave scope-sensitivity unjustified. Some people believe that impartial altruism should be scope-insensitive, i.e., that the numbers should not count (e.g., Taurek 1977 and Timmermann 2004). If you can save person a1 or persons b0 and b1 but not all three, either option is permissible, or you must perform an impartial lottery (regarding the latter, cf. the literature on probabilistic social choice rules, e.g., Intriligator 1973, Fishburn 1975, Pattanaik and Peleg 1986, and contemporary proponents such as Florian Brandl). The standard argument against such views, and for non-probabilistic scope-sensitivity, has run roughly as follows (e.g., Kavka 1979, in response to Taurek 1977):
(i) saving a1 = saving b1
(ii) saving b1 < saving b0 and b1
(iii) If X = Y and Y < Z, then X < Z
Therefore:
(iv) saving a1 < saving b0 and b1
This argument generalizes to any natural numbers m < n of people in the conclusion. Scope-insensitive views have two ways out. The first is to introduce incomparability: “saving a1 || saving b1.” The second is to deny the transitivity premise (iii). If these moves are legitimate in the infinite domain, why would they be illegitimate here? In other words, what is our reply to scope-insensitive altruists who make these moves—quite consistently—in the finite domain as well?
Such arguments create pressure to conclude that all we have is a mere intuition favoring (or disfavoring) scope-sensitivity—cf. Muñoz (2024). As Muñoz argues, scope-insensitive views can lay equal claim to “counting each person for one.” The scope-sensitive view that we are required to save b0 and b1 over a1 gives each person an equal additive vote. The scope-insensitive view that we may permissibly save b0 and b1 or a1 gives each person equal veto power to block a requirement to save the other side. The scope-insensitive view that we are required to perform an impartial lottery gives each person an equal probabilistic vote, i.e., equal probabilistic influence over the outcome. It is disputed whether, in the b0 and b1 vs. a1 case, this should lead to a (½, ½) lottery or to a (⅔, ⅓) lottery between the options, i.e., whether one should toss coins or draw lots. While lotteries violate non-probabilistic scope-sensitivity, they can be individually stakes-sensitive: If b0 and b1 each have less personal welfare at stake than a1, the lottery should plausibly be proportionally weighted in a1’s favor.
Hence, scope-insensitive views can be impartial, altruistic, welfarist, and entirely consequence-focused (i.e., they can deny the moral relevance of distinctions such as doing/allowing, intending/foreseeing, etc.). One might object that lottery procedures depart from an exclusive focus on welfare consequences. But consequentialist, scope-sensitive aggregation, or stakes-weighted additive voting, is a social choice procedure too (cf. Henning 2023). Nor does the justification for lotteries need to involve fairness. Lotteries can be justified as an impartial social choice rule within welfarist beneficence.
Though I cannot argue for this here, I share DiGiovanni’s epistemological aversion to relying on mere intuitions, best guesses, or heuristics. This leads me to ask: Why, precisely, must impartial altruists be scope-sensitive? The above arguments suggest that the notion of equal consideration may be quite indeterminate. This threatens to induce a troubling form of moral uncertainty that affects the heart of impartial altruism. I leave its conceptual analysis, implications for cluelessness, and resolution as an exercise for the reader.
31
More posts like this
326
If wild animal welfare is intractable, everything is intractable.
There's something that bothers me with bracketing, that I'd be interested to hear a reply to. It seems to affect bottom up bracketing as you have presented it here. I am unsure if it affects top down bracketing as well, but it sounds like it might.
Suppose I have an action available to me that will have the following 3 effects, impossible to disentangle:
My understanding is that bracketing would say not only is this act permissable, but that I should do it. This is because it benefits me, and I am clueless about its effects on every other beneficiary (the person who is injured by the glass in 100 years may well be compensated by a benefit from 3).
But this feels wrong! If I just consider the effects that I am not clueless about, the bad seems to clearly outweigh the good. The same intuition that motivated bracketing in the first place would seem to say that this act is wrong.
I can buy the plausibility of a theory that says it is indeterminate whether I should take this act or not, due to the uncertainty bundled up in 3. And I can buy a theory which says the act is wrong (since that is what my intuition strongly points towards). But a theory that says it is wrong not to take this act feels like it has gone wrong somewhere.
Apologies if this is naive and has already been addressed at length somewhere. I would be interested to know how proponents of bracketing respond to this.
Hi Toby––thanks, it's a good question. If you expected your broken glass to badly hurt a person tomorrow, I suppose you wouldn't feel clueless about the net effect on the person's lifetime wellbeing. If so, why feel clueless if the nasty injury occurs later? If injuries from broken glass in the ground are still happening in that future society, it can't have changed that much. (Otherwise, why assume that the broken glass would cause harm, rather than nothing or a benefit, in the first place?)
But suppose we are clueless about the net effect on the injured person's lifetime wellbeing. In that case, if we "ask" the person (from our ex ante perspective, given our imprecise credences) whether she prefers that we remove the glass rather than leaving it in the ground, her welfare-based answer is "no." Consequently, it would seem adequate to bracket her out and bracket you in (from a "consequentialist" point of view). The person supplies no contrastive reason favoring either option; you do.
I should also say, though, that not leaving the glass in the ground seems quite overdetermined by the reasons we have. E.g., deontological reasons not to actively cause serious foreseeable harm to a person (or violate their right to bodily integrity), and reasons not to risk doing so, provide a pretty clear verdict here. (Such reasons are relevant when it comes to actively causing the existence of future people, too.) You can't cut a person without their consent and then (over)compensate them for it.
Just to check: then we should also bracket me out in the snack case, right? I presumably don’t know that eating the tasty snack determinately improves my lifetime welfare overall. So your version of bracketing wouldn't actually recommend #1, as Toby assumed. It would remain silent in the decision situation he depicted.
To be action-guiding, person-based bracketing requires "knowing" the net effect on the lifetime well-being of at least one (lifetime) person, right?