When Does EA's Neutrality Reach Its Limits?

Abstract

Effective Altruism (EA) presents itself as politically neutral because it ranks causes and interventions by their expected contribution to impartial welfare rather than beginning from a party, ideology, or institutional programme. This paper argues that such neutrality does not eliminate politics; it relocates political choice into intervention selection, evidence standards, institutional design, expert delegation, ownership, and the allocation of authority. The relocation has real virtues: it disciplines symbolic politics, protects immediate aid, rewards measurable learning, and recognises the value of technical competence. It also reveals three commitments requiring political justification: methodological individualism, high epistemic thresholds for action-guiding evidence, and a primarily instrumental valuation of democracy. These generate, respectively, collective-action and institutional-feedback problems, evidential threshold conservatism, and a plutocracy risk when private wealth becomes political influence. The argument then extends to global catastrophic risk, advanced AI, extreme political scale, and transformative technology. Expert-dependent governance can remain democratic delegation when collective ends and coercive terms remain publicly authorised, contestable, and revisable; global capacity does not by itself select a world state; and technological abundance does not resolve questions of access, ownership, membership, or residual scarcity. EA therefore needs a political philosophy of accountable expertise, institutional legitimacy, and equal political standing.

Contents

• Introduction
• Part I: The Postponement of the Political Agenda
• Part II: First Hidden Commitment - Methodological Individualism
• Part III: Second Hidden Commitment - High Epistemic Thresholds and Evidential Threshold Conservatism
• Part IV: Instrumental Democracy and the Plutocracy Risk
• Part V: GCR Governance–Expertise, Engineering, and Political Authority
• Part VI: Global AI Governance at Extreme Scale
• Part VII: Transhumanism and the Political Economy of Abundance
• Conclusion: When Doing Good Scales into Governance
• Bibliography

Introduction

Effective Altruism is both a philosophy and a social movement organised around a demanding practical question: how can limited resources do the most good? MacAskill defines it as “(i) the use of evidence and careful reasoning to work out how to maximize the good with a given unit of resources, tentatively understanding the good in impartial welfarist terms, and (ii) the use of the findings… to try to improve the world” (MacAskill, 2019: 14). Impartial welfarism gives equal moral weight to everyone’s well-being, irrespective of nationality, distance, or temporal location, while maximisation directs altruistic resources towards the interventions expected to produce the greatest benefit at the margin (Singer, 2015a: vii; Zuolo, 2019: 3).

This orientation has an obvious moral attraction. It resists the assumption that good intentions, emotional proximity, partisan identification, or symbolic commitment are enough. Direct aid may save lives even when structural reform is unavailable; measurable evidence may reveal that an appealing intervention does not work; specialised knowledge may be indispensable to controlling catastrophic risk; technological innovation may expand health, freedom, and productive capacity. A political assessment of EA should preserve these achievements rather than treating effectiveness, evidence, expertise, or innovation as disguises for domination.

The difficulty is that EA’s procedural neutrality does not leave politics behind. Because it refuses to assign priority in advance to a political ideology or institutional form, it treats political action as one candidate intervention among others. Advocacy, public-capacity building, redistribution, institutional reform, and systemic transformation must enter the same comparative framework as direct health programmes, cash transfers, or technical research. Politics is therefore postponed until it can be translated into expected impact. This protects EA from romanticising political action, but it also changes politics’ status: questions about collective authority, public equality, and institutional membership become inputs into an evaluative procedure rather than the prior framework within which common decisions are authorised.

The paper argues that this relocation exposes three commitments at the centre of EA’s practical reasoning. First, methodological individualism asks what a donor, career-chooser, voter, or organisation should do at the margin. That standpoint can make duty concrete and protect urgent assistance, yet it can fail under assurance problems, thresholds, hidden causal dependence, and institutional feedback. Individually reasonable choices can then become collectively self-defeating. Second, high epistemic thresholds protect action from fantasy philanthropy and politically attractive but ineffective projects. Yet interventions do not enjoy equal access to randomised, local, and measurable evidence. When political power helps determine which experiments can be conducted and which effects become legible, evidential discipline can become evidential threshold conservatism. Third, EA can give democracy a strong consequential defense, while remaining formally open to its constitutive value. But when democracy is valued primarily through welfare-producing effects, public authorship and equal political standing are harder to represent. The problem becomes acute when wealthy private actors use philanthropic resources to shape laws and institutions, creating a plutocracy risk.

These three arguments become more, not less, important as EA turns towards global catastrophic risk and transformative technology. Serious GCR governance requires forecasting, causal modelling, safety engineering, monitoring, and technically competent administration. The relevant distinction is not between expertise and democracy, but among epistemic expertise, administrative delegation, and political authority. Experts may determine feasible means and exercise bounded discretion without acquiring the right to set collective ends, acceptable risks, rights constraints, distributive burdens, or coercive terms. The constructive aim is accountable expertise rather than expert sovereignty.

Global scale then raises a constitutional question. Transboundary risk may require globally capable institutions, but global coordination, global governance, global government, a world state, and AI-assisted administration are not interchangeable. Functional necessity does not by itself establish historical inevitability or normative legitimacy. Nor does the inadequacy of present institutions for a hypothetical society of one trillion persons prove democracy impossible. It instead intensifies questions of representation, agenda control, accountability, membership, and the design of institutional correction. The same logic extends to transhumanism and abundance. EA is not identical to transhumanism, but the projects overlap where interventions alter human capacities and future populations. Automation, employment-independent income, enhancement, monetary design, land and other residual scarcities, and scarce digital assets all return the analysis to ownership, access, transition risk, and equal standing.

The argument proceeds in seven parts. Part I explains the postponement of the political agenda. Parts II–IV analyse methodological individualism, evidential threshold conservatism, and instrumental democracy with its plutocracy risk. Part V develops a model of accountable expert delegation for GCR governance. Part VI compares globally capable institutional forms and stress-tests democracy under extreme scale and AI-assisted administration. Part VII examines the overlap between EA and transhumanism through the political economy of automation, abundance, ownership, and residual scarcity.

Part I: The Postponement of the Political Agenda

EA appears apolitical because it begins with cause-neutrality rather than party, nation, class, or ideology. Its guiding question is not which political camp should win, but where limited resources can do the most good. Cause-neutrality leaves open which problems deserve priority; means-neutrality leaves open any permissible instrument, including direct service, entrepreneurship, or lobbying, if it produces the greatest expected good (MacAskill, 2019: 21). This neutrality is procedural rather than axiological: EA does not suspend judgment about the good, but refuses to assign priority in advance to a party, class, nation, institutional form, or familiar mode of assistance. Politics is therefore not an axiomatic starting point but a downstream variable. MacAskill describes EA as “being open in principle to focusing on any problem… and being open in principle to using any (non-side-constraint violating) means to addressing that problem” (MacAskill, 2019: 21). Rather than asking how to achieve “socialist liberation” or “libertarian freedom,” it asks “how can we… succeed in having the greatest positive impact we can?” (MacAskill, 2015: 5).

This posture has a serious defence. It requires agents to specify an objective and compare interventions rather than allowing partisan identity, symbolic commitment, or emotional proximity to substitute for impact. Kucz and Rosol describe EA as avoiding “constant quarrels over values, virtues, and obligations” (Kucz and Rosol, 2020: 61) while seeking “an ethical-philanthropic framework… attentive to consequences but that also acknowledges and incorporates a range of hard-to-measure substantive and procedural values” (Kucz and Rosol, 2020: 76). Rubenstein similarly writes that EA shifts attention from the contrast between altruism and self-interest to the contrast between “doing the most good and everything else” (Rubenstein, 2016: 516), urging donors to focus on effects rather than good intentions, warm-glow feelings, or the intrinsic value of actions (Rubenstein, 2016: 515). Kucz and Rosol therefore characterise effectiveness as a common ground for ethical evaluation (Kucz and Rosol, 2020: 61). The discipline is morally important: a donation does not become effective merely because its object is familiar, politically expressive, or emotionally vivid.

Expected-value analysis operationalizes this discipline by combining the value of possible outcomes with their probability. Once politics enters this framework, advocacy loses any automatic priority and competes with other interventions on expected impact. EA’s three-factor heuristic asks how much welfare is at stake in a problem’s scale, how few resources its neglectedness leaves committed to it, and how much additional progress its tractability permits per unit of resources (MacAskill, 2015). A legislative campaign, an institutional reform, or a revolutionary project must therefore be compared with direct interventions rather than presumed superior because it is more political.

This comparison also explains EA’s attraction to what is called mistake theory. Mistake theory treats many political disagreements as disputes about information, models, or means to broadly positive-sum ends; conflict theory instead emphasises incompatible interests and relations of domination (Brennan, 2025). EA’s dominant grammar can recognise conflict, but it more readily asks how a problem can be solved than who benefits from the structure producing it. Brennan captures the associated strategy through the idea of “pull the rope sideways”: rather than joining a polarised struggle, agents seek neglected interventions that parties are less likely to oppose (Brennan, 2025). Alexander gives this posture a collective-action defence. Controversial political commitments can lead effective altruists to “canceling ourselves out” when some fund the left and others the right, while “no one actively fights to preserve poverty or malaria” (Alexander, 2013). Partisan alignment may also alienate participants and reduce the movement’s common capacity. In this register, politics appears less as a struggle over power than as a set of empirical disagreements about what works (MacAskill, 2013).

The postponement of the political agenda follows from this structure. Political action is not rejected; it is required to justify itself as one intervention among others. The move protects EA from romanticizing politics or substituting political identity for moral impact. Yet it changes politics’ status. Politics becomes a candidate technology for producing welfare rather than the privileged arena in which citizens authorize the basic terms of their common life.

Two assumptions support this move. The first is consequentialist commensurability: diverse goods can be compared on a common scale of value. In EA practice, that scale is commonly aggregate welfare, represented through measures such as Quality-Adjusted Life Years, Disability-Adjusted Life Years, Well-Being-Adjusted Life Years, or cost-effectiveness ratios (Gold, Stevenson, and Fryback, 2002: 115). The second is institutional instrumentalism: social, legal, and political institutions are valued insofar as they causally produce better outcomes (Singer, 2002: 122). Under these assumptions, political struggle does not receive independent priority as an arena of rights, justice, or collective empowerment. A lobby, policy campaign, or anti-capitalist revolution must compete in expected value with vertical health interventions such as bed nets or deworming tablets (MacAskill, 2015: 35). This is both the virtue and the danger of EA’s posture: it refuses to romanticize political action, but it initially grants politics no special authority.

The contrast between a public health ministry and a private bed-net program makes the differential legibility clear. An EA-aligned foundation might choose between strengthening a ministry’s long-term administrative capacity and funding an NGO able to distribute anti-malarial bed nets immediately. The ministry may build public capacity, local accountability, and citizens’ claims on the state, but its effects are diffuse and uncertain. The NGO offers a clearer sequence: money in, nets out, estimated disease burden reduced. Under a strict impact framework, the second option is easier to count. The political option has not been excluded; it has been postponed until it can speak the language of intervention effectiveness.

Singer’s reply establishes the strongest objection to interpreting this postponement as hostility to systemic reform: “if large-scale reform offers some prospect of reducing poverty… then effective altruists will try to assess its chance of doing good” (Singer, 2015c). EA is therefore not conceptually committed to apolitical charity. It can, in principle, support lobbying, policy reform, institutional redesign, or ambitious systemic change. The issue is not exclusion but translation: political projects must enter the same comparative language of evidence, expected impact, and effectiveness as other interventions.

This explains how a movement without a substantive political platform can still generate a distinctive political orientation. Its neutrality depends on assumptions about who acts, what evidence should guide action, and why political institutions matter. These assumptions often appear methodological rather than political, but they shape which agents count as decision-makers, which interventions appear comparable, and which political values enter the calculus directly rather than through their welfare effects. The next three parts examine them as methodological individualism, high epistemic thresholds for action-guiding evidence, and the instrumental valuation of democracy.

Part II: Methodological Individualism and the Problem of Collective Agency

The first hidden commitment is methodological individualism. The institutional critique of EA provides the starting point. In its influential formulation, the critique holds that EA has neglected institutional change that might address the root causes of poverty while focusing on individuals directing resources to organisations that directly aid people living in poverty (Berkey, 2018: 143; Gabriel, 2017: 457-473). The target is not simply ineffective charity. It is the broader pattern by which global poverty is framed as a problem to be alleviated through direct, measurable, privately chosen interventions rather than through changes in the institutions that structure vulnerability.

Direct-aid cases make this background picture of agency especially visible. Deworming programmes, insecticide-treated bed nets, cash transfers, and other optimised interventions are attractive because their effects are measurable and attributable. Rubenstein notes that EA urges donors to focus on effects rather than intentions, warm-glow feelings, or the intrinsic value of actions (Rubenstein, 2016: 515-516). Critics worry, however, that this discipline narrows the moral field. If global poverty is sustained by trade rules, debt regimes, weak state capacity, extractive political economies, or failures of democratic accountability, routing money through efficient NGOs may treat symptoms while bypassing causes (Srinivasan, 2015: 5-6; Berkey, 2018: 143-144).

The deeper target is therefore EA’s background picture of agency. EA usually asks what a single resource-possessing agent can do on the margin, rather than how historical and political structures produced the crisis. Srinivasan argues that EA’s utilitarian calculations tend to presuppose that “everyone else will continue to conduct business as usual,” and that “the individual, not the community, class or state,” is the proper object of moral theorizing (Srinivasan, 2015: 5). Berkey summarizes a similar worry in Herzog and Mills: effective altruists often picture individuals as isolated agents rather than as interconnected participants in a social world capable of changing institutions (Berkey, 2018: 11, 144). In this sense, methodological individualism functions as EA’s first hidden commitment.

This starting point has a serious defence. Against Andrew Kuper’s charge that urging the bourgeoisie to become charitable “as individual actors” cannot produce deep structural change, Singer replies that where no credible path to such change is available, direct rescue still matters: “When we can’t make deep structural changes, it is still better to help some people than to help none” (Singer, 2002: 126). His Schindler example makes the same point. Schindler did not alter “the structure of the Nazi genocide,” but he saved particular people and was right to do so (Singer, 2002: 126). Singer later argues that a minimally acceptable ethical life requires using a substantial part of one’s spare resources to make the world better (Singer, 2015a, 2015b). The addressee of moral demand is the donor, consumer, voter, career-chooser, or activist with resources.

Methodological individualism therefore should not be treated as a simple moral failure. It makes duty psychologically concrete and refuses to let affluent agents hide behind structural abstractions. People in need cannot wait until the theory of the state is repaired. EA defenders can also object that individuals cannot be blamed for failing to control states or restructure capitalism. McMahan compares such blame to condemning a doctor who treats war victims for failing instead to eliminate the root causes of war (McMahan, 2016: 97, quoted in Berkey, 2018: 161). This reply protects immediate aid from a perfectionism that would sacrifice present victims to an abstract revolutionary horizon.

Yet the defence does not remove the collective problem. Collins argues that group contexts create cases in which individually and marginally benefiting others may need to take a back seat in moral reasoning (Collins, 2019: 211). MacAskill and Pummer likewise note that if EA is construed as a wholly individualistic project, everyone may individually do their best while collectively doing less good than they could have done together (MacAskill and Pummer, 2020: 5-6). Kissel, extending Kuper’s critique, draws the political conclusion: it is not enough to say that all persons have equal moral claims on us; we must also ask how to organise ourselves politically and economically to meet those claims (Kissel, 2017: 13).

The challenge is not that effective altruists pursue private interests. By hypothesis, they care about impartial welfare. The difficulty is that impartial welfare can still be pursued from an individualised decision standpoint: each agent asks what she should do, given what others are likely to do, to produce the greatest expected marginal increase in aggregate welfare. The payoffs below should therefore not be read as private benefits. They represent expected aggregate welfare, action-guiding estimates of aggregate welfare, or welfare contributions under a strategic condition. The problem is individualised impartial maximisation under strategic interdependence. Three families of mechanisms show why this standpoint can fail.

Coordination and threshold failure

The first family includes assurance failure, epistemic divergence, failures of common knowledge, and non-additivity. A Stag Hunt captures the assurance problem. Mutual cooperation yields the best collective outcome, but unilateral cooperation is risky. Each agent would prefer systemic change if others also commit to it; without sufficient confidence that they will, the safer individually rational choice may be marginal aid. Agents trying to benefit others as much as possible may therefore have reason to signal willingness to coordinate rather than simply maximise their own marginal contribution (Collins, 2019: 210–215).

Table 1. Assurance failure in systemic change

If each donor expects others to fund marginal aid, systemic change looks too risky. Each may choose marginal aid not because she values reform less in principle, but because she assigns low credence to uncoordinated success. The result is a stable but suboptimal equilibrium.

Epistemic divergence complicates coordination even when agents share the same impartial objective. One group may believe that deregulation, growth, technological acceleration, or scientific innovation will maximise expected welfare; another may favour regulation, labour protection, redistribution, or democratic control. This is not a conflict between private payoffs but between empirical models. If opposed groups fund incompatible political campaigns, their resources may cancel out even though both act from impartial motives.

A related failure arises from the absence of common knowledge. Agents may agree about values and even about the best collective plan while lacking shared information about who will act. The thin-ice rescue case illustrates the structure: one rescuer may save the child, while several uncoordinated rescuers may cause the ice to collapse. The EA analogue occurs when funders independently rush into the same opportunity because each thinks that otherwise no one will. Over-contribution can exceed absorptive capacity, duplicate effort, reduce the marginal value of later contributions, and leave other opportunities neglected. The failure is not low altruism but ignorance of the total pattern of action.

Non-additivity adds a threshold structure. Many political and institutional goods are not produced by smooth marginal increments. Labor organizing, anti-monopoly restructuring, sovereign debt reform, climate regulation, democratic state-building, and social norm change may appear ineffective before a threshold is crossed; after it is crossed, earlier contributions become part of the explanation of success. If each donor asks only what her marginal dollar is expected to achieve under current conditions, she may prefer the smoother return of direct aid. Methodological individualism can therefore mistake the flat part of a threshold curve for evidence of inefficacy.

Hidden causal dependence

The second family concerns displacement, duplication, background funding, and the hidden-zero problem. Here the difficulty is not failed assurance among equal agents but a causal parameter omitted from the individual donor’s calculation. Budolfson and Spears identify one version: even if a top-rated charity is genuinely effective, an ordinary donor’s contribution may have no marginal effect when a billionaire funder is already prepared to top up the charity to its operating target. The ordinary donor then reduces the billionaire’s contribution rather than increasing the charity’s budget.

The same structure appears in duplication and background funding. A donor may reasonably believe that she is adding resources to a neglected opportunity while lacking information about other funders’ commitments, total funding, or absorptive capacity. A contribution that appears additive from the individual standpoint may be displaced, duplicated, or rendered marginally worthless by actions outside the donor’s view. This is not a Stag Hunt: the problem is hidden causal dependence. It challenges methodological individualism because the value of an individual contribution cannot be established without locating that contribution within the wider funding system.

Institutional feedback

The third family concerns state substitution and the long-run effects of aid on public capacity. A direct intervention may produce genuine short-term benefits while weakening the political and institutional conditions required for long-term welfare. Clough and Muyskens argue that well-funded private NGOs can provide parallel health, education, or welfare services that reduce pressure on governments to build durable public systems, redirect accountability from domestic publics to external donors, and demobilize citizens who exit inferior public services (Clough, 2015; Muyskens, 2017: 4).

The objection does not deny the immediate gains. In the institutional-feedback case, the donor may accurately observe lives saved per dollar while missing effects that emerge at the system level: weaker state capacity, lower democratic pressure, diminished public accountability, and deeper dependence on external provision. Marginal aid can therefore be locally beneficial yet systemically harmful over time. This is why the defence of immediate rescue and the institutional critique can both be true. Individuals remain right to help when structural transformation is unavailable, but a movement organised around doing the most good also needs to ask whether its interventions alter the conditions under which future help will be needed.

Together, the three families expose the limits of methodological individualism without denying the moral importance of individual action. Individual subjective rationality, even when sincerely directed at impartial welfare, is not sufficient for collective effectiveness. A movement that aims to do the most good needs not only better individual expected-value calculations but also a theory of coordination, common knowledge, causal interdependence, institutional feedback, and collective agency.

The resulting danger is collective self-defeat. A theory is collectively self-defeating when widespread acceptance of its decision procedure causes agents, as a group, to achieve the theory’s own aims less well than they would under another set of motives or rules (Venkatesh, 2023: 56; Parfit, 1984: 28). As Gibbard, Regan, Dietz, and others emphasise, even when each person individually produces the best outcome available given what others are doing, they may together produce an outcome worse than one they could have achieved collectively (Dietz, 2019: 109, quoted in Koh, 2024: 47; Freiman, 2025: 213). That is the structural danger of individually rational rabbit hunting.

Part III: High Epistemic Thresholds and Evidential Threshold Conservatism

The second hidden commitment is a high epistemic threshold for action-guiding evidence. Faced with thousands of possible charities, causes, careers, and political projects, EA uses empirical research to restrict the option space. Cote and Steuwer therefore note that its restrictiveness partly follows from one of its virtues: a simple, evidence-based procedure for directing altruistic resources (Cote and Steuwer, 2023: 153). EA asks not only which intervention has the largest imaginable upside, but whether the estimate is supported by evidence strong enough to guide action.

Cote and Steuwer describe a high evidentiary threshold: recommending or funding a charity without solid evidence of effectiveness is morally irresponsible; such evidence should control for selection bias, and RCTs often receive privileged status as the gold standard for estimating causal efficacy and average impact size (Cote and Steuwer, 2023: 154-155). Interventions may therefore be rejected not because their expected value has been shown to be low, but because it has not been established in the accepted evidential form.

This threshold has a strong defence: it protects EA from fantasy philanthropy, marketing-driven charity, and self-indulgent politics. MacAskill advises donors to examine what a charity does, its cost-effectiveness, the robustness of its evidence, its trials and self-evaluation, and its capacity to use additional funds (MacAskill, 2015: 108-109). Strong evidence of fair cost-effectiveness may be preferable to weak evidence of extreme cost-effectiveness, since weak estimates are often optimistic (MacAskill, 2015: 112-113). Scared Straight illustrates the safeguard: before-and-after data suggested success, while randomised trials suggested harm (MacAskill, 2015: 113).

Askell distinguishes the balance of evidence - how decisively it supports a proposition - from its weight - how much relevant evidence we possess (Askell, 2019: 39-40). The same credence may rest on extensive evidence or near-total ignorance: 0.5 might reflect strong evidence that a coin is fair or uncertainty about its bias. Askell captures the difference through estimate resilience: an estimate is more resilient when new evidence would change it less (Askell, 2019: 40).

On an evidence-favoring view, the value of an investment depends both on standard expected value and on the weight of evidence supporting it. If two interventions have equal expected value, the better-evidenced one is preferable, all else equal (Askell, 2019: 43–44). On an evidence-neutral view, total value is simply standard expected value, irrespective of evidential weight (Askell, 2019: 43–44). Askell defends a sophisticated evidence-neutral position: when two interventions have comparable expected value, investment in the less-evidenced option may be justified by the information it generates, especially where the estimate is high-variance and low-resilience (Askell, 2019: 48–49).

EA’s official epistemology therefore contains internal resources for exploration. Always exploiting the best-evidenced intervention may prevent discovery, while indiscriminate exploration wastes resources. Low evidential weight is not automatically a reason for rejection when expected value and information value are sufficiently high (Askell, 2019: 48–49). In near-term evaluation, however, Cote and Steuwer note that EA charity evaluation often requires RCT-based support before recommendation and asks whether micro-level effects generalise when programmes scale (Cote and Steuwer, 2023: 155).

The political difficulty is that interventions do not have equal access to the evidential forms privileged in near-term evaluation. Deworming, bed-net distribution, cash transfers, and corporate cage-free campaigns have identifiable treatment units, measurable outcomes, and plausible comparison groups. Political and institutional interventions often do not. One cannot straightforwardly run an RCT on union power, land reform, sovereign debt cancellation, anti-monopoly restructuring, democratic state capacity, constitutional change, or an entrenched norm. Their counterfactuals may be historically unavailable, their effects delayed and causally overdetermined, and the intervention may transform its evaluative background.

Bisson’s discussion of Clough makes the asymmetry concrete. An RCT may show whether bed nets lowered malaria among recipients, but it is far less likely to show whether the same program demobilized political pressure on the government to build a broader malaria-eradication system (Bisson, 2021: 50). That effect operates through public expectations, political organization, and state capacity, and is poorly captured by a local comparison.

Nonlinearity deepens the problem. Cote and Steuwer argue that myopic marginalism often assumes that benefits vary continuously with scale: small funding increases produce small expected benefits (Cote and Steuwer, 2023: 152-153). That assumption works reasonably well for bed nets but not for threshold-dependent reforms. Social conformity traps such as footbinding, female genital mutilation, and child labour can persist while individuals have reason to conform; once a sufficiently large minority defects, the equilibrium can collapse. Footbinding lasted roughly a thousand years but ended within a generation once a sufficiently large minority abandoned it (Cote and Steuwer, 2023: 157). Before the threshold, contributions may appear ineffective; after it, they help explain transformation.

I call the resulting selection effect evidential threshold conservatism. The term does not mean that effective altruists are temperamentally conservative or hostile to systemic change. It identifies a procedural tendency: transformative, non-linear interventions can be deprioritized because they cannot provide high-certainty empirical proof before being attempted at scale.

A serious objection is that EA officially remains open to systemic change: it does not say, “Never change institutions.” MacAskill says EA is “clearly open to systemic change in both principle and practice” (MacAskill, 2019: 25); Berkey argues that institutional reform is supported if it does the most good (Berkey, 2018: 143; Berkey, 2021: 101); and Broi urges critics to examine assumptions about diminishing marginal returns and low-hanging fruit rather than claim that EA’s core theory excludes reform (Broi, 2019). This blocks a crude institutional critique, but it does not remove the evidential asymmetry. The demand to show that institutional change is among the best uses of resources becomes politically selective when interventions differ systematically in their ability to produce the accepted forms of proof.

Gabriel and McElwee describe the resulting “missing middle”: in global poverty, EA favours low-value/high-confidence interventions such as vertical health programmes and cash transfers while hesitating about advocacy, citizen empowerment, and rights-based initiatives; yet in existential risk, EA has sometimes accepted high-value/low-confidence work (Gabriel and McElwee, 2019: 99-101). Advocacy sharpens the inconsistency because there is no accessible counterfactual world in which the same campaign did not occur (Gabriel and McElwee, 2019: 104).

Addressing this middle requires more than lowering standards. Syme argues that systemic change needs historical, interpretive, and holistic analysis because social systems are complex forms of intentional activity (Syme, 2019: 100-102). Lechterman similarly argues that methods used to identify promising assistance programmes do not transfer smoothly to political reform (Lechterman, 2020: 88). Advocacy works through persuasion, coalition-building, agenda-setting, litigation, lobbying, electoral mobilisation, and reallocations of power; its effects are delayed and causally overdetermined (Lechterman, 2020: 112-113). Comparative, historical, interpretive, and structural evidence answer different causal and institutional questions.

The political economy of evidence production deepens the asymmetry. Evidence is produced by institutions with budgets, permissions, incentives, and access. Existing institutions readily evaluate health programmes, cash transfers, bed nets, or small curricular changes. Trials of land redistribution, unionisation, debt restructuring, anti-monopoly enforcement, or anti-corruption campaigns are often obstructed because they threaten powerful interests. The absence of gold-standard evidence may therefore be endogenous to the status quo.

This requires distinguishing ordinary from structurally produced evidence gaps. In ordinary cases, weak evidence is a reason for caution: an intervention may have been tested and failed or lack a plausible mechanism. In structurally obstructed cases, the social order whose transformation is at issue also limits evidence production. Evidential threshold conservatism arises when both are treated alike: “We lack strong evidence that S works” becomes “We should deprioritise S in favour of D.” Where an intervention cannot become testable without first building the political power it seeks to create, absence of evidence is not evidence of absence; it reflects an epistemic environment shaped by power rather than evidence of ineffectiveness.

This is Syme’s system affirmation bias. Narrow empiricism risks affirming existing systems because it privileges forms of knowledge those systems are willing and able to generate (Syme, 2019: 95-96). The bias is procedural rather than necessarily psychological: interventions that challenge existing conditions must prove themselves before receiving the means to become testable. As Srinivasan argues, EA’s calculations often presuppose that others continue “business as usual,” leaving moral agents to make piecemeal interventions within the existing order (Srinivasan, 2015: 5-6).

The answer is not epistemic permissiveness. EA is right to reject the idea that political passion establishes effectiveness. Revolutionary, state-building, and anti-capitalist projects can fail, be captured, produce authoritarian outcomes, or leave the worst-off worse off. Syme emphasizes that systemic change requires controversial judgments about interpretation, strategy, positionality, and ethics, not romantic confidence that radical action must be good (Syme, 2019: 102-108). Sebo and Paul likewise warn that transformative interventions can involve radical uncertainty and may change the evaluator’s preferences and standpoint (Sebo and Paul, 2019: 67).

The constructive alternative is active evidence creation through democratic experimentalism. Evidence consumption selects the currently best-supported intervention; evidence creation invests in inquiry, experimentation, and institution-building that make opaque interventions more assessable. If political barriers help explain the evidence gap, waiting for evidence reproduces the barrier. EA could instead support localised, transparent, democratically accountable “experiments in living” under public scrutiny (Mill, quoted in Sebo and Paul, 2019: 61): participatory budgeting, cooperative ownership, community health governance, labour-organising pilots, public-capacity partnerships, municipal anti-poverty reforms, or democratic accountability mechanisms.

Democratic experimentalism would combine quantitative evidence with qualitative interpretation, track public capacity and institutional learning, model thresholds and coalition effects, and allow affected communities to help define success and decide whether pilots should scale. Evidence is institutionally produced, politically enabled, and normatively interpreted. EA is right that systemic change requires evidence; its critics are right that the existing evidential order is biased against some forms of systemic change. The task is to create a politically legitimate evidential practice capable of testing structural transformation without romanticizing revolution or freezing the status quo.

Part IV: Instrumental Democracy and the Plutocracy Risk

The third hidden commitment concerns democracy. EA is not anti-democratic. From an impartial-welfarist perspective, democracy has a powerful instrumental defence: democratic institutions can improve information flows, align governments more closely with citizens’ interests, expose policy failure, constrain abuse, permit peaceful correction, and protect liberties that contribute to well-being. Economists have argued that democracy causes growth (Acemoglu, Naidu, Restrepo, and Robinson, 2019); Sen famously argued that democratic institutions and a free press help prevent famines (Sen, 1982; Rubin, 2009); and empirical studies suggest a positive association between democratic participation and life satisfaction (Orviska, Caplanova, and Hudson, 2014: 493). If democracy reliably protects welfare-relevant goods, EA has strong reason to support it.

Singer develops this consequential defence explicitly. It was “no accident,” he argues, that Bentham and his utilitarian followers supported democratic reform, because democracy helps align governmental interests with those of the governed (Singer, 2015c). He applies the same structure to justice, freedom, equality, and knowledge. Most effective altruists, he suggests, value these goods because they help create communities in which people live better lives, suffer less oppression, possess greater self-respect, enjoy more freedom to pursue their aims, and face less suffering and premature death (Singer, 2015a). Singer also stresses that there is no EA “party line” on whether democracy has intrinsic rather than merely instrumental value; some effective altruists likely think it does, while others do not (Singer, 2015c). This pluralism matters. It blocks the claim that EA is necessarily committed to treating democracy, justice, or equality as mere instruments.

The political question concerns the form and limits of this defence. Democracy can be valued instrumentally because it produces good outcomes, or constitutively because public equality, participation, and collective self-rule are themselves components of a legitimate political relationship. The distinction can be clarified through weak and strong republicanism.

By weak republicanism, I mean a view that values democratic participation, civic virtue, and accountable institutions primarily because they secure non-domination, contestability, the rule of law, and protection against arbitrary power. Participation need not be intrinsically valuable on this view. It is a means by which citizens prevent rulers from exercising uncontrolled power. This corresponds to the “instrumental turn” in contemporary republicanism associated with Skinner, Sunstein, and Pettit: participation and civic virtue are justified as instruments for securing political liberty understood as independence from arbitrary rule (Lovett, 2022: 9).

EA can accommodate weak republicanism readily. If democratic institutions improve information, constrain elites, expose mistakes, reduce domination, and force rulers to register the interests of people they might otherwise ignore, they become valuable political technologies for producing welfare-relevant goods. Singer’s Benthamite defence has this structure. Koh’s utilitarian reconstruction does as well: if every person’s welfare matters, institutions should be responsive to each person’s interests, and democracy may be among the best mechanisms for ensuring that responsiveness (Koh, 2024: 125). On this account, democracy receives robust support without requiring the claim that participation is valuable independently of its consequences.

By strong republicanism, I mean a view on which participation, public authorship, or civic self-rule is itself part of citizens’ freedom and equality, rather than merely a means of securing non-domination or good administration. Citizens are not merely welfare-bearers whose interests should be competently served; they are co-authors of the laws and institutions under which they live. Active political participation and public life can therefore be constitutive components of human flourishing (Lovett, 2022: 8–9).

Strong republicanism identifies a loss that cannot be captured solely by measuring outcomes. A benevolent expert regime, enlightened foundation, or efficient administrative apparatus may deliver excellent results while still failing to treat citizens as political equals. The objection is not merely that experts can make mistakes because they lack dispersed knowledge. Nor does it require citizens themselves to perform every technical analysis involved in public health, AI governance, macroeconomics, or biosecurity. The deeper issue concerns authority. Citizens may reasonably rely on experts while retaining the status of those in whose name institutions act, to whom decisions must be justified, and by whom governing authority remains contestable. When citizens are reduced to recipients of optimised policy, something is lost even where welfare improves: they are governed for, but not governed with.

EA’s formal axiological openness provides a serious reply. MacAskill defines EA as the project of using evidence and careful reasoning to determine how to maximise the good, only “tentatively” understanding the good in impartial-welfarist terms (MacAskill, 2019: 14). EA therefore need not be committed to hedonism, preference utilitarianism, or any single theory of value. In principle, its evaluative framework could incorporate justice, freedom, equality, non-domination, democratic legitimacy, civic authorship, or public equality as independently valuable goods.

The difficulty lies in the relationship between formal openness and operational practice. EA’s comparative method works most readily when value can be represented through welfare, suffering reduction, lives saved, healthy life-years, subjective well-being, preference satisfaction, or quality of life. QALYs, DALYs, WELLBYs, cost-effectiveness ratios, and related measures make diverse interventions portable, aggregable, and rankable. This gives the movement considerable practical power: it can compare interventions rather than relying on moral enthusiasm alone.

The same machinery, however, creates an operational tendency toward quality-of-life welfarism. This is not necessarily EA’s foundational doctrine, and it need not amount to crude hedonism. The point is that standard metrics privilege goods that can be translated into welfare, health, preference satisfaction, or life quality. They can represent the value of preventing malaria, extending healthy life, transferring cash, or reducing suffering more easily than the value of public authorship, equal political standing, or participation in collective self-government. Democratic authorship is not naturally a QALY; political equality is not simply an increment in reported life satisfaction, even where it contributes to well-being.

The resulting problem is one of axiological clarification rather than decisive refutation. If democratic participation, public equality, and co-authorship matter only because they improve welfare, EA can endorse weak republicanism while declining the stronger claim. If they are independently valuable, EA must explain how they enter practical comparison without being translated entirely into welfare effects. Singer’s pluralism leaves both possibilities open, but the evaluative practices examined here tend to operationalize the instrumental account more readily than the constitutive one. Values resistant to welfare translation need not be denied to be disadvantaged; they may instead be bracketed, softened, or treated as constraints whose force must itself be justified through consequences.

Longtermist political philosophy intensifies the tension. Schmidt and Barrett distinguish instrumental defences of democracy, which appeal to its consequences, from intrinsic defences grounded in freedom, equality, fairness, and collective self-rule (Schmidt and Barrett, 2025: 477–478). Under conditions of extremely high moral stakes, democratic theories may face pressure toward what Halstead calls “high-stakes instrumentalism”: if enough expected value is at risk, democratic delay, disagreement, or procedural inclusion may appear morally costly (Schmidt and Barrett, 2025: 478; Halstead, 2017). Longtermist concerns about existential risk, AI governance, value lock-in, totalitarianism, and future welfare make that pressure especially acute.

Yet longtermism also supplies strong reasons to preserve democratic institutions. Schmidt and Barrett emphasise that institutions have enormous scale, persistence, and effects across policy domains (Schmidt and Barrett, 2025: 466–468). Democracies may reduce totalitarian risk, resist value lock-in, improve epistemic feedback, expose concentrated power, and make catastrophic political abuse less likely. Brennan further argues that effective altruists tend to be distinctively liberal and often regard liberal democracy as rare, precious, and historically fragile (Brennan, 2025: n.p.). Longtermism can therefore weaken democracy’s position when procedure is treated as overrideable by sufficiently large stakes, but strengthen it when democratic institutions are understood as safeguards against irreversible governance failure. The relevant question is not whether technical expertise should be used. It is whether high stakes convert expert judgement into an entitlement to displace democratic authorisation.

The education-foundation thought experiment isolates the issue. Suppose a foundation discovers a remarkably cost-effective way to improve education in a poor region. It funds curricula, teacher incentives, assessments, and digital platforms. Test scores rise, dropout rates fall, and measured welfare improves. Yet suppose the foundation has effectively redesigned a public education system without meaningful local authorisation. Parents, teachers, and citizens were treated as data sources or implementation partners rather than co-authors of the institution governing their children. The defect is not exhausted by the possibility of bad outcomes. It concerns political standing: who had authority to define the problem, choose the metrics, design the solution, and determine which trade-offs were acceptable?

Saunders-Hastings’s critique of welfare paternalism explains why the problem persists even where the intervention helps. She defines paternalism as intervention that restricts, manipulates, or circumvents an agent’s choices because the paternalist judges that agent less able to choose well (Saunders-Hastings, 2019: 116–117). Her objection does not deny the importance of welfare improvement, especially for people at very low welfare levels (Saunders-Hastings, 2019: 115). It concerns the political and relational form through which benefits are delivered. Welfare-promoting philanthropy can remain objectionably hierarchical when donors substitute their judgement for that of beneficiaries.

Nor must welfarism be abandoned to recognise this problem. Saunders-Hastings argues that even theories that do not regard egalitarian relations as intrinsically valuable have reason to value them instrumentally (Saunders-Hastings, 2019: 116). Hierarchical philanthropic relationships can damage material welfare, psychological welfare, self-respect, and local agency (Saunders-Hastings, 2019: 128). Donors also possess cognitive limitations and biases; the history of philanthropy gives reason to doubt that superior data or resources generally establish superior judgement about what affected people should receive (Saunders-Hastings, 2019: 128–129). The limiting principle is therefore not that every paternalistic intervention is impermissible. It is a presumption in favour of minimising paternalism among comparably effective options and choosing forms of assistance that register and respect beneficiary choice (Saunders-Hastings, 2019: 129–130).

The democratic stakes become sharper when philanthropy turns into political advocacy. A wealthy donor funding malaria prevention raises questions about accountability, beneficiary choice, and institutional effects, but is not necessarily attempting to alter laws binding an entire population. A donor financing lobbying, litigation, think tanks, ballot initiatives, regulatory design, or electoral advocacy is trying to shape the terms under which others must live. The decision can no longer be evaluated only as a private allocation of charitable resources.

This is the plutocracy risk. It is not a claim that wealthy effective altruists are corrupt, selfish, or insincere. It is a structural concern about unequal political power. Democratic citizens are supposed to stand as political equals; their judgements should not count for more merely because they possess wealth, status, celebrity, or privileged access. Lechterman develops the point through political subordination. Under conditions of radical inequality, using superior wealth to overcome political opponents relies on an arbitrary source of strength and treats those opponents as obstacles rather than as equal persons to be reasoned with (Lechterman, 2020: 31). The concern is stronger in transnational advocacy, where affluent outsiders may shape institutions in societies whose residents possess fewer resources to resist and must bear the consequences of reform (Lechterman, 2020: 31–32).

This produces a genuine dilemma for calls to make EA “more political.” Institutional critics urge effective altruists to move beyond bed nets and cash transfers toward laws, institutions, and global structures. But political engagement financed through concentrated private wealth can expand precisely the kind of unequal power those critics oppose. EA may escape technocratic quietism only by entering politics through donor power. The question is therefore not simply whether advocacy has high expected impact, but whether its means of influence preserve citizens’ equal political standing.

The conclusion is not that wealthy donors must never support advocacy, that affected communities possess all relevant expertise, or that democratic institutions should reject technical competence. Political giving instead requires a theory of legitimacy alongside a theory of impact. At minimum, legitimate elite-funded advocacy requires transparency, public justification, meaningful participation by affected communities, contestability, limits on domination, and institutional checks preventing private benevolence from becoming private rule. The next part develops the relationship between expert competence and political authority in greater detail.

EA’s third hidden commitment is therefore not hostility to democracy, but a conditional and often instrumental valuation of it. That commitment has a serious defence: democratic institutions protect welfare-relevant goods, and consequential arguments for democracy are powerful. But once EA actors seek to shape public institutions, they must explain why citizens remain authors rather than merely beneficiaries of optimised policy. The political question is not only “What works?” It is also “Who may decide what counts as working, for whom, and by what authority?”

Part V: GCR Governance–Expertise, Engineering, and Political Authority

Serious governance of global catastrophic risks is necessarily expert-dependent and engineering-intensive. Advanced AI, biosecurity, nuclear systems, and technical monitoring cannot be governed responsibly through unstructured mass opinion. EA’s emphasis on evidence, forecasting, causal modelling, and safety engineering is therefore a genuine governing resource. Shulman and Thornley illustrate the relevant tasks: AI alignment, interpretability, safety engineering, and testing; biosafety training and risk assessment; microbial forensics and pathogen monitoring; and the verification and regulatory capacity needed to detect breaches and enforce safeguards (Shulman and Thornley, 2025: 276–278). Democratic government cannot dispense with these capacities without becoming less able to protect those subject to it.

The political problem begins by distinguishing three functions. Epistemic expertise estimates probabilities, causal mechanisms, system behaviour, and technical feasibility. Engineering and administrative delegation designs controls, monitors compliance, specifies feasible means, and exercises bounded discretion. Political authority determines collective ends, acceptable risk, rights constraints, the distribution of burdens, and the terms of coercion. Technical dependence establishes the functional necessity of the first two functions, but not the historical inevitability of any particular regime or the normative legitimacy of expert rule. The boundary between means and ends is porous: where administrative discretion substantially redefines authorised aims or redistributes risks and burdens, renewed political authorisation is required.

Estlund’s expert/boss fallacy identifies the invalid inference. Some people may know more about better and worse political decisions, but superior knowledge does not by itself create a right to rule, a duty in others to obey, or permission to coerce (Estlund, 2008: 3–5, 40–44). Legitimacy concerns the permissibility of coercive enforcement, whereas authority concerns whether a command can itself create a moral requirement to comply (Estlund, 2008: 41–42). Expertise alone establishes neither. Reich’s question is therefore well posed: would EA, if designing institutions from scratch, prefer good-maximising technocrats to democratic government (Reich, 2015)? The answer cannot be supplied by showing only that technocrats are more competent at some tasks.

Christiano supplies a positive alternative. Democracy publicly realises equality where people disagree in good faith about justice and the common good, yet complex states necessarily require a division of labour (Christiano, 2008: 75–83, 257–259). Democratic institutions can authorise broad ends, acceptable risks, rights constraints, distributive principles, and coercive terms, while specialised agencies determine technically feasible means and exercise the discretion that implementation requires. Such administrative authority is instrumental to democratically authorised aims and remains defensible when constrained by political institutions and monitored by civil society. When agencies depart from or substantially redefine those aims, complexity no longer justifies delegation; the change requires renewed authorisation.

Citizens need not reproduce expert reasoning to judge whether this authority remains legitimate. Moore distinguishes reflective acceptance from blind belief: non-experts can assess source reliability, institutional remit, the quality of explanations, the authority relationship, and the public consequences of expert action (Moore, 2017: 59–66). Continuing authority also depends on real opportunities for scrutiny and contestation. Associations, social movements, and counter-experts can articulate neglected problems and identities, expose implicit framings and interests, and organise oversight that isolated citizens could not supply (Moore, 2017: 95–104). Urgent action may temporarily require citizens and officials to act on expert conclusions they cannot independently verify; it does not justify eliminating the institutions through which those conclusions are later examined.

Plural composition and organised dissent also have conditional epistemic value. Landemore argues that, under specified conditions, differences in perspectives, interpretations, heuristics, and predictive models can enable cognitively diverse groups to outperform groups selected only for individual ability; representation can reproduce some of this diversity at a deliberatively manageable scale (Landemore, 2013: 97–110). This is not the claim that demographic diversity automatically produces cognitive diversity or that democracy always outperforms expertise. Deliberation can generate informational cascades, suppressed dissent, polarisation, and groupthink. Red teaming, devil’s advocacy, renewal, and institutionalised dissent are partial safeguards rather than guarantees (Landemore, 2013: 118–125).

Technocracy should therefore be defined not as the use of experts but as a claim about political representation. In Caramani’s ideal type, technocracy treats the public interest as unitary and objectively discoverable, construes politics as identifying correct solutions rather than mediating plural interests, bypasses intermediary institutions, and weakens vertical and horizontal accountability (Caramani, 2017: 54–55, 59–65). Ferguson’s anti-politics mechanism names the same danger at the level of problem framing: questions of power and authority are translated into technical-administrative problems (Ferguson, 1990: 256), while optimisation can displace inquiry into the structures producing the problem (Rubenstein, 2016: 517). EA’s apparent neutrality then relocates political choice into expert definitions of objectives, evidence, feasibility, and acceptable trade-offs.

The institutional test is positive. Expertise remains democratic delegation when it operates under a public mandate; distinguishes technical assessment from decisions about ends, rights, risks, and burdens; offers explanations sufficient for public judgement; permits organised contestation and counter-expertise; incorporates plural perspectives and dissent; and remains monitored and revisable. It becomes technocratic sovereignty when experts treat competence as self-authorisation, redefine their mandate without renewed approval, present contested distributions as technical necessities, or use urgency and secrecy to escape continuing accountability. High stakes may justify advance delegation, bounded discretion, and altered timing of scrutiny, but not permanent expert sovereignty. Part VI therefore asks which global AI-governance structures can preserve this relationship between indispensable capacity and equal political standing.

Part VI: Global AI Governance at Extreme Scale

The strongest case for a global AI government is conditional but serious. If national fragmentation permits regulatory arbitrage, weakest-link failure, strategic races, or unenforceable safeguards, a globally centralised authority may gain decisive coordination and enforcement advantages. It could establish common rules, organise monitoring and verification, and impose consequences that fragmented arrangements cannot sustain. Shulman and Thornley show that catastrophe prevention already uses treaties, technical monitoring, verification, specialised safety research, and regulatory capacity (Shulman and Thornley, 2025: 276–278). Their argument supports the need for globally capable institutions, but it does not show that these functions require a unitary government.

Global coordination aligns the conduct of multiple actors, while global governance creates cross-border rules and institutions. Global government adds standing political authority with general governing powers; a world state possesses ultimate global sovereignty and a monopoly over legitimate coercion; AI-assisted administration is instead a capacity that any of these arrangements might use. These distinctions preserve three separate questions: what is functionally necessary for risk control, what institutional order is historically likely to emerge, and what form of authority is normatively legitimate.

Four ideal types distribute the relevant advantages and dangers differently. A unitary global government offers the clearest route to uniform rules and central enforcement, but concentrates coercive and epistemic power, creates the most demanding problem of democratic authorisation, and makes correction difficult if defective values become locked in. A global federation divides sovereignty between global and lower levels, combining common authority over specified threats with more proximate representation, but jurisdictional conflict and blurred responsibility may weaken enforcement and accountability. Polycentric governance preserves overlapping centres, experimentation, and multiple points of contestation, while risking inconsistent standards and negotiated rather than uniform compliance. A function-specific GCR authority concentrates expertise and enforcement within a narrow mandate, limiting the conversion of technical necessity into general sovereignty, but may either lack sufficient power or expand beyond its original remit.

Structural pressure does not make a world state inevitable. Shannon reconstructs Wendt as arguing that anarchy, increasingly destructive technologies, and struggles for recognition drive political development toward a world state (Shannon, 2005: 581–582). Shannon’s objection is that every transition still depends on agency and political choice. Major powers may possess the capacity to construct a global state while retaining strong incentives to preserve sovereignty, autonomy, and influence. A world state can therefore be possible without being historically necessary; its formation would require persuasion, power, institutional construction, and contingent support (Shannon, 2005: 582–587).

Dahl identifies the democratic cost borne by every globally empowered arrangement. Size and complexity make delegation unavoidable, but delegation can become so extensive that citizens cease effectively to control important final decisions (Dahl, 1999: 19–23). Enlarging a political unit may increase its capacity to address problems beyond the reach of smaller governments while reducing ordinary citizens’ influence over the larger authority. This does not imply that international institutions should be rejected. They may provide governing capacity, rights protection, or other benefits that outweigh democratic losses, but those losses must be stated, scrutinised, reduced where possible, and assessed under explicit standards rather than concealed by democratic labels (Dahl, 1999: 23, 32–34).

AI-assisted administration may alter this capacity problem, but it is not a constitution. AI systems could operate within any of the four models. Ngo and Bales frame AI alignment as the technical problem of making advanced systems act in accordance with human values and safety constraints (Ngo and Bales, 2025: 411). The further questions–whose values, selected through what procedure, and authorised by whom–are political inferences. Better information processing cannot itself determine legitimate ends, acceptable risk, rights constraints, or burden distribution.

A society of one trillion persons would make current territorial parliaments, direct mass deliberation, and existing agenda processes inadequate. That establishes the failure of present institutional forms, not a fixed numerical limit on democracy. Democratic capacity must instead be disaggregated into information aggregation, agenda control, representation, accountability, and membership: whether distributed knowledge can reach decision-makers; who decides which issues and alternatives receive attention; whether manageable representative bodies remain responsive to the wider population; whether those exercising authority can be identified, challenged, and replaced; and who is entitled to equal political standing.

Landemore offers a conditional response to the representation problem. Representative institutions can preserve some of a wider population’s cognitive diversity at a scale where deliberation remains feasible, especially when they are renewed and responsive to distributed perspectives (Landemore, 2013: 104–110). This does not mean that cognitive diversity is identical to demographic diversity or that representation guarantees epistemic success. Deliberation can produce informational cascades, hidden information, polarisation, and groupthink; accountability, institutionalised dissent, red teaming, and devil’s advocacy are partial remedies rather than assurances (Landemore, 2013: 118–123).

BCI introduces a further conditional possibility. If it substantially reduced translation and communication costs, it would weaken one practical obstacle to global deliberation and might expand democratic bandwidth. It would not thereby settle agenda and attention, access and equality, manipulation and surveillance, or ownership and political authorisation. Greater bandwidth can support democratic institutions without supplying democratic authority.

Globally capable governance may therefore be functionally necessary while its constitutional form remains a political choice. Each model must be judged by its coordination and enforcement capacity, its fit with the accountable-expertise criteria developed in Part V, its democratic costs, and its vulnerability to concentration and lock-in. Institutions have effects whose scale, breadth, and persistence make these design choices themselves longtermist priorities (Schmidt and Barrett, 2025: 466–470). Part VII turns from the organisation of global authority to who is governed, who owns transformative infrastructures, and how technological abundance and residual scarcity are distributed.

Part VII: Transhumanism and the Political Economy of Abundance

Part VI ended by asking who is governed, who owns transformative infrastructures, and how abundance and residual scarcity are distributed. The reviewer is right that these questions bring EA into contact with transhumanism as a political-philosophy problem. EA and transhumanism are not identical: one is a practical project of using evidence and reasoning to do the most good; the other concerns technologies that alter human capacities and forms of life. They overlap when EA evaluates interventions that extend healthspan, enhance cognition, automate production, or reshape future populations. Expected welfare then cannot settle who authorises transformation, controls its infrastructure, receives its benefits, or bears its risks.

Democratic transhumanism offers a constructive alternative to both comprehensive prohibition and unregulated adoption. Transformative technologies may enlarge bodily and cognitive liberty, but their political meaning depends on whether development and access are democratically governed or controlled by concentrated private power. Hughes connects freedom to use enhancement technologies with safety regulation, equality, and broad access, while recognising that the same technologies can support liberty or deepen oppression and exploitation (Hughes, 2004: 191–197). Fukuyama’s more cautious position converges on the institutional point that technologies combining benefits with risks require political institutions capable of regulating their development and use (Fukuyama, 2002: 181–182). The choice is among political arrangements for innovation, not between innovation and politics.

Healthspan extension makes the distributive issue concrete without settling it for every enhancement. Kuruc and Manley acknowledge that anti-aging treatments would probably reach wealthier users first. They argue that unequal initial uptake is not necessarily a reason to prevent a beneficial technology; it creates a problem of widening access, and some non-rival health and knowledge benefits may diffuse differently from rival goods (Kuruc and Manley, 2025: 440–441). This reply is strongest when paired with an institutional question: who owns the technology, who finances diffusion, and what prevents temporary inequality from hardening into a durable hierarchy of capacities?

Automation raises the same question at the level of income. If it persistently weakens wage labour as the principal mechanism through which people receive a share of production, employment-independent income may become central to economic security and equal citizenship. UBI is one possible response, alongside social dividends, public or shared ownership of productive infrastructure, universal basic services, and other transfers. The choice turns not only on aggregate welfare, but on whether people remain dependent on private owners or gain a durable claim to the productive order.

The reviewer’s monetary-financing proposal is a serious hypothesis. Money-financed UBI need not produce persistent general inflation if real output expands sufficiently relative to additional nominal demand. But the mechanism depends on whether supply arrives when and where demand rises, whether unused productive capacity exists, whether bottlenecks constrain expansion, and how velocity, expectations, fiscal and monetary credibility, and asset prices respond. Productivity growth makes non-inflationary financing conceivable; it does not establish it automatically. Consumer-price stability could also coexist with inflation in assets or scarce factors.

Technological abundance is similarly domain-specific. Lower production costs for reproducible goods do not eliminate rival land and valuable locations, congestible energy and compute capacity, ecological limits, or positional goods such as attention, status, and political influence. Some scarcities are also created or intensified through property rules, access restrictions, planning, and infrastructure control. Abundance may therefore shift conflict from manufactured goods toward bottlenecks, rents, and the institutions allocating access. This does not imply that land or any other scarce factor must follow a particular price path.

A credibly scarce digital asset supplies another possible mechanism. By attracting portfolio demand, it could absorb nominal purchasing power that might otherwise flow into consumer goods or physical assets. Yet liquidity absorption is not consumer-price stability. Demand may instead produce digital-asset inflation, volatility, unequal windfalls, ownership concentration, or increased private monetary power; later spending by asset holders could return purchasing power to other markets. Bitcoin or another scarce digital asset is therefore a serious institutional hypothesis, not an established inflation hedge or self-executing monetary solution.

The common thread is ownership and authority. The same transformation can widen liberty when access is broad and control contestable, or deepen hierarchy when enhancement, data, productive capital, compute, land, and monetary assets are concentrated. Abundance changes what politics governs rather than ending politics. EA’s apparent neutrality relocates political decisions into income institutions, infrastructure ownership, transition risks, residual scarcity, and political membership. A political philosophy adequate to technological transformation must ask not only whether an intervention increases expected welfare, but who authorises it, who can access it, who bears its failures, and whether its institutional form preserves equal political standing.

Conclusion: When Doing Good Scales into Governance

EA’s political significance lies less in an explicit programme than in the structure of its practical reasoning. By asking which intervention produces the greatest expected good, it refuses to grant automatic priority to familiar causes, ideological identities, or political symbolism. That discipline is worth preserving. Direct aid can remain obligatory when structural transformation is unavailable; strong evidence can prevent harmful intervention; experts and engineers can supply capacities without which catastrophic risks cannot be governed; technological innovation can enlarge health, liberty, and material possibility. The paper’s argument is not that these goods conceal politics, but that their use at scale must be politically authorised.

The three hidden commitments explain why. Methodological individualism makes responsibility actionable, yet individual marginal optimisation can fail when outcomes depend on coordination, thresholds, background funding, or institutional feedback. High epistemic thresholds protect against wishful thinking, yet evidence is produced within institutions whose powers and incentives make some interventions easier to test than others. An instrumental defence of democracy can be powerful, yet it does not by itself explain why citizens should remain co-authors rather than beneficiaries of well-designed policy. When private wealth supplies the means of institutional reform, the demand for “more politics” can therefore reproduce unequal political power.

The later analysis deepens rather than displaces these conclusions. GCR governance makes specialised knowledge and engineering indispensable, but indispensability concerns functions before it concerns authority. Experts can estimate risks, design controls, and administer complex rules while remaining delegates under publicly authorised ends, rights, burdens, and coercive terms. Global risk likewise creates a case for globally capable institutions without selecting a unitary world government by conceptual necessity. Centralised, federal, polycentric, and function-specific arrangements distribute coordination, enforcement, democratic control, concentration risk, and reversibility differently. Extreme population scale and AI-assisted administration alter the technological conditions of democracy, but they do not answer who sets agendas, who remains accountable, or who belongs to the political community.

Transformative technologies expose the same political structure in economic form. Enhancement may widen liberty or entrench hierarchy depending on access and control. If automation persistently weakens wages as the main distributive mechanism, employment-independent income may become central, but UBI remains one institutional option among several. Productivity may under specified conditions offset some inflationary pressure from monetary financing, and scarce digital assets may absorb some portfolio demand, but neither mechanism settles distribution, asset inflation, or monetary power. Abundance in reproducible goods also leaves rival land and location, congestible infrastructure, ecological capacity, status, attention, and political influence contested. The future does not remove scarcity or politics; it changes their objects.

EA therefore needs more than an invitation to support institutional reform. It needs an account of collective agency, evidence production, democratic value, expert delegation, global institutional design, ownership, access, and membership. Such an account would not replace impact assessment. It would identify the political conditions under which impact can be pursued without converting epistemic advantage, technical capacity, or private wealth into an unaccountable right to rule.

When doing good scales into governance, those subject to transformative institutions must retain equal political standing through public authorisation, accountable delegation, contestable ownership, and the continuing power to revise the terms of their common life.

Bibliography

Acemoglu, Daron, Suresh Naidu, Pascual Restrepo, and James A. Robinson. 2019. ‘Democracy Does Cause Growth.’ Journal of Political Economy 127 (1): 47-100.

Alexander, Scott. 2013. ‘Efficient Charity: Do Unto Others…’ Effective Altruism Forum.

Askell, Amanda. 2019. ‘Evidence Neutrality and the Moral Value of Information.’ In Effective Altruism: Philosophical Issues, edited by Hilary Greaves and Theron Pummer. Oxford: Oxford University Press.

Berkey, Brian. 2018. ‘The Institutional Critique of Effective Altruism.’ Utilitas 30 (2): 143-171.

Berkey, Brian. 2021. ‘The Philosophical Core of Effective Altruism.’ Journal of Social Philosophy 52 (1): 93-115.

Bisson, Keven. 2021. Effective Altruisms and the Institutional Critiques Moral Frameworks. Antwerp: University of Antwerp Press.

Brennan, Jason. 2025. ‘The Effective Altruist Approach to Politics (Effective Altruism Definitions Sequence).’ Online essay.

Broi, Antonin. 2019. ‘Effective Altruism and Systemic Change.’ Utilitas 31 (3): 262-276. https://doi.org/10.1017/S0953820819000153.

Caramani, Daniele. 2017. ‘Will vs. Reason: The Populist and Technocratic Forms of Political Representation and Their Critique to Party Government.’ American Political Science Review 111 (1): 54-67.

Christiano, Thomas. 2008. The Constitution of Equality: Democratic Authority and Its Limits. Oxford: Oxford University Press.

Clough, Emily. 2015. ‘Effective Altruism’s Political Blind Spot.’ Boston Review, July 14, 2015.

Collins, Stephanie. 2019. ‘Beyond Individualism.’ In Effective Altruism: Philosophical Issues, edited by Hilary Greaves and Theron Pummer, 202-215. Oxford: Oxford University Press.

Cote, Nicolas, and Bastian Steuwer. 2023. ‘Better Vaguely Right Than Precisely Wrong in Effective Altruism: The Problem of Marginalism.’ Economics & Philosophy 39: 152-169. https://doi.org/10.1017/S0266267122000062.

Dahl, Robert A. 1999. ‘Can International Organizations Be Democratic? A Skeptic’s View.’ In Democracy’s Edges, edited by Ian Shapiro and Casiano Hacker-Cordon, 19-36. Cambridge: Cambridge University Press.

Dietz, Alexander. 2019. ‘Effective Altruism and Collective Obligations.’ Utilitas 31 (1): 106-115. https://doi.org/10.1017/S0953820818000158.

Estlund, David M. 2008. Democratic Authority: A Philosophical Framework. Princeton, NJ: Princeton University Press.

Ferguson, James. 1990. The Anti-Politics Machine: Development, Depoliticization, and Bureaucratic Power in Lesotho. Cambridge: Cambridge University Press.

Freiman, Christopher. 2025. ‘Is Effective Altruism Anti-political?’ Social Philosophy and Policy 42 (1): 204-222. https://doi.org/10.1017/S0265052525000111.

Fukuyama, Francis. 2002. Our Posthuman Future: Consequences of the Biotechnology Revolution. New York: Farrar, Straus and Giroux.

Gabriel, Iason. 2017. ‘Effective Altruism and Its Critics.’ Journal of Applied Philosophy 34 (4): 457-473.

Gabriel, Iason, and Brian McElwee. 2019. ‘Effective Altruism, Global Poverty, and Systemic Change.’ In Effective Altruism: Philosophical Issues, edited by Hilary Greaves and Theron Pummer. Oxford: Oxford University Press.

Gold, Marthe R., David Stevenson, and Dennis Fryback. 2002. ‘HALYs and QALYs and DALYs, Oh My: Similarities and Differences in Summary Measures of Population Health.’ Annual Review of Public Health 23: 115-134.

Halstead, John. 2017. ‘High Stakes Instrumentalism and Democracy.’ Working paper.

Hughes, James. 2004. Citizen Cyborg: Why Democratic Societies Must Respond to the Redesigned Human of the Future. Boulder, CO: Westview Press.

Kissel, Joshua. 2017. ‘Effective Altruism and Anti-Capitalism: An Attempt at Reconciliation.’ Essays in Philosophy 18 (1). https://doi.org/10.7710/1526-0569.1573.

Koh, Jeremias. 2024. "Utilitarianism and Effective Altruism." PhD diss., National University of Singapore.

Kucz, Mateusz, and Piotr Rosol. 2020. ‘The Ethical, Political and Economic Challenges of Effective Altruism.’ Ethics in Progress 11 (1): 61-76. https://doi.org/10.14746/eip.2020.1.3.

Kuper, Andrew. 2002. ‘More Than Charity: Cosmopolitan Alternatives to the Singer Solution.’ Ethics & International Affairs 16 (1): 107-120.

Kuruc, Kevin, and David Manley. 2025. ‘The Ethics, Economics, and Demographics of Delaying Aging.’ In Essays on Longtermism: Present Action for the Distant Future, edited by Hilary Greaves, Jacob Barrett, and David Thorstad, 428-448. Oxford: Oxford University Press.

Landemore, Helene. 2013. Democratic Reason: Politics, Collective Intelligence, and the Rule of the Many. Princeton, NJ: Princeton University Press.

Lechterman, Theodore M. 2020. ‘The Effective Altruists’ Political Problem.’ Polity 52 (1): 88-115. https://doi.org/10.1086/706867.

Lovett, Frank. 2022. ‘Republicanism.’ The Stanford Encyclopedia of Philosophy (Fall 2022 Edition), edited by Edward N. Zalta and Uri Nodelman. https://plato.stanford.edu/archives/fall2022/entries/republicanism/.

MacAskill, William. 2013. ‘What Warren Buffett’s Son Doesn’t Understand About the World.’ Quartz, September 11, 2013.

MacAskill, William. 2015. Doing Good Better: How Effective Altruism Can Help You Help Others, Do Work that Matters, and Make Smarter Choices about Giving. New York: Gotham Books.

MacAskill, William. 2019. ‘The Definition of Effective Altruism.’ In Effective Altruism: Philosophical Issues, edited by Hilary Greaves and Theron Pummer, 10-28. Oxford: Oxford University Press.

MacAskill, William, and Theron Pummer. 2020. ‘Effective Altruism.’ In International Encyclopedia of Ethics. Wiley-Blackwell.

McMahan, Jeff. 2016. ‘Philosophical Critiques of Effective Altruism.’ The Philosophers’ Magazine 73: 92-99.

Moore, Alfred. 2017. Critical Elitism: Deliberation, Democracy, and the Problem of Expertise. Cambridge: Cambridge University Press.

Muyskens, Kathryn. 2017. ‘The Other Half of Effective Altruism: Selective Asceticism.’ Essays in Philosophy 18 (1). https://doi.org/10.7710/1526-0569.1575.

Ngo, Richard, and Adam Bales. 2025. ‘AGI Safety from First Principles.’ In Essays on Longtermism: Present Action for the Distant Future, edited by Hilary Greaves, Jacob Barrett, and David Thorstad. Oxford: Oxford University Press.

Orviska, Marta, Anetta Caplanova, and John Hudson. 2014. ‘The Impact of Democracy on Life Satisfaction.’ Journal of Happiness Studies 15 (2): 493-510.

Parfit, Derek. 1984. Reasons and Persons. Oxford: Oxford University Press.

Reich, Rob. 2015. ‘Response to Effective Altruism.’ Boston Review.

Rubenstein, Jennifer C. 2016. ‘The Lessons of Effective Altruism.’ Ethics & International Affairs 30 (4): 511-526.

Rubin, Olivier. 2009. ‘The Merits of Democracy in Famine Protection - Fact or Fallacy?’ European Journal of Development Research 21: 699-717.

Saunders-Hastings, Emma. 2019. ‘Benevolent Giving and the Problem of Paternalism.’ In Effective Altruism: Philosophical Issues, edited by Hilary Greaves and Theron Pummer, 115-136. Oxford: Oxford University Press.

Schmidt, Andreas T., and Jacob Barrett. 2025. ‘Longtermist Political Philosophy: An Agenda for Future Research.’ In Essays on Longtermism: Present Action for the Distant Future, edited by Hilary Greaves, Jacob Barrett, and David Thorstad. Oxford: Oxford University Press.

Sebo, Jeff, and L. A. Paul. 2019. ‘Effective Altruism and Transformative Experience.’ In Effective Altruism: Philosophical Issues, edited by Hilary Greaves and Theron Pummer. Oxford: Oxford University Press.

Sen, Amartya. 1982. Poverty and Famines: An Essay on Entitlement and Deprivation. Oxford: Oxford University Press.

Shannon, Vaughn P. 2005. ‘Wendt’s Violation of the Constructivist Project: Agency and Why a World State Is Not Inevitable.’ European Journal of International Relations 11 (4): 581-587.

Shulman, Carl, and Elliott Thornley. 2025. ‘How Much Should Governments Pay to Prevent Catastrophes? Longtermism’s Limited Role.’ In Essays on Longtermism: Present Action for the Distant Future, edited by Hilary Greaves, Jacob Barrett, and David Thorstad. Oxford: Oxford University Press.

Singer, Peter. 1999. A Darwinian Left: Politics, Evolution, and Cooperation. London: Weidenfeld & Nicolson.

Singer, Peter. 2015a. The Most Good You Can Do: How Effective Altruism Is Changing Ideas about Living Ethically. New Haven, CT: Yale University Press.

Singer, Peter. 2015b. ‘The Logic of Effective Altruism.’ Boston Review, July 1.

Singer, Peter. 2015c. ‘Reply to Effective Altruism Responses.’ Boston Review, July 1.

Srinivasan, Amia. 2015. ‘Stop the Robot Apocalypse: The New Utilitarians.’ London Review of Books 37 (18): 3-6.

Syme, Timothy. 2019. ‘Charity vs. Revolution: Effective Altruism and the Systemic Change Objection.’ Ethical Theory and Moral Practice 22 (1): 93-120.

Venkatesh, Siddharth. 2023. Utilitarianism and the Social Nature of Persons. PhD diss., University of Oxford.

Zuolo, Federico. 2019. ‘Beyond Moral Efficiency: Effective Altruism and Theorizing about Effectiveness.’ Utilitas 32 (1): 1-14.