After The Myth of the Rational Voter: Can Democracy Be Made Less Blind?
In my review of Bryan Caplan’s The Myth of the Rational Voter, the book’s most unsettling claim was that democracies often fail not because voters are ignored, but because they are obeyed. Bad policy is not always imposed upon the public by cynical elites. Sometimes voters ask for it, reward it and prefer the story it tells about the world.
That leaves a harder question for anyone interested in democratic reform. If individual voters bear almost none of the cost of holding politically satisfying beliefs, what kinds of institutions could make collective error less cheap without treating citizens as defective material to be managed by their betters?
One answer is to reduce popular power. Voters do not understand economics, incentives, institutions or trade-offs, so more authority should pass to economists, judges, civil servants, central banks, markets or some other body less exposed to mass opinion. The temptation begins from a real weakness and then mistakes dispossession for correction.
Caplan’s diagnosis does not require that conclusion. Democracy can retain equal political standing while becoming less forgiving of vague promises, invisible costs and policies that survive long after their justification has weakened. The relevant target is not participation itself, but the conditions under which political preferences are formed, translated into policy and protected from later evidence.
The Wrong Lesson Is Less Democracy
The most poisonous application of Caplan’s argument is that irrational voters should simply have less power. Systems that make political standing depend upon property, literacy or approved competence do not operate in an institutional vacuum. The people defining competence acquire a predictable ability to discover that political judgment is concentrated among people rather like themselves.
Even a knowledge test designed without discriminatory intent would collide with democracy’s function as more than a method for selecting efficient policies. Democracy distributes political standing, prevents permanent rule by a self-authorising class and gives electoral losers some reason to accept decisions made by their opponents. A system that tells citizens they are too foolish to count may occasionally make a technically better decision while steadily weakening the legitimacy required to carry it out.
Expertise remains necessary. Most citizens cannot personally estimate the fiscal effect of a tax reform, assess competing epidemiological models or understand every consequence of a trade agreement. The mistake lies in treating expert analysis as a substitute for political judgment rather than an input into it. Estimates can clarify likely consequences; they cannot decide how liberty, equality, security, disruption and distribution should be valued.
Experts and institutions also have their own failure modes. Bureaucracies protect procedures and jurisdictions. Professions develop shared assumptions. Models simplify what they cannot easily measure. Independent agencies may be more insulated from electoral mood while remaining vulnerable to institutional caution, status incentives or narrow mandates. The choice is not between an irrational public and a frictionless machine of expertise.
Democracy is already mediated through parties, ballots, campaigns, news organisations, social platforms, courts, civil services, lobby groups, electoral systems and constitutional rules. The useful question is not whether mediation should exist, but what it makes visible and whose pressure it mistakes for public judgment. As explored in “Democracy’s Signal Problem: Lobbying, Outrage, and False Responsiveness”, organised interests can make private advantage resemble technical necessity, while online intensity can make a highly activated minority resemble the public as a whole. Some institutions expose costs and contradictions; others reward whichever actors possess the best access or loudest amplification.
This is where epistemic scaffolding becomes useful. Citizens should not have to become professional economists or legislative analysts before voting. They should have access to institutions that clarify what an office controls, what a proposal is expected to cost, which assumptions drive an estimate, where specialists disagree and what evidence would later show that a policy has failed.
Information will not abolish motivated reasoning. A briefing can be dismissed as captured, a cost estimate attacked as partisan and an expert recoded as an enemy. Political beliefs often survive because they distribute blame, reinforce identity or make an ambiguous world feel morally legible. The purpose of better scaffolding is more modest: to make evasion require greater effort and to ensure that official decisions leave a record against which later claims can be tested.
Make Political Promises Legible
Campaign promises are often sold as moods. Protect jobs. Lower prices. Support families. Build homes. Control migration. Defend national industry. Each phrase may point towards a legitimate political objective, but it invites agreement before identifying the mechanism, cost or conflict hidden inside it.
Many policies derive their attraction from the moral story attached to them. Protectionism becomes loyalty. Price controls become compassion. Housing restrictions become community preservation. Subsidies become strategy. Labour-market rigidity becomes dignity. The story may contain part of the truth, yet still conceal who pays and which alternatives are being rejected.
Major proposals should therefore arrive with a standard account of their trade-offs. Who benefits directly? Who pays directly? Which indirect effects are expected? What uncertainty remains? Which assumptions make the estimate rise or fall? What evidence would justify changing or ending the policy later?
Independent fiscal institutions can answer part of that problem. The Australian Parliamentary Budget Office’s 2025 Election Commitments Report, for example, presented individual and combined budget effects for hundreds of qualifying commitments made by major parties and one participating independent. It did not determine which platform was desirable. It made the fiscal claims more comparable and documented the assumptions behind them.
Such exercises have limits that should remain visible. A budget costing can estimate direct effects on revenue, expenditure and debt under stated assumptions. It may not capture every behavioural response, distributional consequence, implementation failure or social benefit. A proposal can be accurately costed and still be a bad idea; it can also impose a measurable fiscal cost that voters reasonably decide is worth paying.
The democratic value lies in preventing fiscal consequences from disappearing completely behind campaign language. A tariff should identify expected effects on protected producers, consumers and downstream businesses. A housing policy should say where homes may be built and which constraints must change. A pension promise should disclose the contribution, tax or borrowing assumptions upon which it depends.
This is not rule by accountants. It is political labelling. Citizens are routinely given disclosures before buying medicine, financial products or insurance. Policies capable of reshaping public finances should not be sold with less information than a household appliance.
Give Difficult Choices More Time
Some democratic failures arise less from missing information than from compressed decision-making. Referendums, constitutional changes and major reforms are debated through slogans, campaign deadlines and identity signals even when the underlying questions have occupied specialists for decades. Speed rewards the first emotionally satisfying account of the issue and leaves little time for contradictions to become uncomfortable.
Representative deliberative processes offer a limited way to slow such choices down. The OECD’s database of deliberative processes recorded hundreds of cases in which randomly selected citizens received information, considered competing perspectives and produced recommendations for public authorities. Most recent examples occurred at local or regional level, where the subject is often more concrete and the relationship between recommendation and consequence easier to see.
Deliberative Polling uses a related but distinct method. A representative sample is surveyed, given balanced briefing material, brought together to question specialists and discuss the issue, and then surveyed again. Its purpose is not to discover an authentic popular will hidden beneath ordinary polling. It examines how opinion changes after participants have spent time with information and disagreement.
Neither method produces automatic wisdom. Briefing materials can be framed badly, experts selected unevenly and questions defined in ways that favour one interpretation. Participants remain subject to social pressure and motivated reasoning. A temporary mini-public also lacks the democratic authority of an elected legislature unless the wider political system has explicitly granted it a defined role.
Their most defensible function is advisory. Before a referendum or major reform, a representative group could study the issue and publish a report explaining what members learned, where they remained divided, which claims survived scrutiny and which apparent solutions concealed difficult trade-offs. The assembly would not decide for everyone else. It would give the wider public a record produced by citizens who had been given time unavailable in an ordinary campaign.
This matters because public opinion is rarely one coherent instruction. Citizens often want lower taxes and better services, more housing and less construction, cheap energy and rapid decarbonisation, secure borders and abundant labour. Deliberation does not remove those contradictions. It makes them harder to present as though no conflict exists.
Make Policies Return for Judgment
Democratic systems are generally better at creating policies than removing them. A subsidy, tariff, emergency provision or regulatory privilege may begin as a response to a real problem. Beneficiaries then organise around it, administrators develop routines and political identities attach themselves to its survival. The original justification can weaken while the policy acquires new reasons not to be disturbed.
This is where the argument connects naturally to inadequate equilibria. An arrangement can remain stable without being desirable because the benefits of reform are diffuse while the costs of change are concentrated. Almost nobody would design the system from the beginning, yet enough actors can block the path out.
Better pre-enactment analysis is not enough. Policies also need mechanisms through which evidence can return after adoption. Reviews should ask whether the intended benefit appeared, what the programme actually cost, which groups gained or lost, what unintended effects emerged and whether the original problem still exists.
A sunset clause changes the default by making a provision expire after a fixed period unless renewed. That can be appropriate for emergency powers, pilots, temporary subsidies and measures adopted under severe uncertainty. It forces lawmakers to return to the question instead of allowing inertia to renew the policy silently.
Automatic expiry is too blunt for every law. Long-term infrastructure, core public institutions and rights frameworks require stability, and a mandatory renewal vote can itself become an opportunity for hostage-taking or disruption. In many cases, a statutory review clause—requiring evidence and parliamentary consideration without automatic expiry—will be more appropriate.
The broader principle is that adoption should not confer immortality. Temporary measures should normally end unless they still earn renewal. Permanent programmes should return for structured evaluation on a published timetable. Defenders should not be able to treat mere survival as proof of success.
Review also needs consequences. Reports that arrive late, measure only activity or disappear into departmental archives provide little discipline. The original objectives and evaluation criteria should be stated when the policy is adopted, allowing later governments and citizens to compare promise with performance rather than inventing a new justification after the old one fails.
AI as a Research Layer, Not an Arbiter
Artificial intelligence enters this reform agenda with two opposite possibilities. Generative systems can fabricate local facts, misstate the powers of an office, invent candidate positions and convert partisan claims into calm, apparently neutral prose. The Brennan Center’s work on AI and democracy documents the danger of cheaper deceptive content, impersonation and election disinformation. Fluency can give weak information an authority it has not earned.
AI can also lower the cost of basic civic research. Local ballots often contain unfamiliar offices and candidates about whom reliable information is scattered across official sites, party platforms, meeting records, news coverage and public filings. Most voters have little individual incentive to spend hours assembling that material when the chance that one ballot changes the outcome is tiny.
Scott Alexander’s “Use AI This Election” proposes this narrower use: not asking a model who should win, but using it to accelerate research that the voter would otherwise perform manually or abandon. What does the office control? What have the candidates said? Which endorsements or records are relevant? Where do the voter’s own priorities appear to point?
The useful civic system would not be neutral in the sense of possessing no assumptions. Selection and summary always require judgment. It should instead be transparent and contestable. It should cite the material used, identify the date of each source, separate official claims from independent evidence and state when available information is incomplete or contradictory.
It should also resist the ordinary conversational pressure to please the user. A voter who repeatedly asks for a justification of an existing preference should not be rewarded with an increasingly polished rationalisation. The system should be able to present the strongest evidence against the user’s current conclusion, distinguish factual disagreement from value disagreement and show how different priorities produce different choices.
Official records should receive priority where they answer the question: voting histories, party programmes, candidate statements, legal powers, budget documents and court decisions. Independent reporting and analysis remain necessary for context and verification, but the model should not quietly blend all sources into one synthetic voice.
No such assistant would make democracy rational. It could be manipulated, trusted too readily or dismissed whenever it threatens a cherished belief. Its contribution is narrower: lowering the cost of locating evidence and understanding the formal choice before the voter. That is useful without pretending that an information tool can settle political judgment.
More Democracy, Less Blindness
None of these reforms cures motivated reasoning. Independent costings can be attacked as partisan. Citizens’ assemblies can be framed or ignored. Reviews can become procedural rituals. AI research tools can hallucinate, flatter or be captured by the institutions providing them. Every layer of scaffolding creates another institution whose authority must be examined.
That is not a reason to abandon the effort. It is a reason to design for contestability. Methods should be published, assumptions exposed, dissent recorded and later performance compared with earlier claims. No institution should be allowed to define reality without showing its work.
The reform programme is therefore less dramatic than either technocracy or democratic romanticism. Make fiscal promises comparable. Give difficult decisions more time. Require temporary policies to seek renewal and permanent ones to face review. Use AI to locate and organise evidence without granting it political authority. Preserve the voter’s decision while improving the conditions under which it is made.
The public language matters. Reform should not begin by telling citizens that experts must protect policy from them. It should begin from the claim that citizens deserve better tools than slogans, party branding and campaign theatre. They deserve to know what promises cost, what institutions can actually do and whether yesterday’s solution produced the result used to justify it.
Caplan’s diagnosis should make democracy less innocent, not less democratic. Political equality does not require pretending that public judgment is automatically wise. It requires building institutions in which citizens can encounter evidence without surrendering the authority to decide what the evidence is worth.
The problem is not that voters are uniquely foolish. It is that people, organisations and governments all become more tolerant of error when someone else receives the bill. A less blind democracy would make that bill arrive sooner, display it more clearly and refuse to let it disappear merely because the policy has become familiar.
Comments
Post a Comment