AI in government is no longer a future policy issue. Public agencies already use algorithmic systems to sort applications, detect fraud, allocate resources, answer citizen questions and support internal decision-making. Some of these tools are simple automation. Others influence access to benefits, education, housing, policing, immigration or public health.
That makes AI rulemaking a democratic question, not only a technical one.
A fair AI system in government cannot be judged only by accuracy, efficiency or procurement cost. It must also answer civic questions: Who is affected? Who can challenge a decision? What values should guide the system when rights, budgets and risks collide? Which uses should be prohibited entirely? Deliberative democracy offers a practical way to answer those questions before systems become embedded in public life.
JustSocial’s manifesto argues that industrial-era institutions are still governing a technological society with outdated habits. Its broader call is for people to become a real governing force, not passive voters or consumers. Fair AI rules in government belong exactly there: citizens should not discover automated public power after it has already been bought, deployed and normalized.
Why AI rules in government need democratic legitimacy
Government AI differs from private AI in one crucial way: citizens often cannot opt out. If a company uses a recommendation system badly, people may have alternatives. If a public agency uses an automated risk score, eligibility model or chatbot that shapes access to state services, the citizen faces the state itself.
That is why AI governance cannot rest on internal ethics statements. Public institutions need clear rules for when AI may be used, when it must be avoided and how citizens can see, contest and correct its effects.
Several global frameworks already point in this direction. The NIST AI Risk Management Framework focuses on mapping, measuring and managing AI risk. UNESCO’s Recommendation on the Ethics of Artificial Intelligence calls for human rights, transparency and public participation. The European Union’s AI Act uses a risk-based model for regulating AI systems, including high-risk uses.
These frameworks are useful, but they do not replace democratic judgment. A national framework can say “high-risk systems require oversight,” but the public still needs to decide what meaningful oversight looks like in a welfare office, school district, immigration authority or city police department.
Deliberative democracy fills that gap by giving citizens time, balanced information and structured dialogue before public decisions are made.
What deliberative democracy adds to AI governance
Ordinary public consultation often favors the loudest, most organized or most technically fluent participants. AI policy can make that worse because the language is specialized, the systems are opaque and the risks are unevenly distributed.
Deliberative democracy changes the conditions of participation. Instead of asking people for quick opinions after reading a policy summary, it brings a representative group of citizens through a process of learning, questioning, discussion and recommendation.
A good deliberative process for AI rules would include:
- Balanced briefings from technologists, civil rights advocates, public servants, affected communities and independent researchers
- Plain-language explanations of how AI systems are used in public administration
- Time for citizens to question experts and challenge assumptions
- Facilitation rules that prevent domination by partisan groups, vendors or government insiders
- A public record showing how citizen recommendations influenced the final rules
This matters because AI governance is full of tradeoffs. Speed may conflict with due process. Fraud prevention may conflict with dignity and access. Predictive tools may promise efficiency while creating feedback loops against vulnerable groups. Citizens need to deliberate over these conflicts, not merely react to them.
For a deeper look at which democratic functions AI should never replace, JustSocial has also explored what should stay human in deliberative democracy and AI. The key distinction is simple: AI can help organize knowledge, but it cannot carry democratic responsibility.
Fair AI rules begin with better public questions
The quality of AI rules depends heavily on the questions government asks before adoption. A weak question sounds like: “Should this agency use AI to reduce processing time?” That framing invites a narrow efficiency answer.
A fairer question asks: “Under what conditions, if any, should this agency use AI in decisions that affect access to public services, and what rights must citizens have when the system is wrong?”
Question design is not a neutral administrative detail. It sets the boundaries of public imagination. If the question assumes deployment, citizens are pushed into discussing safeguards only. If the question allows refusal, redesign or delay, they can exercise real judgment.
| AI rulemaking issue | Narrow administrative question | Better deliberative question |
|---|---|---|
| Eligibility systems | Can AI process claims faster? | When should AI assist eligibility decisions, and what human review is required? |
| Public chatbots | Can chatbots reduce staff workload? | What information may a chatbot provide, and when must it transfer people to a human? |
| Fraud detection | Can models identify suspicious cases? | How do we prevent error, bias and intimidation while protecting public funds? |
| Predictive enforcement | Can AI help target inspections or policing? | Which predictive uses violate fairness, dignity or equal treatment even if technically effective? |
| Procurement | Which vendor offers the best tool? | What public values must any tool satisfy before government buys it? |
This is where civic participation becomes constitutional in spirit, even when the topic looks technical. Citizens are not writing code. They are setting the terms under which public power may use code.
JustSocial’s work on fair question design in deliberative democracy is especially relevant here because AI policy can be captured long before a vote, hearing or procurement process begins.
Discursive democracy and the public language of AI
Deliberation needs more than a well-run citizens’ panel. It also needs a wider public culture where arguments can be heard, tested and revised. That is the domain of discursive democracy.
Discursive democracy treats public reasoning as a core part of legitimacy. People do not only count preferences. They exchange reasons. They explain why a rule is fair, why a risk is unacceptable or why a benefit matters. In AI governance, this is essential because many systems are difficult to see. Their effects may appear as a denied application, a delayed service, a risk label or a recommendation accepted by a busy official.
A discursive approach forces government to speak in reasons the public can examine. “The model is accurate” is not enough. Accurate for whom? Compared with what baseline? With what appeal rights? Under what data limits? At what human cost?
This also connects to the manifesto’s argument that public institutions should use modern technology without treating people as passive data points. The goal is not anti-technology. It is public technology under public reason.
A practical model for deliberative AI rulemaking
Governments do not need to choose between expert regulation and public participation. Fair AI rules need both. Experts can explain technical risks, legal obligations and operational realities. Citizens can judge what risks are acceptable in a democratic society and what uses cross the line.
A practical model could work in five stages.
| Stage | Purpose | Democratic value |
|---|---|---|
| Public AI inventory | List current and proposed AI systems used by agencies | Visibility before judgment |
| Impact mapping | Identify who is affected, what rights are involved and what errors could cause harm | Fairness and accountability |
| Citizen deliberation | Convene representative panels to review evidence and recommend rules | Informed civic participation |
| Legislative or administrative adoption | Turn recommendations into binding rules, procurement standards or agency policies | Public authority and enforceability |
| Continuous review | Revisit systems as data, technology and social conditions change | Adaptation without secrecy |
The first stage is often neglected. Citizens cannot deliberate over systems they do not know exist. A public AI inventory should include the agency using the system, the purpose, the vendor if applicable, the affected population, the decision points influenced by AI and the appeal process.
The second stage should not be outsourced entirely to vendors. Vendor documentation may be useful, but government has a duty to examine rights, institutional incentives and social effects. AI impact assessments should be public by default, with narrow exceptions for genuine security risks.
The third stage is where deliberative democracy earns its place. A citizens’ assembly or panel can hear from multiple sides, ask questions and produce recommendations that are more reflective than polling and more representative than open-comment processes.
The fourth stage is the test of seriousness. If citizen recommendations disappear into a PDF, the process becomes civic theater. Governments should publish a response explaining which recommendations were adopted, rejected or modified and why.
The fifth stage is necessary because AI systems change. Models drift. Data changes. Agencies repurpose tools. A rule that was adequate in 2026 may fail by 2028 if the system expands into new decisions.
What fair AI rules should cover
Deliberative processes should not be vague town halls about “AI and the future.” They should produce usable public standards. The exact rules will differ by country, city and agency, but several categories deserve public judgment.
First, governments need use limits. Some AI applications may be acceptable for translation, document search or service navigation but unacceptable for final decisions about rights, liberty or essential benefits. The line should not be drawn by vendors or agency convenience alone.
Second, citizens need notice. If AI materially influences a public decision, the affected person should know. Hidden automation undermines trust because people cannot contest what they cannot see.
Third, there must be human responsibility. This does not mean a human rubber-stamps the model. It means a named public authority remains accountable for the decision, understands the system’s limits and can override it.
Fourth, appeal rights must be real. A citizen should be able to challenge an AI-influenced decision without needing a data science degree. The process should be accessible, timely and capable of correcting both individual errors and systemic problems.
Fifth, public agencies need audit duties. Models used in government should be tested for disparate impact, error rates, security risks, data quality and unintended consequences. Where independent audit is needed, the results should be summarized publicly.
Sixth, procurement rules must change. Governments should not buy systems whose core logic, data practices or performance claims cannot be meaningfully examined. Trade secrecy cannot become a shield against democratic accountability.
These categories match the basic concern behind JustSocial’s political movement: public systems should be designed around citizen empowerment, not bureaucratic inertia.
Why a political movement is needed, not only policy papers
AI governance will not become democratic by default. The path of least resistance is procurement first, public explanation later. Agencies face pressure to cut costs. Vendors market efficiency. Politicians often prefer visible modernization over slow institutional redesign.
That is why a political movement matters. Fair AI rules require organized public pressure, not only expert reports. Citizens, technologists, educators, lawyers, public servants and community leaders need to insist that government AI is governed in the open.
The JustSocial manifesto speaks of moving beyond institutions that reduce people to voters, taxpayers and consumers. AI could either deepen that reduction or help reverse it. If public AI is built without meaningful civic participation, citizens become subjects of automated administration. If it is governed through deliberative democracy, citizens become co-authors of the rules.
This does not mean every AI configuration should go to a public vote. It means the core standards, boundaries and accountability structures should be shaped in public, by people who have been given the time and information to deliberate well.
JustSocial’s guide to safe uses and red lines for AI in democracy develops a similar distinction: AI can assist democratic systems, but the public must define the limits.
The role of academia and independent knowledge
The manifesto’s proposal to make academia a stronger independent branch of public life is especially useful for AI rulemaking. Citizens need expert knowledge, but they also need protection from expert capture.
Universities, public research institutes and independent auditors can help deliberative bodies understand model risk, statistical bias, administrative law, political theory and human rights. Their role should be advisory rather than ruling. In a democratic process, knowledge informs judgment. It does not replace it.
A balanced deliberative process should include disagreement among experts. AI vendors should be heard, but so should civil society groups, frontline public workers, affected communities and scholars who study failure cases. Citizens should see where consensus exists and where uncertainty remains.
This is also how deliberative democracy avoids both technocracy and populist simplification. It does not say “let experts decide,” and it does not say “ignore expertise.” It creates a structured meeting point between knowledge and public judgment.
From industrial bureaucracy to democratic technology
Many public institutions still operate as if technology is an internal administrative upgrade. A new platform is bought, workflows are adjusted and citizens are informed after the fact. That approach may work for office software. It fails when AI changes how public power sees and sorts people.
The manifesto’s critique of industrial-era public systems is relevant because AI can easily become another layer of bureaucratic distance. Instead of a person lost in paperwork, the citizen becomes a person lost in a model output. The interface looks modern, but the democratic relationship remains weak.
Fair AI rules should move in the opposite direction. They should make government more explainable, more contestable and more responsive. If AI helps summarize public comments, translate documents, identify service delays or reveal unequal outcomes, it can strengthen democratic administration. If it hides responsibility, accelerates punishment or replaces public reasoning with opaque scoring, it weakens democracy.
The difference is not the technology alone. It is the political structure around it.
Frequently Asked Questions
What is deliberative democracy in AI governance? Deliberative democracy in AI governance is a process where representative citizens learn about AI systems, question experts, discuss tradeoffs and recommend rules before government adopts or expands AI tools.
Why should citizens help shape AI rules in government? Citizens should help shape AI rules because public AI can affect rights, services and accountability. Technical experts can explain how systems work, but the public must judge what uses are fair and legitimate.
How is discursive democracy different from deliberative democracy? Discursive democracy focuses on the broader public exchange of reasons, arguments and narratives. Deliberative democracy is usually more structured, with selected participants, evidence, facilitation and formal recommendations.
Does fair AI governance mean banning AI in government? No. Fair AI governance means deciding where AI may help, where strict safeguards are required and where certain uses should be off limits. The goal is democratic control, not automatic rejection.
What should every government AI rule include? At minimum, government AI rules should include public notice, human responsibility, appeal rights, impact assessment, audit duties, procurement transparency and clear limits on high-risk uses.
A democratic test for public AI
The central question is not whether government should use AI. It is whether government can use AI without weakening the citizen’s standing before the state.
Deliberative democracy gives governments a way to answer that question in public, with citizens rather than around them. It turns AI rulemaking from a procurement issue into a civic act.
If JustSocial’s vision is to help build a more continuous, participatory public life, fair AI rules are one of the first places to start. The systems that shape public decisions must be governed by public judgment. Anything less asks citizens to trust power they cannot see, question or meaningfully change.
