Finland’s citizens’ initiative system allows a group of citizens to submit either a proposal that law-drafting should begin or a complete bill containing the proposed legal sections. If an initiative gathers 50,000 verified statements of support within six months, Parliament must consider it, although Parliament remains free to amend or reject it. The route is unusually concrete: a public concern can arrive at the legislature not only as a petition, but as text that resembles the beginning of a law.
That possibility exposes an old inequality. People experience the consequences of law every day, but few know how to identify the responsible level of government, distinguish an administrative failure from a legislative gap, compare policy options or express an idea in clauses that lawyers can examine. Organisations with money can hire people who do this translation. An individual tenant, parent, care worker or small association usually cannot.
Generative AI can narrow that gap. It can help a person move from “this keeps happening to us” to a documented problem, a jurisdictional question, several policy options and a draft that can be criticised. But it should not be described as letting everyone manufacture laws tailored to private preference. The democratic promise is more exacting: AI could help more people prepare proposals that are responsive to lived needs and fit for public argument.
Drafting is not lawmaking
A model can arrange words in legislative form. It cannot confer legal competence, reconcile constitutional rights, allocate public money or establish that a proposal has democratic support. Those acts belong to institutions and procedures whose legitimacy comes from more than fluent text.
Nor should “tailored” mean that every person receives a private rule. Law normally has to define a class of situations, apply criteria consistently and remain open to challenge. A personal difficulty is therefore a starting point, not the final scope. A parent worried about dangerous traffic outside one school might begin with a specific crossing, but a serious proposal must ask which schools face the same conditions, who would be affected by a restriction, which authority controls the road and what evidence would justify action.
This is where AI can be useful. It can preserve the concrete experience while repeatedly widening the question: Is this an isolated error, a failure to enforce an existing rule, a poorly designed service, or a genuine gap in law? Who benefits from the current arrangement? Who would carry the cost of changing it? What would count as a counterexample?
From a complaint to a public problem
The first useful output is not a bill. It is a problem statement. A good assistant can interview a citizen about what happened, separate observation from assumption and turn scattered details into questions that evidence could answer. It can help record frequency, affected groups, present remedies, administrative responsibilities and the result the person actually wants.
That resembles the disciplined opening of professional policy work. The European Commission’s better-regulation process begins with problem definition, objectives, policy options, evidence, consultation and expected impacts before a legislative proposal is finalised. Its public “Have Your Say” system also accepts feedback throughout the policy lifecycle. AI could make those conceptual tools usable at a much smaller scale without pretending that a citizen has become the Commission, a ministry or a parliamentary counsel.
The second output is a route map. Depending on the place and subject, the appropriate path might be a municipal consultation, a petition, a citizens’ initiative, contact with a representative, a complaint to an existing authority or a request to enforce rules already on the books. At EU level, for example, a European citizens’ initiative can invite the Commission to propose legislation within EU competence; it does not itself enact a law. Reaching one million signatures across the required spread of member states obliges examination and a response, not adoption.
An AI can help locate and compare these routes, translate official instructions and make a checklist. It must show its sources and label the route unverified until checked against the current official procedure. Jurisdiction is not a detail to be filled in later. It determines whether the proposed actor has power to do anything at all.
Use the model to produce alternatives, not certainty
Once the problem and authority are clearer, the model can do something that hurried political debate often neglects: generate materially different options. One option may change a law. Another may improve enforcement, information, funding or service design. A third may preserve the current rule while creating a limited pilot. The citizen can then compare who gains, who pays, how compliance would be observed and what unintended behaviour each option might encourage.
This is also a natural place for adversarial testing. Ask the model to construct cases involving a person unlike the original proposer: someone with a disability, a rural resident, a small organisation, a person with limited digital access or somebody who reasonably opposes the measure. Ask what evidence would disprove the proposed explanation. Ask whether a powerful actor could exploit an exception intended for a vulnerable group.
Alkemata has previously argued that institutions should test a policy before it tests the public. A citizen-facing drafting assistant could use the same discipline: create synthetic cases, historical edge cases and community-designed scenarios before real rights or duties change. If part of a proposal can be represented computationally, rules as code can help expose ambiguity, but executable logic should remain a traceable companion to authoritative legal text.
The danger of authoritative nonsense
Legal language is especially good at making generated errors look official. A model can invent a statute, misstate the powers of an institution or combine rules from different jurisdictions into a confident answer. Empirical studies of questions about United States federal cases have found high rates of legal hallucination in general-purpose models; later testing of specialist legal research products found fewer but still significant errors. Those studies did not test citizen legislation directly, but they establish the relevant warning: legal fluency is not reliable evidence of legal accuracy.
The remedy is not a larger warning below the chat box. It is a different working object. Every legal claim should be paired with an official source, retrieval date and status: verified, uncertain or disputed. The assistant should quote only the minimum necessary text and link to the complete authority. Draft clauses should contain annotations explaining their purpose and unresolved dependencies. Before formal submission, a qualified person familiar with the jurisdiction must review competence, compatibility with superior law, drafting conventions and consequences.
AI should also preserve disagreement. If the assistant repeatedly rewrites objections until they fit the proposer’s preferred answer, it has turned participation into persuasion. The better pattern is the one described in Do Not Make Human–AI Agreement the Goal: keep the competing view visible, identify the evidence that would matter and record why a judgement changed.
The danger of personalised politics
Cheap drafting does not automatically distribute political power. Well-organised groups could use models to flood consultations with thousands of polished variations, creating the appearance of broad participation. People with better models, data and lobbying knowledge could still dominate. Training data may reproduce yesterday’s administrative categories and make unfamiliar needs harder to express. A system optimised to produce agreeable text may hide conflict that democratic institutions need to hear.
Participation therefore cannot be measured by document volume. The OECD’s guidance on citizen participation distinguishes open submissions from representative deliberative processes in which a broadly representative group receives evidence, weighs trade-offs and develops recommendations. AI may help participants understand material and formulate amendments, but selection, facilitation, time, accessibility and the treatment of minority positions remain institutional design questions.
A defensible citizen-law tool would consequently maintain a non-AI route, publish its sources and versions, disclose substantial machine assistance, limit duplicate automated submissions and make distributional questions mandatory. It would help people find one another around a shared problem, not generate the illusion that each person can order a law to specification.
A public-law workbench
The useful product is less like a chatbot that answers “write me a law” and more like a workbench. It begins with a record of lived experience. It produces a jurisdiction map, an evidence register and a public problem statement. It compares legislative and non-legislative options. It tests affected groups and difficult cases. Only then does it prepare a clause skeleton with plain-language annotations and questions for a lawyer, public official or citizens’ group.
Crucially, each stage remains revisable. If the authority is wrong, the route map changes. If consultation reveals that the problem was framed too narrowly, the objectives change. If an edge case exposes disproportionate harm, the proposed rule changes. The model’s speed becomes useful because revision is cheap, not because its first answer is presumed correct.
The accompanying capsule turns that process into a small exercise. It helps a reader create a Public-Law Proposal Card for one concrete concern, identify the responsible jurisdiction, compare three possible interventions and run a paper consultation before drafting formal clauses. It does not produce legal advice or authorise a submission.
A modest democratic gain
AI will not solve the unequal distribution of political attention, and democracy should not become a competition between automated proposal generators. Its more credible contribution is humbler: give more people access to the intermediate skills that turn experience into an examinable public proposition.
A citizen who can name the problem, locate the authority, show the evidence, compare alternatives, anticipate objections and present a revisable draft is not replacing a legislator. They are arriving at the democratic conversation better equipped. The question for institutions is whether they will build procedures capable of hearing that contribution without confusing fluent text with public consent.
Sources and further reading
- Finland Ministry of Justice: citizens’ initiatives, briefly in English
- European Commission: how the European Citizens’ Initiative works
- European Commission: Better Regulation
- OECD Guidelines for Citizen Participation Processes
- Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models
- Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools
Subscribe to Alkemata for new viewpoints, explainers and deep dives on technology for human capability.
Turn this idea into a project
The companion capsule helps you turn one lived problem into a Public-Law Proposal Card: a public problem statement, a provisional jurisdiction map, an evidence register, three intervention options and a small consultation test. It deliberately postpones formal clauses until authority, evidence and affected groups are visible.
Download the capsule and attach it to your AI assistant, or expand the text below and copy it into a new conversation. Add one sentence about your situation. The capsule will guide the next steps.
Capsule version 1.0 · Experimental; desk-designed and not field-validated. This is a proposal-development method, not legal advice.
Read or copy the complete capsule
ALKEMATA CAPSULE Capsule ID: alk-public-law-proposal-001 Title: Citizen Law Proposal Workbench Version: 1.0 Date: 30 September 2026 Status: Experimental; desk-designed and not field-validated Source article: Let Citizens Draft the First Version of the Law Planned canonical URL: https://alkemata.com/2026/09/30/let-citizens-draft-first-version-law/ WordPress post ID: 1252 Scope: Early-stage citizen proposals for public rules, policies or legal reform Not legal advice: This capsule does not determine whether a proposal is lawful, constitutional, within an institution's power or suitable for formal submission. # Purpose Help a person turn one lived problem into a Public-Law Proposal Card that can be discussed, criticised and improved before anyone attempts formal legal drafting. The first session should produce: 1. a public problem statement; 2. a provisional jurisdiction and participation route; 3. an evidence-and-uncertainty table; 4. three materially different intervention options; 5. a small paper consultation to test the proposal. The aim is not to generate a law on demand. It is to make the reasoning between personal experience and a public proposal visible. # Activation Help the reader apply this capsule to their situation. If no objective is supplied, explain the three available missions briefly and ask at most three essential questions: 1. In what place and level of government did the problem occur? 2. What happened, to whom, and what outcome would be better? 3. Is the reader seeking a new law, a change to an existing rule, better enforcement, or help deciding which route applies? If enough context is already available, begin. Teach only the concepts needed for the next decision. Produce the first useful artifact rather than a general essay. Label missing facts, jurisdictional uncertainty, assumptions and hypothetical examples. Follow the reader's choices. Do not invent legislation, cases, institutional powers, public support or evidence. Never present a generated clause as valid law. Do not submit, publish, contact officials or collect signatures without the reader's explicit authorisation. # Essential knowledge ## 1. A personal need can reveal a public problem Begin with the person's experience, but do not assume that its solution should become a law. Ask whether the difficulty is: - an isolated error; - failure to enforce an existing rule; - an inaccessible or badly designed service; - absence of information or resources; - conflict between existing rules; - or a genuine gap requiring a new or amended law. A public proposal should explain which class of people or situations it covers and why the distinction is justified. ## 2. Jurisdiction comes before drafting The relevant authority might be municipal, regional, national, federal or supranational. It might be legislative, administrative, regulatory or judicial. Similar words can have different legal effects in different places. Use current official sources to determine: - which body has power over the subject; - whether citizens may make a formal initiative, petition or consultation response; - eligibility, format, language, signature and deadline requirements; - whether a complete bill is permitted or only a request for action; - privacy, funding-disclosure and campaigning rules. Until these facts are checked, label the route PROVISIONAL. ## 3. Separate four kinds of material Lived experience: what the reader or affected people observed. Evidence: records, public data, research and testimony supporting or challenging the explanation. Legal authority: constitutions, statutes, regulations, official procedures and authoritative decisions. Proposal: the reader's preferred interpretation, objective and intervention. AI can help organise all four, but it must not convert one into another. A compelling story is not proof of prevalence; a cited law is not proof that it is enforced; a popular proposal is not necessarily lawful. ## 4. Drafting is downstream Do not begin with formal clauses. First establish: - the problem and baseline; - affected groups; - competent authority; - policy objective; - alternative interventions; - costs, benefits and distribution; - implementation and enforcement; - review, appeal and correction; - evidence gaps; - conditions for revision or repeal. Formal drafting becomes useful only when those elements are sufficiently clear for critique. ## 5. Law must remain contestable A proposal should make it possible to know: - who makes a decision; - what facts and criteria matter; - how affected people are notified; - how mistakes can be corrected; - how a decision can be reviewed or appealed; - who is responsible for harm; - how the rule will be evaluated. Do not optimise only for efficiency. Consider agency, accessibility, comprehensibility, privacy, distribution, responsibility and resilience. # Missions ## Mission A — Turn an experience into a public problem Use this when the reader knows what is going wrong but not what kind of change is needed. Output: a one-page problem statement separating observation, explanation, missing evidence and desired outcome. ## Mission B — Test an existing proposal Use this when the reader already has a petition, manifesto, campaign demand or draft clause. Output: a proposal stress test covering authority, affected groups, alternatives, difficult cases, implementation and evidence. ## Mission C — Prepare a conversation draft Use this when the problem, authority and policy choice are reasonably established. Output: a plain-language proposal and an annotated clause skeleton for discussion with a qualified lawyer, public official, representative body or citizens' group. # Workflow ## Step 1 — Describe the lived situation Input: the reader's account and any non-sensitive documents they choose to provide. Action: write two short sections titled "What was observed" and "What is presently inferred." Remove identifying details that are not necessary. Output: a factual situation note. Move on when: the reader confirms that observation and inference have not been confused. ## Step 2 — Form the public problem Input: the confirmed situation note. Action: identify the affected class, current baseline, recurring mechanism and intended public outcome. Ask what evidence would show that the case is exceptional rather than systemic. Output: a public problem statement of no more than 200 words. Move on when: it describes a shared situation without claiming unsupported prevalence. ## Step 3 — Map authority and participation route Input: location, governmental level, subject and desired type of change. Action: consult current official sources. Distinguish a petition, citizens' initiative, consultation response, administrative complaint, enforcement request and ordinary political advocacy. Output: a Jurisdiction and Route Map with source links, retrieval dates and unresolved questions. Move on when: each claimed power or procedural requirement has an official source, or is clearly marked unverified. ## Step 4 — Build the evidence table For each important proposition record: | Claim or question | Material available | Source | Supports / challenges | Status | |---|---|---|---|---| | What is happening? | | | | verified / uncertain / disputed | | Who is affected? | | | | | | Why does it happen? | | | | | | Does an existing remedy work? | | | | | | What would change the outcome? | | | | | Move on when: the proposal's essential assumptions and missing evidence are visible. ## Step 5 — Compare three interventions Generate three genuinely different options: A. improve enforcement or service delivery without changing the law; B. amend or create a rule; C. run a limited pilot, collect evidence or use a non-regulatory measure. For each option assess authority, expected benefit, affected groups, cost, implementation, misuse, accessibility, review and reversibility. Output: an options table and a reasoned provisional choice. Move on when: the preferred option is not simply the first idea rewritten three times. ## Step 6 — Run opposition and edge cases Construct at least four tests: - a person who benefits; - a person unintentionally burdened; - a person with limited money, mobility, language or digital access; - a powerful actor who might exploit the rule. State the strongest good-faith objection. Identify what evidence could change the recommendation. Output: an opposition note and proposed corrections. Move on when: disagreement remains visible and the reader decides what to revise. ## Step 7 — Prepare the Public-Law Proposal Card Use the template below. If formal clauses are requested, add only an annotated skeleton. Every clause should state its intended function, dependency and unresolved legal question. Move on when: the card can be understood without the conversation that produced it. ## Step 8 — Seek human and institutional review Identify the kind of review required: affected community members, legislative counsel, a local authority, subject expert, rights organisation or elected representative. Do not claim endorsement. Preparing a document does not authorise submitting it. # Small reversible experiment: a paper proposal clinic Hypothesis: Three contrasting readers can identify a hidden assumption or affected group before formal drafting begins. Procedure: 1. Complete a first Public-Law Proposal Card. 2. Remove unnecessary personal information. 3. Ask three people to review it: one person affected by the problem, one plausible critic and one person who would implement the proposal. 4. Ask each person what they believe the rule would change, who might lose, what is unsupported and what they would revise. 5. Record disagreement rather than averaging it away. 6. Revise the card once. Resources: the card, source links and sixty to ninety minutes. No software or AI is required. Observe: whether reviewers understand the same objective, identify missing evidence and expose different consequences. Stopping conditions: stop if the discussion exposes confidential information, imminent risk, targeted hostility, an unresolved conflict of interest or a need for professional legal representation. Next action: verify the jurisdictional route or seek qualified drafting review. Do not treat a successful discussion as proof of public support. # Public-Law Proposal Card Title: Version and date: Jurisdiction and governmental level: Observed situation: Public problem: People and situations covered: People consulted: Current rule, service or practice: Competent authority and official source: Participation route and requirements: Policy objective: Option A — non-legislative: Option B — legislative: Option C — pilot or evidence-gathering: Preferred option and reasons: Evidence supporting it: Evidence challenging it: Affected groups and distribution: Implementation responsibility: Notification, explanation and accessibility: Correction, review or appeal: Costs and dependencies: Misuse and difficult cases: Measure of success: Review or expiry condition: Unresolved legal questions: Next reviewer or action: # Optional annotated clause skeleton Use only after the previous sections are complete. 1. Purpose — the public outcome sought. 2. Scope — people, institutions and situations covered. 3. Definitions — only terms requiring a precise legal meaning. 4. Duties or powers — who must or may do what. 5. Decision criteria — facts that may lawfully matter. 6. Procedure — notice, reasons, time limits and accessibility. 7. Safeguards — privacy, human review, correction and appeal. 8. Implementation — responsible body, resources and commencement. 9. Monitoring — evidence to collect and public reporting. 10. Review, amendment or expiry — when the rule must be reconsidered. For every item include: - intended function; - evidence or policy reason; - superior-law dependency; - foreseeable edge case; - question for qualified counsel. # Verification Before recommending formal drafting, verify: - Is the institution legally competent? - Are all legal citations retrievable from official sources? - Is current law distinguished from the proposed change? - Is the problem supported by more than one unverified account? - Were non-legislative options considered? - Are affected and excluded groups visible? - Can a person understand, correct and challenge an adverse decision? - Are implementation cost and responsibility named? - Could a powerful actor exploit the proposed exception? - What evidence would contradict the proposal? - Is there a review, amendment or expiry mechanism? - Has qualified jurisdiction-specific review been identified? Counterexample: If a delayed benefit resulted from a single incorrect database entry, a new entitlement law may add complexity without fixing the data-correction process. Use the administrative correction route first. Do not use this capsule as a substitute for urgent legal advice, legal representation, constitutional review, criminal-law drafting, taxation advice or assessment of an individual's rights and deadlines. # Manual route Print the Public-Law Proposal Card and complete it with official documents, conversations and handwritten notes. The method does not require AI. A librarian, community organisation, representative's office or legal clinic may help locate current procedures. # MCP access Alkemata's public MCP endpoint is: https://alkemata.com/wp-json/mcp/alkemata-public It is anonymous and read-only. After the source article is published, a compatible client can: 1. call alkemata-search-published-content with the article title; 2. retrieve WordPress post ID 1252 using alkemata-get-published-content; 3. read this complete capsule in the article's "Turn this idea into a project" section; 4. preserve the canonical article URL and capsule version. Drafts and private posts are not exposed. MCP improves retrieval; it does not validate the legal accuracy of the material or authorise action. # Portable checkpoint At the end of each session return: Objective: Capsule ID and version: Jurisdiction: Participation route: Verified facts and sources: Provisional assumptions: Preferred intervention: Important objection: Current artifact: Unresolved legal questions: Human review required: Next action: # Optional return formats Project passport: objective, capsule/version, jurisdiction, problem, affected groups, preferred option, stage, first test, next milestone and help needed. Field report: what was actually attempted, participants, observations, evidence, disagreement, unexpected consequences, remaining uncertainty and proposed correction. Never describe a simulated result as field evidence. Help request: one precise question, jurisdiction, expertise sought, relevant public sources, current artifact and expected contribution. Before sharing, remove personal or confidential information. Preparing a report does not authorise sending it. A verified route for optionally discussing a well-prepared project with Alkemata is:Collaborate# Sources Finland Ministry of Justice, Citizens' initiatives — briefly in English: https://www.kansalaisaloite.fi/fi/ohjeet/briefly-in-english European Commission, European Citizens' Initiative — how it works: https://citizens-initiative.europa.eu/how-it-works_en European Commission, Better Regulation: https://commission.europa.eu/law/law-making-process/better-regulation_en OECD Guidelines for Citizen Participation Processes: https://www.oecd.org/content/dam/oecd/en/publications/reports/2022/09/oecd-guidelines-for-citizen-participation-processes_63b34541/f765caf6-en.pdf Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models: https://arxiv.org/abs/2401.01301 Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools: https://arxiv.org/abs/2405.20362 END OF CAPSULE
For example, after transferring it, say: “A dangerous junction outside our school has produced repeated near misses. Help me determine whether this is an enforcement, street-design or rule-making problem, and create the first Public-Law Proposal Card without inventing local powers.”
Use it through MCP
After this article is published, a compatible AI client can connect anonymously to https://alkemata.com/wp-json/mcp/alkemata-public, search for this title and retrieve post ID 1252. The complete capsule is embedded here, so it is returned with the published article. Drafts are not exposed through the public MCP service.
If you develop a proposal and want to prepare a careful request for expertise, use the Alkemata Collaborate page. Remove private information and decide what you want to share before sending anything.
Featured image: “Oslo Parliament, Norway” by Claudia Regina, licensed under CC BY-SA 2.0. No modifications.
