Against Utilitarianism
Six fractures in Singer's moral arithmetic
A popular provocation in utilitarian circles claims that you ought to save an astronomical number of shrimp — \(10^{100}\) is the usual figure — over a single human life, and that disagreement is irrational or even evil. The argument presents itself as moral arithmetic at its most fearless: shrimp are sentient, suffering is suffering, multiply a tiny moral weight by a large enough number and it must eventually outweigh anything. Most people meet this with a flat no and are told their refusal is squeamishness — an inability to follow the math where it leads.
The refusal is correct, and it is not squeamishness. The shrimp argument is utilitarianism with the upholstery removed: aggregation, sentience-as-currency, and the tyranny of large numbers, operating exactly as designed. It fails not because the multiplication is done wrong but because the framework licensing the multiplication is wrong — and the ways it is wrong are worth cataloguing precisely, because they are the same fractures that run through the whole edifice of Peter Singer’s moral system, from famine relief to animal liberation to population ethics.
I gave Effective Altruism its charitable reading in Judging Goodness: a high-agency moral identity, admirable in its rigor and its follow-through, wrong only about its own universality. This chapter is the attack that reading deferred. Singer’s utilitarianism sits almost perfectly opposite to Axionic ethics. The two frameworks occasionally converge on the same practical recommendation, but that is superficial. Their foundations are different. Their units of moral concern are different. Their definitions of harm are different. Their tolerances for coercion are different. Their shape is different all the way down.
Singer begins with welfare — more precisely, with suffering, pleasure, and the impartial aggregation of consequences across all affected beings. The moral question becomes: what action produces the best overall experiential outcome, counted from the standpoint of everyone equally? That question drives the rest of the system. It drives the demand for impartiality, the tendency toward aggregation, the extreme demandingness, and the willingness to override local attachments, ownership, and personal boundaries whenever doing so improves the total.
Axionic ethics begins somewhere else entirely: with agency, coherence, evaluability, consent, and the rejection of coercion. Harm is not defined primarily as a reduction in aggregate welfare. Harm is damage to agency, distortion of evaluative integrity, violation of consent, manipulation of a being as if it were an instrument. The central moral fact is not that experience can be scored and summed. It is that agents exist as structured centers of valuation and action, and that any ethical system treating them as interchangeable containers for utility has already committed the basic philosophical error.
From these two starting points, six fractures open. Take them in order.
Aggregation
The first fracture is aggregation, and it is not a peripheral detail — it is the engine of utilitarianism. One person’s losses can be justified by another person’s gains if the overall result is sufficiently positive. Once aggregate welfare becomes the master variable, individuals cease to matter as sovereign centers of value and become locations where value can be produced, reduced, traded, or sacrificed. Singer would resist that formulation. He would say utilitarianism takes everyone seriously by counting everyone equally. The problem is that counting everyone equally in an aggregate calculus is precisely what makes everyone fungible. The structure is egalitarian at the level of arithmetic and annihilating at the level of personhood.
Singer’s shift from preference utilitarianism toward hedonic utilitarianism only sharpens the problem. Preference utilitarianism at least leaves some room for agency, authorship, and consent to matter as part of what is being optimized. Hedonic utilitarianism strips the moral currency down further: what finally matters is pleasure and suffering, and agency survives, if at all, as an instrument for producing better experiences. That is not a refinement. It is a cleaner statement of the reduction.
The shrimp argument shows what aggregation looks like at full extension, and in doing so it exposes the hidden assumption: that ethical values aggregate linearly and indefinitely, so that trivial moral weights multiplied by astronomical numbers must eventually surpass the moral significance of a person. But ethical significance is not strictly additive. There are thresholds beyond which piling up negligible moral weights never yields meaningful ethical claim — you do not get to swamp a sovereign agent by accumulating utility dust. A framework in which any value, however small, can be multiplied past any other value, however great, has not discovered a daring truth about shrimp. It has confessed that it contains no categorical distinctions at all — that everything in it trades against everything else at some exchange rate. Axionic ethics rejects the currency, not just the price. Agents are not buckets of welfare. They are not entries in a moral spreadsheet. They are not morally dissolvable into a global objective function. A moral framework that authorizes the sacrifice of one agent’s sovereignty for the sake of a sufficiently large total elsewhere has not discovered compassion. It has discovered an accounting trick.
Nor does resisting the trade require the strawman of infinite human value. Critics note that people accept small risks for trivial pleasures, and conclude that we already price human life finitely — so the shrimp multiplication must eventually win. But nothing in the refusal depends on infinity. Human standing is finite and qualitatively superior: grounded in agency and complexity, not in a bigger number on the same scale. Recognizing a difference in kind is not an inconsistency in a difference of degree.
Demandingness
The second fracture is demandingness. Singer is famous for pushing utilitarianism toward its most severe implications: if you can prevent suffering at relatively little cost to yourself, you are morally required to do so, and the requirement iterates until your marginal sacrifice equals the marginal benefit to distant strangers. This logic can be softened rhetorically, but it cannot be escaped without weakening the theory. Once aggregate welfare is the criterion, every retained luxury becomes morally suspect. Every local preference stands under permanent indictment. Every asymmetry between what you could give and what you do give becomes a potential moral failure.
This is not ethical seriousness. It is a category error masquerading as moral rigor. The existence of need elsewhere does not automatically generate a claim on your agency. It may generate an opportunity — for charity, solidarity, alliance, voluntary aid; those are real and often admirable. What it does not generate is a universal mortgage on the lives of other agents. Singer’s framework turns beneficence into standing obligation because it begins by assuming that value is globally aggregable and impersonally rankable. Axionic ethics denies both premises. The full dismantling of the Shallow Pond — why need does not create claim, and why moral pressure of this kind is coercion in emotional currency — is the work of Against Moral Extortion; here the structural point suffices: the demand only binds if the aggregation does, and the aggregation fails first.
Coercion as a Ledger Cost
The third fracture is where the conflict becomes politically explosive. Singer’s framework tends naturally toward redistribution: if resources can be extracted from some and reallocated to produce more welfare for others, utilitarianism will usually regard that as justified and often required. The state becomes a tool for outcome optimization, and the fact that coercion is involved is morally secondary — regrettable perhaps, costly perhaps, but instrumentally legitimate within the calculus. Coercion enters the ledger as one cost among others, to be netted against the gains.
Axionic ethics treats coercion very differently. Coercion — the credible threat of actual harm used to gain compliance — is not another input on the cost side. It is itself a morally salient kind of harm, because it directly attacks agency: it forces alignment through fear, dispossession, or domination rather than through consent, persuasion, exchange, or voluntary cooperation. Any ethic that treats coercion as a routine instrument of moral optimization has already subordinated agents to outcomes. That is not a policy disagreement. It is a foundational divergence about what ethical violation even is.
Utilitarians sometimes reply that strong protections for rights, property, consent, and family attachment can be defended instrumentally, because they tend to produce better outcomes in the long run. That reply does not dissolve the conflict; it restates it. Under utilitarianism, agency is still derivative — protected when useful, relaxed when inconvenient, overridden when the numbers grow large enough. Agency is not a helpful heuristic inside a welfare machine.
The same subordination of process to outcome explains why utilitarianism is structurally drawn toward manipulation and paternalism. A utilitarian can justify deception, censorship, nudging, or forced compliance whenever the expected consequences look favorable enough — this is the standing temptation of every consequentialist system: once outcomes govern, process becomes negotiable. Hedonic utilitarianism makes the danger clearer still. If what finally matters is the balance of felt experience, then managed belief, sedation, and coerced benevolence become easier to justify whenever they improve the aggregate hedonic result. Axionic ethics treats epistemic distortion itself as anti-agentic harm. To deceive someone for their own good is still to invade their evaluative process. To manipulate their beliefs for a socially beneficial end is still to treat them as a substrate to be managed. That is not compassion made rigorous. It is paternalism with better branding.
Impartiality
The fourth fracture is Singer’s demand for a very strong form of impartiality — the standpoint he calls the point of view of the universe. Your child, a stranger, and a distant population are all meant to count in fundamentally the same way. Particular attachments are morally permissible only under constraint; left to themselves, they look suspiciously like bias. The theory pushes toward a view in which personal commitments are always in danger of appearing ethically provincial.
Axionic ethics has no interest in this fantasy of view-from-nowhere moral cognition. Agents are situated. They have histories, commitments, identities, projects, relationships, and bounded spheres of authority. These are not contaminants to be scrubbed out by moral mathematics; they are part of the actual structure of agency. A father’s special concern for his child is not a local irrationality awaiting justification before an impartial tribunal. It is one of the ways agency manifests in the world. A framework that trains people to distrust every thick human attachment unless it can be defended in aggregate welfare terms is not elevating ethics. It is thinning out the moral world until only arithmetic remains.
Moral Realism by Stealth
Beneath all of this sits a metaphysical fracture. Singer writes as though suffering and flourishing generate reasons that are objective, impersonal, and universally binding. His utilitarianism is framed in secular terms, but structurally it inherits the full ambition of moral realism: the world contains morally relevant states; those states impose demands; ethical thought consists in properly recognizing and maximizing the good across all affected beings. Suffering becomes a universal currency of obligation — value located in the experience itself, commanding anyone who learns of it.
That is precisely the myth this volume opened by dismantling. Value is not a substance in the furniture of the universe. It does not hang over reality issuing commands. Note what is being rejected: unconditional realism — value floating free of every agent, binding everyone from nowhere. The grounded, conditional realism this volume ultimately endorses (objective conditionals over the structure of agency) is a different animal, and Singer’s system cannot retreat to it, because his currency — suffering, aggregated impartially — refuses exactly the agent-relative conditioning that makes the grounded version true. Values are agent-bound: they arise within evaluative structures, and they are interpreted, held, enacted, revised, and defended by agents under conditions. This does not make ethics arbitrary. It makes ethics conditional, structured, and real at the level where real valuation actually occurs — the same discipline that governs every other kind of claim, because all truth is conditional and moral truth is no exception. Singer’s system tries to leap over this layer by turning suffering into obligation directly. Axionic ethics regards that move as philosophical smuggling: it imports objectivity through the back door and calls the result compassion.
The same stealth realism powers the shrimp argument’s second load-bearing assumption: that sentience alone confers moral standing, so that any creature that can suffer enters the ledger on the same page as you. But the capacity to feel pain is not the capacity that grounds sovereign moral standing. Ethical significance derives largely from agency — the capacity to model futures, own policies, choose, and bear the consequences of choosing — not solely from sensory experience. Sentience without that structure is morally relevant without being morally sovereign; where the line falls, and why it is a threshold of capacity rather than a bias of species, is the argument of Sapientism. The shrimp argument needs sentience and sovereignty to be the same thing. They are not, and no multiplication closes a difference in kind.
To be clear about what the rejection of stealth realism does not claim: the Axionic prohibition on coercion does not treat agency as a mystical substance with objective cosmic sanctity — that would be realism through a different door. The point is structural. A system that normalizes domination, forced compliance, or epistemic capture destroys the conditions under which agents can remain evaluable as agents at all. The prohibition is constitutive of inter-agent order, not a borrowed relic of the realism I have just rejected.
Population Ethics
The sixth fracture is where utilitarianism’s structure produces its most famous embarrassment. Derek Parfit’s Repugnant Conclusion runs: a world with a billion people living excellent lives — deep relationships, rich experiences, long lifespans — is worse than a world with ten billion people whose lives are barely worth living, so long as total happiness is greater in the second world. The conclusion is repugnant precisely because it sacrifices quality for quantity, trading a smaller paradise for a larger mediocre swamp. And yet under classical utilitarianism it is not just acceptable — it is optimal. Enough tiny welfare increments outweigh severe harms; enough barely positive lives dominate fewer excellent ones; enough aggregate gain justifies almost anything if the arithmetic runs long enough. These are not accidental embarrassments. They reveal what happens when you reduce ethics to scalar optimization over a sufficiently large domain.
The problem is not in the math. It is in the premise. Population ethics requires you to believe there exists some global moral vantage point from which whole worlds can be evaluated and compared — that “World A is better than World B” can be true even though no agent lives across both, no one experiences both, and there is no mind whose preferences actually encompass both sets of lives. That assumption is false. All value is vantage-relative. There is no “good overall”; there is only “better from the standpoint of X.” The Repugnant Conclusion assumes a meta-agent capable of evaluating entire population states from nowhere. That agent does not exist. Nature is indifferent, God is silent, and reason operates on values without generating any.
Every major paradox in population ethics — totalism, averageism, person-affecting views, the asymmetries about non-existence — stems from this one error: treating morality as if it operates on global facts instead of agent-relative evaluations. Drop the assumption and the entire debate shifts. Whether to bring a new person into existence is a real question, but it only makes sense relative to a valuer — the would-be parent, the existing society, the person who would come to exist and value or regret their own life; it has enough structure of its own that it gets its own chapter. What you cannot ask is whether the total outcome is better in an absolute sense, because there is no total agent whose preferences you would be maximizing. In some contexts agents will prioritize quantity — spreading a culture, maximizing resilience. In others, quality — sustaining a high-trust, high-agency civilization. These are different vantage points, not contradictions. Population ethics, done honestly, stops being a search for the best world and becomes a branch of decision theory for agents with long-term, population-level preferences. That is not a weakness. It is the only coherent foundation. The Repugnant Conclusion is only a dilemma if you believe there is a single answer to “which world is better?” Once you see there isn’t, the dilemma evaporates. You can’t optimize for everyone, because there is no such thing as everyone. There are only agents, perspectives, and values.
And this is the fracture that finishes the shrimp argument too. You do not get to swamp violations of agency by piling up enough remote utility. You do not get to erase coercion by multiplying beneficiaries. You do not get to transmute domination into righteousness by enlarging the denominator. The shrimp-versus-human trade is the Repugnant Conclusion in a lab coat: clever arithmetic mistaken for moral clarity, running on a vantage point that no one occupies.
Two Different Questions
The contrast can now be stated cleanly. Singer asks: what action maximizes aggregate welfare across all affected beings, counted impartially? Axionic ethics asks: what preserves or violates agency, evaluability, consent, and non-coercive order among agents? These are not two versions of the same ethical project. They are rival conceptions of what ethics is about.
That is why the occasional overlap in policy tells us very little. Both frameworks condemn cruelty. Both support voluntary charity. Both can favor institutions that reduce misery. But the agreement is unstable because the reasons are different: Singer supports these things because they improve the total; Axionic ethics supports them when they preserve or enlarge agency without violating consent or legitimizing coercion. Once the background conditions change, the divergence reappears immediately.
Singer’s utilitarianism moralizes optimization. Axionic ethics moralizes sovereignty. Singer sees the ethical world as a field of sentient welfare to be improved from an impartial vantage; Axionic ethics sees it as a world of agents whose integrity cannot be collapsed into a single score. Singer asks what we owe the total. Axionic ethics asks what we may do to each other. Singer is willing to spend persons for outcomes — and treats that willingness as moral courage. Axionic ethics treats it as the central warning sign.
This is why Singer’s moral seriousness so often produces conclusions that feel monstrous to anyone who still takes agency seriously — a hundred duodecillion shrimp against your daughter’s life, and a charge of irrationality if you hesitate. The monstrosity is not incidental. It comes from the structure of the theory. Once morality becomes aggregate optimization, persons are always in danger of being outvoted by arithmetic.