These conditions are foundational and they are failing.

1. Democratic legitimacy presupposes epistemic conditions

The standard accounts of democratic legitimacy, whether deliberative, proceduralist, or participatory, all assume a shared informational environment. Citizens need not agree, but they must be reasoning about something real. Dewey saw this. Habermas built it into his theory of communicative rationality. Even accounts that locate legitimacy entirely in fair procedure implicitly require that participants are engaging with a common reality.

This is a claim about minimum conditions. Without them, the vote aggregates something, but it is not clear that what it aggregates carries the authorizing weight democratic theory assigns to it.

2. Those conditions are failing structurally

Three vectors of deterioration are operating simultaneously.

Fragmentation has dissolved the shared informational anchors that once gave citizens enough common ground to disagree productively. Citizens increasingly inhabit non-overlapping factual worlds. Disagreement no longer takes the form of competing evaluations of shared evidence.

Incentive misalignment means that the dominant platforms for public discourse are optimized for engagement, and engagement and accuracy are not the same objective. The result is epistemic selection pressure: a structural condition in which the beliefs that survive and circulate are those that maximize behavioral response, instead of those that accurately represent the world.

Fabrication has become cheap and verification has become expensive. This asymmetry widens as AI tools improve and has no obvious equilibrium.

3. The mechanism is optimization

The epistemic deterioration described above is the predictable output of systems designed to maximize engagement operating at global scale and continuous operation. The information environment selects for beliefs. What it selects for is not, in any straightforward sense, the truth.

Verification does not scale at the same rate as fabrication. Corrections spread more slowly than the content they correct. Expert authority is contested by the same mechanisms that produce the misinformation it is trying to correct. The system has no self-correcting tier.

4. Repair faces constraints that compound

The natural response to this diagnosis is repair by regulating platforms, investing in institutions, improving media literacy, to restore the shared epistemic baseline. Working Paper 3 takes this response seriously and examines what it actually requires.

Repair requires simultaneous progress across four independent systems: platform incentives, verification capacity, institutional trust and individual cognition, each of which faces its own resistance. The constraints interact, so failure in one undermines progress in others.

Incentive lock-in means platforms have strong pressure to revert to engagement optimization regardless of regulatory intervention. The repair asymmetry means fabrication will continue to outpace verification under current technological trajectories. Trust fragmentation means the institutions required to arbitrate disputes are themselves contested by the conditions that made repair necessary. Cognitive limits mean individual epistemic improvement does not produce the population-level convergence democratic legitimacy requires.

Repair interventions face a further problem. In conditions of deep trust fragmentation, interventions designed to restore epistemic conditions are available to be read as power consolidation. The effort to restore shared reality is processed through the distorted epistemic conditions it is trying to correct.

Local improvements are possible. Whether repair can occur at the scale democratic legitimacy requires is a different question, and the answer is less comfortable than the repair literature acknowledges.

5. If repair is insufficient, the problem is theoretical

The repair project operates within democratic theory. It takes the normative framework as given and asks what empirical conditions must be restored for that framework to function.

If those conditions cannot be reliably restored, the problem is about the relationship between the framework and the conditions under which it was developed, conditions that no longer obtain and may not be recoverable.

The theory of democratic legitimacy may require revision to remain coherent under conditions its architects did not anticipate.

6. Revision is a live theoretical space

Working Paper 4 maps five possible responses to the epistemic problem, in order of how much they preserve the participatory basis of democratic legitimacy.

Epistemically constrained democracy restructures deliberation through devices like citizens’ assemblies without abandoning the participatory framework. Information-gated participation ties voting rights to demonstrated epistemic competence. Technocratic augmentation increases the role of experts and insulated institutions in political decision-making. Managed information environments have legitimate authority actively shape the epistemic conditions under which citizens form beliefs. Decentralized governance accepts non-convergence and reduces the scale at which governance must operate.

Every path that improves the epistemic quality of collective decision-making does so by reducing the equal standing of citizens whose inputs are epistemically compromised. There is no revision path that dissolves that tension. There are only different positions within it.

None of these paths solves the theoretical question of what justifies governance when its traditional participatory foundations are absent. That question is the most consequential open problem in contemporary political theory, and it has not been adequately confronted.

7. The Legitimacy Project is an investigation

This project does not advocate for any of the revision paths above. Its aim is to establish the problem with the precision it deserves, as a precondition for evaluating the available responses.

The evaluation proceeds according to empirical criteria. Feasibility asks whether a governance form can actually be implemented given current platform economics, regulatory capacity and technological trajectories. Stability asks whether it can persist over time without requiring constant coercion or the suppression of the mechanisms that would otherwise undermine it. Justificatory coherence asks whether the operating logic of the system is publicly defensible to those subject to it, since a system that depends on citizens not understanding how it works fails to be legitimate even if it produces good outcomes.

None of the revision paths currently satisfies all of the criteria with confidence.

A political theory that presupposes epistemic conditions that cannot be reliably restored must eventually confront that fact. What replaces the inherited assumptions remains an open question. The Legitimacy Project’s contribution is to establish that it is no longer hypothetical.

Working Papers 1–4 are available on this Substack. Working Paper 1 establishes the epistemic preconditions of democratic legitimacy. Working Paper 2 explains the mechanism of deterioration. Working Paper 3 examines the limits of repair. Working Paper 4 maps the space of revision.

BEYOND REPAIR

Rethinking Democratic Legitimacy Under Epistemic Constraint

The erosion of epistemic conditions does not eliminate the need for governance. It eliminates the comfort of inherited assumptions about how governance is justified.

By The Legitimacy Project | April 2026 | Working paper

I. FROM INSTITUTIONAL DESIGN TO THEORETICAL REVISION

The first three working papers in this series established a sequence. Democratic legitimacy presupposes epistemic conditions. Those conditions are failing under systematic structural pressure. The repair project, which attempts to restore them from within the democratic framework, faces constraints that are deep enough to cast doubt on its sufficiency at democratic scale. The present paper asks what follows from that sequence if the doubt is warranted.

The repair project operates within democratic theory. When it confronts the limits identified in working paper 3, it tends to respond by refining the repair strategy rather than questioning the framework. That response is reasonable if the gap between what democratic legitimacy requires and what the current information environment can produce is bridgeable by institutional means. The analysis in working paper 3 suggests it may not be. If that assessment holds, the problem shifts from institutional design to theoretical revision, and the question becomes how to think about legitimacy under conditions where its traditional foundations do not reliably obtain.

This paper maps the space of possible responses without advocating for any of them. The appropriate analogy is to a systems theorist examining design options under a given set of constraints, rather than to a political theorist defending a preferred model. Each response involves tradeoffs that are real and significant, and the paper’s aim is to make those tradeoffs explicit. Identifying the available options and their costs is a precondition for any serious subsequent evaluation, and that evaluation has not yet been adequately undertaken in the existing literature.

I proceed as follows. Section 2 specifies the constraints that any viable response must satisfy. Sections 3 through 7 examine five revision paths in order of how much they preserve the participatory basis of democratic legitimacy. Section 8 identifies the tradeoffs that cut across all five paths. Section 9 develops the deeper theoretical problem that revision raises. Section 10 states the position of this project in relation to that problem.

II. WHAT ANY VIABLE SYSTEM MUST ADDRESS

Before examining revision paths, it is necessary to specify what any viable governance system must accomplish, independent of its particular structure. This is a question about functional requirements, specifically what any system that aspires to govern a modern political community must meet to remain stable and justifiable over time.

Any viable system must provide mechanisms for collective decision-making on questions that affect the population as a whole, since the coordination problems that governance exists to solve do not disappear under epistemic constraint. It must generate at least a working basis of legitimacy, meaning a reason, even if attenuated or contested, why those subject to decisions accept their binding force rather than simply resist. It must maintain some relationship to epistemic adequacy, in the sense that decisions must be connected, however imperfectly, to accurate information about the conditions they are meant to address. And it must be stable enough to persist over time without constant crisis and renegotiation.

These requirements pull against each other in ways that the revision paths below make visible. A system that maximizes epistemic adequacy may sacrifice legitimacy. One that maximizes legitimacy under non-convergent epistemic conditions may sacrifice epistemic performance. The task is to understand the tradeoff structure, not to dissolve it.

III. REVISION PATH 1: EPISTEMICALLY CONSTRAINED DEMOCRACY

The revision path most continuous with existing democratic theory introduces epistemic constraints on democratic participation without abandoning the participatory framework. Citizens retain formal equality before the law and in electoral processes, but deliberative mechanisms are structured to filter out or weight down the influence of epistemically degraded inputs. The clearest contemporary expressions of this path are deliberative mini-publics, randomly selected citizen assemblies that deliberate on constrained agendas under conditions designed to favor informed discussion over reactive preference expression.1

The appeal of this path is that it attempts to restore the epistemic conditions democratic legitimacy requires by restructuring the deliberative process rather than by altering the information environment. It accepts that the broader epistemic environment cannot be repaired and attempts to create insulated deliberative spaces that are less subject to the selection pressures described in working paper 2. Within those spaces, the conditions for genuine deliberation, including shared information, expert input, and sustained reasoning under low cognitive load, are manufactured by design.

The tradeoff is between epistemic performance and equality of participation. Deliberative mini-publics improve the quality of deliberation by limiting participation to a randomly selected subset of the population and by structuring their information environment in ways that the broader population does not experience. The legitimacy of their outputs depends on the representativeness of the selection and on the acceptance of their authority by those not included. Both of these conditions are fragile in exactly the conditions of trust fragmentation that make repair difficult. A population that distrusts expert institutions has limited grounds for trusting the outputs of a deliberative process designed and administered by those institutions.2

IV. REVISION PATH 2: INFORMATION-GATED PARTICIPATION

A more direct response to the epistemic problem ties participation to demonstrated informational competence. On this path, some form of qualification, whether verified knowledge of the issues at stake or assessed reasoning ability, mediates access to consequential decision-making. The idea has a long history in political philosophy, from Mill’s plural voting scheme to contemporary proposals for epistocracy, and it is the revision path that most directly targets the mechanism of epistemic failure.3

If the problem is that democratic decision-making aggregates preferences formed under epistemically distorted conditions, a system that conditions participation on epistemic competence addresses the problem at its source. Decisions would be made by citizens whose beliefs are at least minimally reliable, and the aggregation problem that working paper 1 identified, that the vote may no longer track what legitimizes it, would be at least partially dissolved.

The design of competence criteria is itself a political act, and whoever controls that design controls which citizens participate. The history of literacy tests and similar qualification mechanisms demonstrates that epistemic criteria are reliably available to be shaped by existing power structures to exclude the populations those structures prefer to exclude.4

Even a well-designed competence criterion faces the further problem that competence, in a non-convergent epistemic system, is itself contested. If citizens disagree about what counts as reliable evidence, they will disagree about what counts as knowledge, and the authority to adjudicate that disagreement must come from somewhere. That authority shift problem, the question of who legitimately decides what competence requires, is the deepest difficulty this path faces and the one its proponents have been least willing to confront directly.

V. REVISION PATH 3: TECHNOCRATIC AUGMENTATION

Technocratic augmentation increases the role of experts and specialized institutions in political decision-making, shifting authority along a spectrum from advisory to influential to determinative. In its weaker forms, this path is already embedded in democratic governance, regulatory agencies, central banks, and judicial bodies all exercise forms of insulated authority over consequential decisions. The revision path under consideration extends that insulation more broadly, in response to the argument that mass participation under epistemically degraded conditions produces systematically worse outcomes than decisions made by smaller groups of appropriately qualified actors.5

If citizen belief formation is subject to the structural distortions described in working papers 1 through 3, and if those distortions systematically bias democratic outputs away from accurate assessments of policy problems, then there are outcome-based grounds for increasing the epistemic quality of decision-making by reducing the role of epistemically compromised inputs. The argument requires only that expert institutions in domains where expertise is well-defined and verifiable produce more reliable outputs than mass deliberation under non-convergent epistemic conditions.

The tradeoff is between epistemic performance and legitimacy, and it is more severe on this path than on the previous two. Technocratic authority derives its legitimacy from a combination of demonstrated competence and democratic delegation. The public accepts expert authority because it has been granted by democratic processes and remains accountable to them. Both sources of legitimacy are under strain in conditions of trust fragmentation. Expert competence is contested by populations whose information environments have systematically cultivated that contestation. Democratic delegation is weakened when the delegating processes are themselves epistemically degraded.6

VI. REVISION PATH 4: MANAGED INFORMATION ENVIRONMENTS

Managed information environments attempt to address the epistemic problem indirectly, by restructuring the conditions under which citizens form beliefs rather than by altering the structure of participation itself. On this path, legitimate authority actively shapes the information environment through curation, filtering, or restriction, with the aim of reducing the epistemic selection pressure that working paper 2 identified as the primary mechanism of deterioration. The goal is to restore the epistemic conditions democratic legitimacy requires rather than to work around their absence.

This path is distinct from the repair project examined in working paper 3. The repair project attempts to restore epistemic conditions through interventions that are themselves subject to democratic authorization and constrained by democratic norms, including norms of free expression. Managed information environments involve a more direct exercise of authority over the epistemic environment, one that may require setting aside those norms to achieve its aims. The distinction matters because the self-undermining problem identified in working paper 3 is most acute on this path. An authority that controls the information environment controls the conditions under which its own legitimacy is assessed. That circularity is potentially manageable in environments of high trust. In conditions of trust fragmentation, it is politically explosive.7

The appeal of this path is that it targets the correct level of the problem. If epistemic conditions are structural rather than individual, then structural interventions are appropriate and individual-level approaches, including deliberative mini-publics and competence criteria, will be insufficient. The cost is that structural intervention in the information environment concentrates the authority to define epistemic quality in whoever administers the intervention, and the history of such authority suggests that it is reliably subject to capture by the political interests of the administering body. The path toward epistemic restoration and the path toward epistemic control are, from the outside, difficult to distinguish.

VII. REVISION PATH 5: DECENTRALIZED GOVERNANCE

The four revision paths examined so far attempt, by different means, to maintain or restore the epistemic conditions that legitimate governance requires at the scale of modern democratic states. Decentralized governance takes a different approach. Instead of attempting to produce epistemic convergence across large and heterogeneous populations, it accepts the reality of non-convergent epistemic systems and responds by reducing the scale at which governance must operate.

If citizens inhabiting different information ecosystems share insufficient epistemic common ground for legitimate collective decision-making at national scale, the response may be to devolve consequential decisions to smaller political units where epistemic communities are more homogeneous and shared informational baselines more achievable. Within such units, the preconditions for democratic legitimacy are at least partially recoverable. The coordination and aggregation problems that make democratic legitimacy difficult at scale are reduced when the population is smaller and more epistemically coherent.8

The costs of this path are concentrated in what it gives up. Many of the governance problems that modern states must address, like climate change, pandemic response, financial stability, and security, are not tractable at the scale of epistemically homogeneous local communities. Decentralization that is genuinely responsive to the epistemic problem would require accepting a significant reduction in the scope of collective decision-making. The problems that require coordination across epistemically divergent populations would either go unaddressed or be addressed by mechanisms that are not themselves democratically legitimate at the required scale.

VIII. WHAT EVERY PATH COSTS

The five revision paths involve tradeoffs that are structural, meaning that movement toward one value requires accepting costs on another. Making those tradeoffs explicit is the precondition for evaluating the paths honestly.

The central tradeoff runs between epistemic quality and equality of participation. Every path that improves the epistemic quality of collective decision-making does so by reducing the influence of epistemically compromised inputs. Reducing that influence means reducing the equal standing of the citizens providing those inputs. The democratic commitment to political equality and the epistemic commitment to reliable belief formation are, under conditions of degraded epistemic environments, in genuine tension. There is no revision path that resolves that tension rather than accepting a particular position within it.

A second tradeoff runs between legitimacy and performance. Governance systems derive their legitimacy partly from outcomes and partly from process. Epistemic constraint on the process tends to improve performance at the cost of process legitimacy. Epistemic openness tends to preserve process legitimacy at the cost of performance. The conditions under which a governance system can have both are precisely the conditions that working papers 1 through 3 have established are currently absent.

A third tradeoff runs between stability and justificatory coherence. A governance system that maintains democratic forms while quietly concentrating epistemic authority in insulated institutions may be more stable than one that openly acknowledges the tradeoffs it has made. A system that is stable only because its operating logic is not publicly acknowledged is, however, justificatorily incoherent. The gap between the democratic self-description and the technocratic reality is itself a source of epistemic distortion, producing citizens whose beliefs about their own governance system are systematically inaccurate, a condition generated by the governance system itself.

IX. WHAT REPLACES PARTICIPATION AS THE SOURCE OF LEGITIMACY

The revision paths and their tradeoffs converge on a theoretical problem that none of them resolves. Democratic legitimacy, in its canonical formulations, derives from the equal participation of citizens in collective self-governance. Participation under conditions of shared epistemic access generates decisions that citizens can recognize as, in some sense, their own, even when they disagree with particular outcomes. It is this participatory basis that gives democratic legitimacy its distinctive character and that distinguishes it from legitimacy grounded in outcomes alone, or in the historical authority of tradition.9

Every revision path examined above either reduces participation, conditions it, or restructures the informational environment in which it occurs. In each case, the participatory basis of legitimacy is attenuated. The question that follows is whether any of the remaining sources of legitimacy are adequate substitutes. Outcomes are insufficient on their own, because a governance system whose legitimacy depends entirely on producing good outcomes has no basis for legitimacy during the periods when its outputs are contested or poor. Expertise is insufficient on its own, because the authority problem means that the designation of expertise requires a legitimating basis that expertise itself cannot provide.

The theory of legitimacy adequate to the current condition has not yet been developed. The revision paths map the available design space, but they do not solve the theoretical problem of what justifies governance when its traditional foundations are absent. That problem is the most consequential open question in contemporary political theory, and it is one the field has been slow to engage with the directness it requires, partly because engaging with it directly requires acknowledging that the epistemic conditions democratic theory assumes are not reliably present.

X. AN INVESTIGATION WITHOUT A CONCLUSION

The Legitimacy Project does not begin from a commitment to any of the revision paths examined above. Its aim across these four working papers has been to establish the problem with the precision and seriousness it deserves, as a precondition for evaluating the available responses. The sequence of preconditions, mechanism of breakdown, limits of repair, space of revision is an investigative structure aimed at clearing the ground rather than arriving at a predetermined destination.

The evaluation of revision paths requires criteria that this project treats as empirical questions rather than as normative commitments: feasibility under actual technological and institutional conditions, stability over time given the incentive structures that actually operate, and justificatory coherence in the sense that the operating logic of the system is at least in principle publicly defensible to those subject to it. These are demanding criteria, and none of the revision paths examined above satisfies all three with confidence. That is the finding, and it is an uncomfortable one.

The discomfort is appropriate, as a political theory that presupposes epistemic conditions that cannot be reliably restored must eventually confront that fact. The alternatives to confronting it are to ignore the problem, which produces governance systems that are less legitimate than they appear, or to acknowledge it only partially, which produces governance systems whose self-descriptions are systematically misleading. Neither alternative is satisfactory, and the cost of each is borne primarily by the populations subject to governance rather than by the theorists describing it.

What replaces the inherited assumptions remains an open question. The Legitimacy Project’s contribution is to establish that it is no longer a hypothetical one.

REFERENCES

Achen, Christopher H. and Larry M. Bartels, 2016, Democracy for Realists, Princeton: Princeton University Press.

Anderson, Elizabeth, 2006, “The Epistemology of Democracy,” Episteme, 3(1-2): 8-22.

Balkin, Jack M., 2018, “Free Speech in the Algorithmic Society,” UC Davis Law Review, 51(3): 1149-1210.

Brennan, Jason, 2016, Against Democracy, Princeton: Princeton University Press.

Caplan, Bryan, 2007, The Myth of the Rational Voter, Princeton: Princeton University Press.

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

Cohen, Joshua, 1989, “Deliberation and Democratic Legitimacy,” in Alan Hamlin and Philip Pettit (eds.), The Good Polity, Oxford: Basil Blackwell, 17-34.

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

Fishkin, James S., 2018, Democracy When the People Are Thinking, Oxford: Oxford University Press.

Hayek, Friedrich A., 1945, “The Use of Knowledge in Society,” American Economic Review, 35(4): 519-530.

Kolodny, Niko, 2014a, “Rule Over None I: What Justifies Democracy?” Philosophy and Public Affairs, 42(3): 195-229.

Kolodny, Niko, 2014b, “Rule Over None II: Social Equality and the Justification of Democracy,” Philosophy and Public Affairs, 42(4): 287-336.

Lafont, Christina, 2019, Democracy Without Shortcuts, Oxford: Oxford University Press.

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

Mill, John Stuart, 1861 [1991], Considerations on Representative Government, Buffalo: Prometheus Books.

Peter, Fabienne, 2009, Democratic Legitimacy, New York: Routledge.

Rousseau, Jean-Jacques, 1762, The Social Contract, Charles Frankel (trans.), New York: Hafner, 1947.

Zittrain, Jonathan, 2019, “The Problem of Infodemic: How Dominant Platforms Can Serve the Public Interest,” in Disinformation, ‘Fake News’ and Influence Campaigns, Matthew Bunker (ed.), London: Routledge.

THE LIMITS OF EPISTEMIC REPAIR

On the Feasibility of Restoring the Conditions Democratic Legitimacy Requires

The repair project is serious, and it deserves a serious examination of what it actually requires.

By The Legitimacy Project | April 2026 | Working paper

I. THE SERIOUS BASELINE POSITION

The previous two working papers established that democratic legitimacy presupposes epistemic conditions that the contemporary information environment is systematically degrading. A natural response to that diagnosis is repair: identify the mechanisms of degradation and intervene to correct them. Regulate platforms. Invest in fact-checking and stronger institutions. Restore, as far as possible, the shared informational baseline that democratic theory assumes.

This is a serious position, and it should be treated as one. The repair project accepts that epistemic conditions have deteriorated and proposes to restore them from within the democratic framework, preserving both the normative commitments of democratic theory and the institutional structures through which those commitments are realized. It is the position most consistent with continuity, and continuity has genuine value when the alternatives are uncertain.

The question this paper asks is whether repair is feasible at the scale democratic legitimacy requires. Repair may be possible in local or partial ways, and that possibility is worth preserving. The more pressing question is whether repair, understood as the systematic restoration of the epistemic conditions democratic legitimacy presupposes, can succeed against a set of independent structural constraints that compound in ways its proponents have not adequately confronted.

I proceed as follows. In section 2, I define repair in operational terms that make its requirements explicit. Sections 3 through 7 examine five structural constraints on those requirements. Section 8 synthesizes the constraints and introduces the distinction between local repair and the scale repair requires. Section 9 opens the question that follows.

II. DEFINING THE PROBLEM OPERATIONALLY

Most discussion of epistemic repair operates at the level of policy proposals, such as content moderation rules, algorithmic transparency requirements, media literacy curricula, and public interest journalism funding. These proposals are worth examining on their own terms, and some may have merit in specific domains. The more fundamental question is what conditions any such proposal must satisfy to count as repair at the level of democratic legitimacy rather than merely improvement at the margins.

Repair, in the relevant sense, requires four things operating simultaneously. Platform incentives must be realigned so that engagement and accuracy are no longer systematically opposed. A shared informational baseline must be restored, meaning some convergence, across sufficiently large populations, in what counts as evidence and which sources can arbitrate disputes. Verification mechanisms must scale to process contested claims at the volume they are produced. And epistemic authorities, the institutions whose function is to adjudicate factual disputes, must recover sufficient trust to perform that function across groups with different prior commitments.

Each of these is a demanding condition on its own. The repair project requires all four simultaneously, because the conditions are interdependent. Restored verification mechanisms are useless if trust in the institutions running them is absent. Realigned platform incentives cannot produce convergent beliefs if the informational substrate remains fragmented. Repair is accordingly a coordinated transformation of the information environment across multiple layers, each of which faces its own structural resistance.

III. CONSTRAINT 1: INCENTIVE LOCK-IN

The first constraint is that platform incentives are not freely adjustable. Working paper 2 established that the divergence between engagement and epistemic quality is a predictable consequence of the platform business model, not a design choice that could be revised without disrupting the underlying economics. That point has a further dimension when examined as a constraint on repair.

The revenue model of dominant information platforms depends on advertising, and advertising revenue scales with engagement. Engagement correlates with emotional activation, and emotional activation is systematically inversely related to epistemic quality. A platform that successfully realigned its optimization target from engagement to accuracy would, under current market conditions, generate less revenue than its competitors. The pressure to revert is structural, not contingent on the preferences of platform managers.1

Regulatory intervention could in principle alter this equilibrium by imposing costs on engagement-maximizing behavior or mandating accuracy-oriented design. The history of platform regulation suggests that such interventions face significant resistance, move slowly relative to the pace of platform development, and tend to be implemented in forms that preserve the core incentive structure while modifying its surface features.2

This observation bears on the feasibility of regulation as a repair mechanism, though it does not settle whether regulation should be attempted. The lock-in is stable enough that dislodging it requires sustained political will and regulatory capacity that have not been demonstrated at the scale required.

The deeper problem is that incentive lock-in affects the repair effort itself. Organizations advocating for platform reform operate within the same information environment they are trying to change. Their communications are subject to the same selection pressures that favor emotionally activating content over careful argument. The reform effort is not insulated from the condition it is trying to correct.

IV. CONSTRAINT 2: REPAIR ASYMMETRY

The second constraint follows directly from the analysis in working paper 2. Fabrication scales cheaply. Verification scales slowly and at high cost. Any repair strategy that depends on verification keeping pace with fabrication faces a persistent and widening asymmetry.

The repair project must outrun production. Under current technological conditions, that requirement is not being met and the trajectory is adverse. The tools that reduce the cost of fabrication are improving faster than the tools that reduce the cost of verification. Fact-checking organizations, however well-resourced, are processing a finite volume of claims against an effectively unbounded production capacity. Automated verification tools reduce the cost of some forms of checking, but they do so against a background of fabrication tools that are improving at least as rapidly and that have been deliberately optimized to evade detection.3

The repair asymmetry is also a coordination problem. Verification at democratic scale requires agreement across institutions about what constitutes sufficient evidence for a claim. In a fragmented information environment, that agreement is itself contested. Different institutions apply different standards, and in conditions of trust fragmentation, the existence of multiple verification regimes provides grounds for dismissing any particular verdict. A system in which verification is decentralized and contested cannot perform the epistemic function that democratic legitimacy requires, regardless of how much resource is invested in individual verification efforts.

V. CONSTRAINT 3: TRUST FRAGMENTATION

The third constraint is that repair presupposes trust that the conditions requiring repair have already degraded. The restoration of shared epistemic baselines requires institutions capable of arbitrating between competing claims with authority that is broadly accepted. Those institutions do not currently exist in the relevant sense, and the mechanisms that would produce them are themselves subject to the conditions that have undermined trust.

Trust in epistemic authorities, such as scientific institutions, journalistic organizations, regulatory bodies, and courts, has declined across most democratic societies over the past several decades. The causes are multiple and contested, but among them are institutional failures that were genuine, manufactured doubt campaigns that were deliberate, and the selection dynamics described in working paper 2 that systematically amplify the impression of expert disagreement beyond its actual extent.4

Whatever the causal mix, the functional result is that there is no neutral arbiter whose authority is accepted across the populations whose beliefs democratic legitimacy requires to be at least partially convergent.

This creates a circularity the repair project has difficulty escaping. Repair requires trusted institutions. Building trusted institutions requires a period of demonstrated reliability. Demonstrating reliability requires operating in an information environment that does not systematically distort the signal. The information environment that distorts the signal is precisely what repair is supposed to fix. The precondition for repair is the outcome repair is supposed to produce.5

A further dimension of this constraint is that corrections are interpreted through identity filters. Working paper 2 noted that citizens are rational with respect to the information environments they actually inhabit. An implication of that observation is that corrections from distrusted sources are processed as evidence of the source’s untrustworthiness rather than as evidence against the corrected claim. Increased investment in correction, delivered through channels that are already distrusted, may reinforce the prior rather than revise it.6

VI. CONSTRAINT 4: COGNITIVE LIMITS

The fourth constraint operates at the level of individual cognition rather than institutional design, but it bears directly on the feasibility of repair strategies that depend on individual epistemic improvement.

A significant strand of the repair literature places weight on media literacy interventions: teaching citizens to evaluate sources, identify manipulation, and apply critical scrutiny to claims they encounter. The intuition behind this approach is that epistemic failure is partly a deficit of skill, and skills can be taught. The empirical literature on motivated reasoning gives substantial grounds for skepticism about this intuition.7

The relevant finding, documented extensively by Kahan and others, is that increased cognitive sophistication does not reliably produce better-calibrated beliefs about contested political questions. Citizens with higher analytical ability are better at finding and processing information that confirms their prior commitments, which means that improvements in individual cognitive skill can amplify instead of correct the divergence that epistemic selection pressure produces. The mechanism that media literacy interventions aim to improve is not straightforwardly improvable in the direction repair requires.8

The relationship between individual epistemic skill and the population-level epistemic conditions democratic legitimacy requires is accordingly not the straightforward one that media literacy programs tend to assume. The conditions are structural, and structural conditions are not dissolved by improving the individuals operating within them. Citizens reasoning more carefully within a non-convergent epistemic system may simply produce better-reasoned divergent conclusions.

VII. CONSTRAINT 5: THE SELF-UNDERMINING PROBLEM

The fifth constraint is the most structurally interesting, and it is the one the repair literature has been slowest to confront. Repair interventions are not epistemically neutral. They are exercises of power over the information environment, and the conditions under which they are legitimate are themselves subject to the epistemic conditions that repair is supposed to restore.

Consider the range of interventions the repair project requires. Platform regulation mandates changes to how information is amplified, which is an intervention in speech at scale. Institutional investment in trusted arbiters means deciding which institutions receive epistemic authority and on what basis. Verification standards, however technically specified, involve judgments about what counts as evidence and who is qualified to assess it. Each of these interventions requires a legitimating basis that the current epistemic environment makes difficult to establish.9

In conditions of trust fragmentation, repair interventions are available to be read as power consolidation rather than genuine epistemic improvement. A government that mandates content moderation is, for a significant portion of the population, suppressing legitimate speech. An institution designated as an authoritative arbiter is, for those who distrust it, having authority conferred by the very entities whose legitimacy is at stake. The repair intervention becomes evidence for the narrative that motivated distrust in the first place. The effort to restore epistemic conditions is itself processed through the distorted epistemic conditions it is trying to correct.10

Repair interventions in conditions of deep trust fragmentation face a legitimacy problem that is internal to the repair project rather than external to it. They can proceed, but they proceed in an environment where their own legitimacy is contested by the mechanisms that made repair necessary.

VIII. LOCAL SUCCESS AND DEMOCRATIC SUFFICIENCY

Each constraint examined in sections 3 through 7 is serious on its own. Together, they describe a repair environment in which progress in any one dimension tends to be offset by resistance in others. Incentive lock-in limits the effectiveness of platform reform. Repair asymmetry limits the effectiveness of verification investment. Trust fragmentation limits the effectiveness of institutional strengthening. The self-undermining problem means that repair efforts can actively worsen the conditions they are trying to correct. These are not parallel failures. They interact: failure in one undermines progress in others, and the repair project requires simultaneous progress across all of them.

Domain-specific improvements have been documented. Certain interventions reduce the spread of specific false claims. Some institutional reforms have improved the reliability of particular verification processes. The question the repair project faces is whether improvement can occur at the scale and simultaneity required for democratic legitimacy. Local success and democratic sufficiency are not the same standard, and the repair literature has been insufficiently clear about which standard its evidence actually supports.

The probability of sufficient simultaneous progress across independent systems such as platforms, institutions, individual cognition, and political will is not simply the product of the probability of progress in each domain. It is lower, because the domains interact in the ways described above. A repair effort that succeeds in improving platform incentives without restoring institutional trust will find its verification outputs dismissed. One that restores institutional trust without addressing cognitive limits will find that trust exploited to reinforce divergence instead of correct it. The interdependence that makes repair coherent as a project is also what makes its full realization deeply constrained.

IX. WHEN THE PROBLEM OUTRUNS THE FRAMEWORK

The repair project operates within democratic theory. It takes the normative framework as given and asks what empirical conditions must be restored for that framework to function. That is the right question if the conditions are restorable. The analysis above suggests that they may not be, at least not by the mechanisms the repair project identifies, operating under the institutional and political constraints that actually obtain.

If that assessment is correct, the gap between what democratic legitimacy requires and what the current information environment can reliably produce is one that further institutional design within the existing framework cannot close. It is a problem about the relationship between the framework and the conditions under which it was developed, conditions that no longer obtain and may not be recoverable.

This does not establish that democratic legitimacy is impossible or that the democratic framework should be abandoned. The framework may require revision to remain coherent under conditions its architects did not anticipate and that its defenders have been slow to confront. The question shifts from how to restore the conditions democratic legitimacy requires to whether political structures must be revised to align with the epistemic conditions that actually obtain. That is the subject of working paper 4.

REFERENCES

Balkin, Jack M., 2018, “Free Speech in the Algorithmic Society: Big Data, Private Governance, and the Future of Public Discourse,” UC Davis Law Review, 51(3): 1149-1210.

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

Goldstein, Josh A., Girish Sastry, Micah Musser, Renee DiResta, Matthew Gentzel, and Katerina Sedova, 2023, “Generative Language Models and Automated Influence Operations: Emerging Threats and Potential Mitigations,” arXiv:2301.04246.

Kahan, Dan M., 2013, “Ideology, Motivated Reasoning, and Cognitive Reflection,” Judgment and Decision Making, 8(4): 407-424.

Kahan, Dan M., Ellen Peters, Maggie Wittlin, Paul Slovic, Lisa Larrimore Ouellette, Donald Braman, and Gregory Mandel, 2012, “The Polarizing Impact of Science Literacy and Numeracy on Perceived Climate Change Risks,” Nature Climate Change, 2(10): 732-735.

Kaye, David, 2019, Speech Police: The Global Struggle to Govern the Internet, New York: Columbia Global Reports.

McGrew, Sarah, Mark Smith, Joel Breakstone, Teresa Ortega, and Sam Wineburg, 2018, “Can Students Evaluate Online Sources? Learning from Assessments of Civic Online Reasoning,” Theory and Research in Social Education, 47(2): 165-193.

Napoli, Philip M., 2019, Social Media and the Public Interest: Media Regulation in the Disinformation Age, New York: Columbia University Press.

Nyhan, Brendan and Jason Reifler, 2010, “When Corrections Fail: The Persistence of Political Misperceptions,” Political Behavior, 32(2): 303-330.

Oreskes, Naomi, 2019, Why Trust Science?, Princeton: Princeton University Press.

Oreskes, Naomi and Erik M. Conway, 2010, Merchants of Doubt, New York: Bloomsbury Press.

Parker, Geoffrey G., Marshall W. Van Alstyne, and Sangeet Paul Choudary, 2016, Platform Revolution, New York: W. W. Norton.

Pennycook, Gordon and David G. Rand, 2019, “Lazy, Not Biased: Susceptibility to Partisan Fake News Is Better Explained by Lack of Reasoning Than Motivated Reasoning,” Cognition, 188: 39-50.

Peter, Fabienne, 2009, Democratic Legitimacy, New York: Routledge.

Wu, Tim, 2016, The Attention Merchants, New York: Knopf.

Zellers, Rowan, Ari Holtzman, Hannah Rashkin, Yonatan Bisk, Ali Farhadi, Franziska Roesner, and Yejin Choi, 2019, “Defending Against Neural Fake News,” Advances in Neural Information Processing Systems, 32.

FROM INFORMATION TO OPTIMIZATION

How the Modern Information Environment Selects for False Belief

The modern information environment selects for beliefs. What it selects for is not, in any straightforward sense, the truth.

By The Legitimacy Project | April 2026 | Working paper

I. NOT MISINFORMATION. SELECTION PRESSURE.

The standard account of epistemic deterioration focuses on false content: fabricated stories and coordinated deception. That account is incomplete in a way that matters. False information has always existed. What has changed is the structure of the environment in which it competes.

The first working paper in this series identified three vectors of epistemic deterioration, among them incentive misalignment, which this paper examines in depth. The argument here is that the information environment is now structured to systematically advantage false and epistemically degraded content over accurate content, independent of any actor’s intentions. The mechanism is optimization, and it operates at scale.

I proceed as follows. In section 2, I describe the incentive structure of contemporary information platforms and explain why it generates selection pressure against epistemic quality. In section 3, I examine how scale transforms what was a local and episodic problem into a global and continuous one. In section 4, I return to the cost structure inversion introduced in working paper 1 and unpack its mechanics. In section 5, I describe the emergent outcome of these combined forces. In section 6, I explain why the standard corrective responses fail. In section 7, I extend the argument to knowledge-producing institutions. In section 8, I state the structural constraint that follows for any political system operating in this environment.

II. WHAT PLATFORMS OPTIMIZE FOR

Contemporary information platforms are optimization systems. They are designed to maximize engagement, understood as the time users spend on platform and the probability they return. Engagement is what platforms sell to advertisers, and it is the metric around which their technical and organizational structures are oriented.1

Engagement and epistemic quality are not the same objective, and the divergence between them is a predictable consequence of optimizing for a behavioral signal that is easier to measure than truth. Content optimized for outrage spreads faster than content optimized for accuracy. Alarming claims outperform measured ones. The platform prefers high-engagement content to low-engagement content, and the correlation between high engagement and epistemic degradation is systematic.2

The result is epistemic selection pressure, a structural condition in which beliefs compete for circulation under platform conditions, and the ones that survive and spread are those that maximize engagement rather than those that accurately represent the world. The selection mechanism is indifferent to truth value in the same way that market competition is indifferent to the welfare of firms it eliminates. What survives is what is fit for the environment. The environment rewards engagement. Accuracy is not part of the fitness function.3

It is worth being precise about what this selection pressure acts on. It acts on the behavioral responses that claims elicit, rather than on their truth value directly. A false claim that triggers outrage spreads. A true claim that is complicated and emotionally flat does not. Over time, the claims that populate a shared information environment are a sample selected for the behavioral responses they produce in an audience operating under conditions of limited attention and high emotional reactivity. That is the population of beliefs from which democratic citizens are drawing their picture of the world.

III. WHY THIS TIME IS DIFFERENT

Incentive misalignment between truth and profitable content is not new. Yellow journalism, partisan presses, and sensationalist broadcasting all produced versions of this dynamic. What distinguishes the present moment is the scale and continuity at which it operates.4

Historical misinformation was episodic. A false story ran in a newspaper, circulated for a time, and eventually competed with corrections in the same or rival publications. The information environment had natural friction: print runs were finite, distribution was slow, and the number of channels was limited. Epistemic disruptions occurred against a background of relative informational stability.

The contemporary information environment operates under conditions of always-on epistemic competition. Content circulates globally and instantaneously. The number of channels is effectively unbounded. There is no off-cycle, no closing edition, no natural pause in which corrections can catch up with fabrications. Citizens are continuously immersed in an environment in which epistemically degraded content circulates alongside accurate content under conditions that systematically favor the former. The background of informational stability that made episodic disruption manageable no longer exists.5

Scale also changes the population dynamics of belief. When misinformation was episodic and local, its effects were bounded. A false claim might shape the beliefs of a community for a period, but it competed with the lived experience of community members, with local institutions, and with other channels of information not subject to the same selection pressures. At global scale and continuous operation, no such countervailing anchors exist for many of the beliefs that matter most for democratic governance. Citizens forming views about climate policy, electoral integrity, or public health are doing so in an environment where selection pressure operates uniformly and without interruption.

IV. VERIFICATION DOES NOT SCALE

Working paper 1 introduced the observation that the cost structure of the information environment has inverted, so that fabrication has become cheap and verification has become expensive. The mechanics of that inversion are more specific than the general observation suggests.

The cost of producing persuasive false content has declined along several dimensions simultaneously. Large language models generate coherent, authoritative-sounding text at negligible marginal cost. Image synthesis produces visually convincing fabrications without the skill or equipment that photographic manipulation once required. Voice cloning and video synthesis extend the same economics to audio and visual media. The production cost of a convincing false claim, which once required journalistic or technical skill, now approaches zero for anyone with access to widely available tools.6

Verification has not undergone the same cost reduction. Assessing whether a claim is true still requires domain expertise, access to primary sources, and the capacity to coordinate across institutions. Fact-checking organizations require trained staff. Scientific replication requires equipment and sustained time. The labor intensity of verification is a structural feature of what the process requires, one that no organizational improvement can eliminate.7

The result is a verification asymmetry, meaning that fabrication scales at a rate that verification cannot match. A single person with access to contemporary AI tools can produce false content faster than any institution can assess and correct it. At the level of an ecosystem, the ratio of fabrication to verification capacity is not stable. It widens as fabrication tools improve and as the volume of content requiring verification grows.8

There is a further dimension to this asymmetry that Chesney and Citron have called the liar’s dividend, referring to how the mere existence of convincing synthetic media undermines the epistemic value of genuine evidence. When fabricated video is indistinguishable from real video, real video loses its evidential force. The production of false content degrades the epistemic value of true content, so the verification burden increases even for claims that are accurate.9

V. NON-CONVERGENT EPISTEMIC SYSTEMS

Selection pressure, always-on competition, and verification asymmetry compound. Their combined effect is the emergence of non-convergent epistemic systems, meaning populations whose belief-forming processes operate under conditions that provide no stable mechanism for resolving disputes or establishing a shared factual baseline.

In a functional epistemic environment, disagreement is resolvable in principle. Citizens may hold different views, but they share enough of a common informational substrate that evidence and institutional adjudication can narrow disputes over time. This is the picture that democratic theory assumes. Citizen disagreement is structured by a common reality to which appeal is possible.10

Non-convergent epistemic systems do not have this property. When citizens draw their beliefs from information ecosystems structured by different selection pressures, operating on different factual priors, with no shared institutional authority capable of adjudicating between them, disagreement does not respond to evidence in the way democratic deliberation requires. Citizens are rational with respect to the information environments they actually inhabit, and those environments are structured to produce divergence rather than convergence.11

This is the mechanism that directly undermines the epistemic preconditions identified in working paper 1. Democratic legitimacy presupposes that citizens share enough of a common informational world to deliberate in good faith. Non-convergent epistemic systems make that presupposition false because the environment in which they form beliefs is structured to prevent it.

VI. WHY THE STANDARD RESPONSES DO NOT WORK

The standard response to misinformation is correction by producing accurate information, labeling false content, and deploying expert authority. There is nothing wrong with these interventions in principle. The problem is that they are responses to a content problem, and what is being described here is a structural one. Correcting individual false claims does not alter the selection pressure that produced them.

Several mechanisms explain why correction is systematically insufficient. Corrections spread more slowly than the content they correct. The same engagement dynamics that amplify false claims work against corrections, which tend to be less emotionally activating and more qualified. Research on the persistence of misinformation suggests that corrections often fail to fully displace false beliefs even when they reach the same audience, and they frequently do not reach the same audience.12

Expert intervention faces its own structural problem. In an environment where epistemic authority is fragmented and contested, the credibility of expert correction is itself a variable rather than a stable resource. A correction issued by a scientific body, a fact-checking organization, or a government agency does not land as authoritative for audiences whose information ecosystems have systematically cultivated distrust of those institutions. The correction may be accurate. It is not, for that audience, epistemically compelling.13

More information does not solve an optimization problem. In most contemporary democracies, accurate information is widely available. The problem is that the environment in which accurate and inaccurate information compete is structured to disadvantage accuracy. Adding more accurate content to that environment does not change the selection pressure. It increases the volume of content competing under conditions that are already adverse to epistemic quality.

VII. WHEN THE ARBITERS ARE ALSO SUBJECT TO SELECTION

The argument so far has focused on how epistemic selection pressure operates on citizen belief formation. The argument extends further. The institutions tasked with producing authoritative knowledge are themselves operating within the same information environment and are subject to versions of the same pressures.

Scientific institutions face replication failures, publication incentives that reward novelty over accuracy, and funding structures that create at least the appearance of conflicts of interest. These are not new problems. What is new is that the mechanisms for publicizing and amplifying these failures operate at the same scale and under the same selection pressures as the broader information environment. A retracted study generates as much circulation as its correction. The impression of expert disagreement is more engaging than the reality of expert consensus, and the information environment systematically produces the former at the expense of the latter.14

Legal and regulatory institutions face related dynamics. Courts produce authoritative adjudications of contested facts, but only within their jurisdictional scope and procedural constraints. When the contested facts that matter for democratic governance circulate outside those constraints, and when the authority of judicial findings is itself contestable in the broader information environment, the epistemic function that legal institutions serve is degraded. The finding exists, but its authority does not automatically transfer to the populations whose beliefs it was meant to settle.15

The press faces the most direct version of the institutional selection problem. Journalism that hedges and qualifies competes poorly in an environment that rewards neither. The same engagement dynamics that affect platform content affect the economic viability of news organizations. The environment in which journalism operates creates systematic pressure toward epistemic degradation. The institutions with the greatest capacity to resist that pressure are also the ones facing the most severe economic stress.16

When the institutions whose function is to arbitrate factual disputes are themselves subject to selection pressure, the epistemic infrastructure loses what working paper 1 called its second tier of support. Citizens form beliefs in an environment structured against epistemic quality. When they look to institutions for authoritative adjudication, they find institutions whose authority is contested and whose outputs are filtered through the same selection mechanisms that distort citizen belief formation in the first place. The system has no self-correcting tier.

VIII. A CONDITION, NOT A DISRUPTION

The picture assembled across sections 2 through 7 is of a system. Epistemic selection pressure, always-on competition, verification asymmetry, non-convergent epistemic systems, the failure of correction, and institutional spillover are interacting components of a single structural condition that emerged from the optimization logic of contemporary information infrastructure.

A disruption has a before and an after. It is an event against a background of relative normality, and responses to it are oriented toward restoration. The condition described here is a structural transformation of the epistemic environment itself. The question of what epistemic conditions prevailed before this transformation is historically interesting but not practically relevant. The environment that democratic theory assumed is not the environment that presently obtains, and there is no evidence that the forces producing the current environment are temporary or self-limiting.

Any political system that depends on reliable belief formation among its citizens must operate within this environment. That includes democratic systems, but it is not limited to them. The constraint is on governance as such, insofar as governance requires a population capable of being informed about and responsive to the conditions it is governing. What the constraint means specifically for democratic legitimacy was the subject of working paper 1. What it means for the available options of repair and revision will be the subject of working papers 3 and 4.

The present paper has established that the epistemic deterioration identified in working paper 1 is the product of an optimization logic operating at scale, rather than the contingent work of bad actors against an otherwise stable epistemic background, and that it is uncorrectable by interventions that leave that logic intact. The question has shifted from whether epistemic conditions are degraded to what forms of governance can survive them.

REFERENCES

Benkler, Yochai, Robert Faris, and Hal Roberts, 2018, Network Propaganda: Manipulation, Disinformation, and Radicalization in American Politics, Oxford: Oxford University Press.

Chesney, Robert and Danielle Keats Citron, 2019, “Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security,” California Law Review, 107(6): 1753-1820.

Citron, Danielle Keats, 2014, Hate Crimes in Cyberspace, Cambridge, MA: Harvard University Press.

Cohen, Joshua, 1989, “Deliberation and Democratic Legitimacy,” in Alan Hamlin and Philip Pettit (eds.), The Good Polity, Oxford: Basil Blackwell, 17-34.

Goldstein, Josh A., Girish Sastry, Micah Musser, Rene DiResta, Matthew Gentzel, and Katerina Sedova, 2023, “Generative Language Models and Automated Influence Operations: Emerging Threats and Potential Mitigations,” arXiv:2301.04246.

Goodin, Robert E. and Kai Spiekermann, 2019, An Epistemic Theory of Democracy, Oxford: Oxford University Press.

Graves, Lucas, 2016, Deciding What’s True: The Rise of Political Fact-Checking in American Journalism, New York: Columbia University Press.

Habermas, Jurgen, 1992 [1996], Between Facts and Norms, William Rehg (trans.), Cambridge, MA: MIT Press.

Kahan, Dan M., 2013, “Ideology, Motivated Reasoning, and Cognitive Reflection,” Judgment and Decision Making, 8(4): 407-424.

Lewandowsky, Stephan, Ullrich K. H. Ecker, Colleen M. Seifert, Norbert Schwarz, and John Cook, 2012, “Misinformation and Its Correction: Continued Influence and Successful Debiasing,” Psychological Science in the Public Interest, 13(3): 106-131.

McChesney, Robert W. and John Nichols, 2010, The Death and Life of American Journalism, Philadelphia: Nation Books.

Oreskes, Naomi, 2019, Why Trust Science?, Princeton: Princeton University Press.

Oreskes, Naomi and Erik M. Conway, 2010, Merchants of Doubt, New York: Bloomsbury Press.

Vosoughi, Soroush, Deb Roy, and Sinan Aral, 2018, “The Spread of True and False News Online,” Science, 359(6380): 1146-1151.

Wardle, Claire and Hossein Derakhshan, 2017, Information Disorder: Toward an Interdisciplinary Framework for Research and Policy Making, Strasbourg: Council of Europe.

Wu, Tim, 2016, The Attention Merchants: The Epic Scramble to Get Inside Our Heads, New York: Knopf.

Zuboff, Shoshana, 2019, The Age of Surveillance Capitalism, New York: PublicAffairs.

THE EPISTEMIC PRECONDITIONS OF DEMOCRATIC LEGITIMACY

A citizenry capable of forming reliable beliefs about the world it is governing is among the foundational presuppositions of democratic theory, and among the least examined.

By The Legitimacy Project | April 2026 | Working paper

I. WHAT DEMOCRATIC THEORY PRESUPPOSES

Democratic theory rests on a substrate it rarely names. Beneath the machinery of elections, representation, and rights lies a set of epistemic presuppositions, namely that citizens can access information relevant to political life, that this information is broadly shared, and that deliberative institutions can process disagreement into actionable consensus, and these presuppositions are foundational features of democratic legitimacy.1

The classical formulations understood this, if incompletely. Dewey’s account of democratic community required a public capable of perceiving and communicating its own problems.2 Habermas’s communicative rationality required conditions under which the force of the better argument could at least in principle prevail.3 Even proceduralist accounts, which locate legitimacy in the fairness of decision rules rather than in the quality of deliberation, implicitly presuppose that participants are reasoning about something real.4

The shared informational environment is, in other words, the medium through which democratic authorization becomes meaningful. A system in which citizens vote freely but cannot distinguish fact from fabrication calls into question whether the votes it produces carry the authorizing weight that democratic theory assigns to them.5

I proceed as follows. In section 2, I identify three vectors along which the background conditions of democracy are currently under stress. In section 3, I develop the tension that follows if those conditions fail. In section 4, I distinguish two possible responses. In section 5, I introduce the Legitimacy Project as an ongoing investigation into which response, if either, remains viable.

II. THREE VECTORS OF EPISTEMIC DETERIORATION

The conditions democracy presupposes are under systematic stress. This is not a claim about moral failure or political polarization in the colloquial sense. It is a structural claim about the information environment in which contemporary citizens form beliefs. Three vectors of deterioration are worth distinguishing.

The first is fragmentation. The collapse of shared informational anchors, including broadcast media, common educational curricula, and regional civic institutions, has been well-documented, but its epistemic consequence is less well understood.6 Fragmentation is more than mere disagreement between citizens. It means that citizens increasingly inhabit non-overlapping factual worlds, such that disagreement no longer primarily takes the form of competing evaluations of shared evidence. It takes the form of competing realities.7

The second is incentive misalignment. The dominant platforms for public discourse are optimized for engagement, and engagement and accuracy are not the same objective. The divergence is not incidental8 , because emotionally activating content spreads faster and generates more revenue than calibrated and qualified claims. The result is a systematic selection pressure against the capacity for nuance and the willingness to revise that accurate belief formation requires.9

The third is scalable epistemic noise. Artificial intelligence now enables the production of persuasive but false content at negligible cost and effectively infinite scale.10 The significance of this is not simply that there is more misinformation. The cost structure of the information environment has inverted, so that fabrication has become cheap and verification has become expensive. This asymmetry has no obvious equilibrium.

Each of these vectors compounds the others. Fragmentation makes verification harder to coordinate. Incentive misalignment ensures that fabrications, once produced, travel further than corrections. Scalable noise raises the volume of what must be corrected to a level that existing institutions cannot process. Taken together, the three vectors constitute a structural account of why the informational substrate of democracy is failing, a description of a system behaving as designed rather than a list of bad actors.

III. PROCEDURALLY INTACT, SUBSTANTIVELY HOLLOW

The question these developments pose isn’t whether elections are being stolen, or whether elites are subverting democratic procedure. The machinery may be functioning exactly as designed. The question is what that machinery is actually producing.

When citizens form preferences through systematically distorted epistemic processes, the preferences they express may be misaligned with the function that democratic theory takes them to serve. The vote is meant to aggregate genuine political will. But if that will is shaped by information architectures optimized for engagement rather than accuracy, by the exploitation of cognitive bias rather than genuine deliberation, then the output of democratic procedure may no longer track what legitimizes it.11

This is a structural question about the relationship between democratic procedure and democratic substance and requires no conspiracy theory.12 A system can be formally democratic, while being substantively degraded in its capacity to produce legitimate collective decisions.

The problem extends further than citizen belief formation. The institutions tasked with producing authoritative knowledge, courts, regulatory agencies, and the press, along with the scientific bodies that inform them, are themselves subject to the same incentive pressures and coordination failures that distort the broader information environment. When those institutions can no longer reliably arbitrate between competing factual claims, the epistemic infrastructure that democratic legitimacy depends on loses a second tier of support. We do not yet have a stable name for this compounded condition. It may be that the condition is genuinely novel, and that the concepts adequate to it have not yet been formed.

What we can say is this. Democratic legitimacy has traditionally been understood to require that the procedure is fair and that participants are at least minimally capable of reasoning about what they are deciding.13 If the second condition fails at scale, and if the institutions meant to shore it up are themselves compromised, the first condition alone is unlikely to be sufficient to generate legitimate outcomes. That is the tension this paper is tracking.

IV. TWO PATHS, NEITHER COMFORTABLE

If the background conditions of democratic legitimacy are failing, the response takes one of two forms. They are not mutually exclusive, but they are different in kind.

The first is repair, meaning the restoration of conditions under which democratic legitimacy can function. This might mean redesigning the platforms and institutions through which information flows, or developing interventions that raise the cost of large-scale manipulation. Repair operates within democratic theory. It accepts the normative framework and attempts to restore the empirical conditions that framework presupposes.14 It is not yet clear, however, that repair is achievable at the scale the problem requires, or that the institutional capacity to coordinate such repair is itself intact.

The second is revision, meaning the reconsideration of political structures themselves in light of the epistemic conditions that actually obtain. This path is harder to discuss without invoking either technocratic elitism or authoritarian temptation.15 But if the conditions that democratic legitimacy requires cannot be restored, then the theory of legitimacy will have to be revised, not abandoned, but brought into alignment with what is empirically possible. The current posture, assuming democratic legitimacy while the conditions for it erode, is itself a choice, and a consequential one.

Maybe the repair path is viable and revision unnecessary. Maybe repair is insufficient and revision unavoidable. Maybe neither path is adequate to the scale of the problem. These are not rhetorical possibilities. They are the live options that the present situation puts on the table, and they cannot be responsibly evaluated without sustained inquiry into what the present situation actually is.16

V. AN ONGOING INVESTIGATION

The Legitimacy Project is an investigation into these questions, conducted without prior commitment to the answers.

The guiding assumption is that political ideology is a variable system constrained by empirical reality. The question is not which ideology is correct. The question is what forms of governance remain possible and justifiable under actual cognitive, technological, and social conditions. That question may have uncomfortable answers.

It may turn out that some governance forms presuppose epistemic conditions that are no longer achievable. It may turn out that the repair of those conditions requires interventions that themselves carry democratic costs. It may turn out that we are navigating a transition whose endpoint is not yet legible. We do not know. What we do know is that the question is not being adequately asked, not by political theory, which tends to assume functioning epistemic institutions,17 not by political science, which tends to treat legitimacy as an empirical variable rather than a normative concept under stress, and not by public discourse, which tends to litigate symptoms rather than diagnose the condition.

This paper is a diagnosis which is, at present, incomplete. Future work will address the mechanisms by which epistemic authority degrades across both public discourse and knowledge-producing institutions, historical precedents for the conditions described here, and the available options for repair or revision. The Legitimacy Project is an ongoing investigation into whether modern political systems, democratic ones especially, remain viable under contemporary epistemic conditions. That investigation begins here.

REFERENCES

Achen, Christopher H. and Larry M. Bartels, 2016, Democracy for Realists: Why Elections Do Not Produce Responsive Government, Princeton: Princeton University Press.

Anderson, Elizabeth, 2006, “The Epistemology of Democracy,” Episteme, 3(1-2): 8-22.

Bartels, Larry M., 2002, “Beyond the Running Tally: Partisan Bias in Political Perceptions,” Political Behavior, 24(2): 117-150.

Benkler, Yochai, Robert Faris, and Hal Roberts, 2018, Network Propaganda: Manipulation, Disinformation, and Radicalization in American Politics, Oxford: Oxford University Press.

Brennan, Jason, 2016, Against Democracy, Princeton: Princeton University Press.

Chesney, Robert and Danielle Keats Citron, 2019, “Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security,” California Law Review, 107(6): 1753-1820.

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

Cohen, Joshua, 1989, “Deliberation and Democratic Legitimacy,” in Alan Hamlin and Philip Pettit (eds.), The Good Polity, Oxford: Basil Blackwell, 17-34.

Dewey, John, 1927 [2012], The Public and Its Problems: An Essay in Political Inquiry, Melvin L. Rogers (ed.), University Park, PA: Pennsylvania State University Press.

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

Goodin, Robert E. and Kai Spiekermann, 2019, An Epistemic Theory of Democracy, Oxford: Oxford University Press.

Habermas, Jurgen, 1992 [1996], Between Facts and Norms: Contributions to a Discourse Theory of Law and Democracy, William Rehg (trans.), Cambridge, MA: MIT Press.

Kahan, Dan M., 2013, “Ideology, Motivated Reasoning, and Cognitive Reflection,” Judgment and Decision Making, 8(4): 407-424.

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

Peter, Fabienne, 2009, Democratic Legitimacy, New York: Routledge.

Wardle, Claire and Hossein Derakhshan, 2017, Information Disorder: Toward an Interdisciplinary Framework for Research and Policy Making, Strasbourg: Council of Europe.