Week 19: AI, Democracy, and Civic Participation

 

Welcome to Week 19 of the AI Literacy Course.

Imagine that a municipality asks residents to comment on a new transport plan. Thousands of responses arrive in several languages.

The municipality uses AI to translate, group, and summarise the responses. The final summary says that most participants support the plan.

But important questions remain:

• Did the summary preserve disagreement?

• Did a rare accessibility concern disappear?

• Were repeated or automatically generated responses identified?

• Could residents check how their contribution was interpreted?

• Did the consultation influence the decision?

• Will the municipality explain what changed?

AI can make participation more accessible and manageable. It can also hide voices, distort priorities, or create the appearance of participation without real influence.

The central rule is:

Show where the decision is, protect every route to participation, preserve disagreement, and require an accountable response.

Learning goals

By the end of this lesson, you should be able to:

• distinguish access, participation, influence, agency, and accountability;

• identify how AI may support or distort civic participation;

• assess AI-supported translation and summarisation;

• recognise hidden political targeting and coordinated synthetic participation;

• distinguish informational public tools from consequential decision systems;

• apply VOICE to a public consultation;

• design feedback, correction, and stop procedures.

1. Democracy is more than elections

Democracy includes:

• free and fair elections;

• freedom of expression and information;

• independent institutions;

• public records and trustworthy journalism;

• civil society;

• public discussion and disagreement;

• equal rights;

• meaningful opportunities to influence public decisions.

AI can affect this whole system.

It may influence:

• what information people receive;

• which voices become visible;

• how public comments are summarised;

• who is targeted by political messages;

• how public services communicate;

• how authorities make or support decisions.

A large number of clicks does not necessarily mean that people have democratic influence.

2. Access, participation, and influence are different

These concepts are connected but not identical.

Access

People can obtain information and reach the participation process.

Examples include an accessible website, a telephone line, a meeting, translated information, or a paper form.

Participation

People can contribute views, questions, experiences, or proposals.

Influence

Contributions can affect a real decision.

Organisers should explain what participants can and cannot change.

Agency

People understand their choices and can participate, question the process, correct errors, or choose another channel.

Accountability

Decision-makers explain what they decided, what evidence they used, and how public contributions affected the result.

A process may provide access and collect thousands of comments while offering little real influence.

3. Begin with the decision

Before asking people to participate, organisers should explain:

• What decision is being made?

• Who has authority to make it?

• Which parts are still open to change?

• Which parts are already fixed?

• How will contributions be used?

• When will the decision be made?

• How will participants receive feedback?

• How can an error be corrected or challenged?

Do not collect opinions when there is no clear plan for using or responding to them.

Participation without response can create frustration and reduce trust.

4. AI can widen access

AI may support civic participation through:

• translation;

• plain-language explanation;

• text-to-speech;

• speech-to-text;

• captions;

• accessible formats;

• search;

• question-and-answer tools;

• summaries of long public documents.

These tools may help more people understand public information and express their views.

However, an AI-generated answer should not become a dead end.

People need:

• a route to the original source;

• the date of the information;

• an explanation of uncertainty;

• an accessible alternative;

• a responsible human contact.

A public chatbot should distinguish general information from an official decision about an individual case.

It should not make a proposal sound final or remove a person’s right to apply, explain circumstances, request review, or use another channel.

5. Summarisation can help and distort

AI can help organise a large number of consultation responses by:

• translating them;

• identifying themes;

• grouping related comments;

• finding repeated questions;

• producing draft summaries.

However, the categories and instructions given to the system influence what becomes visible.

An AI summary may lose:

• rare concerns;

• minority languages;

• disagreement;

• irony;

• uncertainty;

• emotional meaning;

• local knowledge;

• unusual but serious risks.

The most common theme is not automatically the most important.

One response might identify a serious safety problem, discrimination, or an accessibility barrier affecting a small group. A democratic process must not treat that contribution as unimportant merely because it is rare.

6. Organise input without turning it into a vote

Consultations are not always referendums.

A theme mentioned 500 times does not automatically outweigh a legal problem, safety risk, or effect on a small community mentioned five times.

AI can help organise input, but it should not decide:

• which rights matter;

• which evidence is strongest;

• which risk is acceptable;

• whose experience deserves attention;

• what the public authority should do.

Human decision-makers must examine:

• frequency;

• evidence;

• severity;

• affected groups;

• rights;

• feasibility;

• disagreement;

• public purpose.

They must then explain how these considerations shaped the decision.

7. Make AI-supported summaries auditable

An accountable consultation should document:

• the consultation question;

• who could participate;

• participation channels;

• languages offered;

• number and type of responses;

• AI tool and version, when available;

• prompts or instructions;

• categories used;

• handling of duplicate responses;

• human review;

• minority and dissenting views;

• material omitted from the public summary;

• privacy protections;

• limitations and uncertainty;

• correction process.

People should be able to understand how many different contributions became a shorter summary.

Where appropriate, publish examples of how themes were formed without exposing personal or confidential information.

8. Protect personal data and political opinions

Consultation responses may contain:

• names;

• contact details;

• political opinions;

• information about health or disability;

• migration history;

• personal stories;

• criticism of employers or authorities;

• information about small communities.

Political opinions linked to identifiable people are sensitive personal data.

Apply the A–P–K process from Lesson 15:

• Is the tool and context approved?

• Does the material identify anyone?

• Is it sensitive, confidential, or potentially harmful?

Removing names may not be enough. A detailed story or a combination of location, occupation, and event may still identify someone.

Original contributions should be preserved only according to a defined purpose, access policy, security plan, and retention period. “Preserve the record” does not mean “publish every response” or “store it forever.”

9. Watch for coordinated synthetic participation

Generative AI can produce thousands of comments, letters, or social-media posts at low cost.

One organisation or individual may create many apparently independent messages to make a position look more popular than it is. This is sometimes called astroturfing.

Synthetic flooding can:

• overwhelm staff;

• hide genuine contributions;

• distort theme counts;

• create false public pressure;

• reduce trust in participation.

Organisers need proportionate ways to identify repeated or coordinated submissions.

Possible methods include:

• examining repeated wording;

• analysing submission timing;

• separating identical responses from unique arguments;

• requiring limited verification for particular processes;

• publishing how duplicates were handled;

• keeping a human review route.

Authenticity checks must not become excessive surveillance or exclude people who share devices, use templates, need assistance, or have limited digital access.

A repeated message may still express a genuine view. The goal is to understand the character of the participation—not automatically silence organised civic action.

10. Political communication requires transparency

AI can help create:

• political text;

• images;

• audio and video;

• translated campaign messages;

• targeted message variants;

• synthetic personas;

• impersonations.

When you encounter political communication, ask:

• Who is speaking?

• Who authorised the message?

• Who paid for it?

• Is it paid political advertising or ordinary political speech?

• Was AI used?

• Is the AI use disclosed where required?

• Why did this audience receive the message?

• What evidence supports its claims?

• Can journalists, opponents, and citizens inspect other versions?

A genuine candidate may use AI assistance. Satire may use synthetic media. A targeted advertisement may contain authentic footage but a misleading claim.

Do not decide trust from an AI label alone. Apply SIFT and the harm-reducing verification process from Lessons 12 and 14 when checking factual claims.

11. Political targeting can fragment public debate

Data-driven targeting can deliver different political messages to different audiences.

Targeting may help campaigns reach interested voters. It may also:

• rely on personal or inferred characteristics;

• exploit fears or vulnerabilities;

• test contradictory promises;

• hide messages from public scrutiny;

• prevent citizens from seeing what other groups were told.

Public transparency can help citizens understand:

• the sponsor;

• the amount spent;

• audience criteria;

• reach;

• dates;

• message versions;

• use of targeting.

EU political-advertising rules have applied fully since October 2025. They require political advertisements to be labelled and include information such as who paid and, where targeting is used, information about the targeted audience.

These rules concern political advertising, not every personal political opinion or ordinary political conversation.

12. Freedom of expression and protection from harm

Democracy requires space for lawful expression, criticism, satire, disagreement, and unpopular views.

It also requires protection against:

• threats;

• harassment;

• fraud;

• impersonation;

• coordinated manipulation;

• unlawful content;

• abuse that drives people out of public discussion.

Automated moderation may misunderstand:

• context;

• satire;

• minority languages;

• reclaimed words;

• political criticism;

• differences in power.

Over-removal may silence lawful voices. Under-enforcement may allow abuse to exclude people from participating.

Responsible moderation procedures include:

• understandable rules;

• reasons for important decisions;

• access to human review;

• correction and appeal;

• transparent reporting;

• appropriate response to repeated abuse.

13. Inclusion requires more than an online form

People face different barriers to civic participation, including:

• language;

• disability;

• limited literacy;

• lack of time;

• limited devices or connectivity;

• low confidence;

• fear of retaliation;

• migration concerns;

• distrust of institutions;

• previous experience of being ignored;

• identification within a small community.

Offering everyone the same online form does not guarantee equal participation.

A consultation may need:

• accessible web forms;

• telephone participation;

• paper forms;

• public meetings;

• interpreters;

• captions;

• screen-reader compatibility;

• community partners;

• trusted local organisations;

• confidential or anonymous channels;

• assistance without loss of the participant’s own voice.

Digital channels should complement—not automatically replace—human routes.

14. Public-sector AI requires public responsibility

Authorities may use AI for different purposes.

Information and access tools

Examples include:

• translation;

• document search;

• plain-language summaries;

• general service chatbots.

These tools still require accuracy, accessibility, source links, privacy safeguards, and human contact.

Consultation tools

Examples include:

• translation of responses;

• theme identification;

• duplicate analysis;

• draft summaries.

These require data protection, auditable methods, preservation of disagreement, and human review.

Consequential systems

Some systems may help determine:

• eligibility;

• benefits;

• inspections;

• risk;

• enforcement;

• access to services;

• other decisions affecting rights.

These require stronger legal, technical, and organisational safeguards.

Certain public-authority or public-service uses of high-risk AI are subject to specific requirements under the EU AI Act, including fundamental-rights assessment in defined situations.

Not every public chatbot is automatically high-risk. The responsible authority must identify which rules apply.

15. A vendor cannot carry democratic accountability

A public authority may buy technology, but it cannot outsource its responsibility to the supplier.

The authority needs sufficient information about:

• the system’s purpose;

• data flows;

• known limitations;

• evaluation results;

• updates;

• human responsibilities;

• incident reporting;

• audit access;

• correction procedures;

• exit and replacement arrangements.

Commercial complexity or secrecy should not make public decisions impossible to explain.

Authorities should also retain appropriate records showing:

• which system or version was used;

• what instructions were applied;

• which information was processed;

• who reviewed the result;

• what final decision was made.

Without records, later correction, appeal, research, or public scrutiny may become impossible.

16. Close the participation loop

A democratic consultation should not end when comments have been collected.

Participants should receive a clear response:

What we heard

Summarise common themes, disagreement, minority concerns, and important evidence.

What we changed

Explain which parts of the proposal changed because of participation.

What we did not change

Identify suggestions that were not adopted.

Why

Give understandable reasons, including legal, financial, safety, evidence, or feasibility considerations.

What happens next

Provide the timetable and responsible decision-maker.

How to challenge an error

Explain how participants can correct a misinterpreted contribution or raise a complaint.

This closing loop turns participation into accountable influence rather than one-way data collection.

17. Civic AI literacy is a shared responsibility

Citizens need skills, but they should not be expected to detect every manipulation or audit every public system alone.

Democratic AI literacy depends on many institutions.

Schools and folk high schools

Teach critical, civic, and practical AI literacy.

Libraries

Provide public access, accessible support, and trusted spaces for questions.

Journalists

Investigate claims, sponsors, targeting, systems, and public decisions.

Public authorities

Provide official sources, explanations, records, and correction routes.

Election bodies

Communicate reliable information about voting and election procedures.

Platforms

Provide transparency, reporting, moderation, and appeal systems.

Civil society

Represent affected communities, monitor power, and support participation.

Researchers

Evaluate systems, evidence, methods, and social effects.

Collective capacity prevents responsibility from being transferred entirely to individual citizens.

18. Use VOICE

Use VOICE to evaluate an AI-supported democratic process.

V — Voice

Ask:

• Who can participate?

• Who is absent?

• Whose contributions are preserved?

• Who can influence the outcome?

• Are disagreement and minority concerns visible?

O — Openness

Ask:

• Is the purpose clear?

• Who owns the decision?

• What role does AI play?

• Who paid or sponsored the process?

• How are data and responses used?

• What are the limits?

I — Information

Ask:

• Can people reach original sources?

• Are dates and uncertainty visible?

• Can factual errors be corrected?

• Does the tool distinguish general information from official decisions?

C — Contestability

Ask:

• Can people question a summary or decision?

• Is human review meaningful?

• Can errors be corrected?

• Is there a complaint or appeal route?

• Who has authority to respond or stop harm?

E — Equity

Ask:

• Which barriers affect participation?

• Which groups may experience unequal errors or exclusion?

• Are several accessible channels available?

• What additional support is needed for meaningful influence?

Classroom activity: Democratic consultation canvas

Use a fictional, nonpartisan local issue, such as a park design, library opening hours, recycling service, or public-transport timetable.

Do not use real political opinions, personal data, or unpublished consultation responses.

Complete this canvas:

Area	Questions
Decision	What real decision is being made, and who owns it?
Influence	What can participants change?
Participants	Who is affected, included, absent, or unsafe?
Channels	What online and offline routes are available?
Information	What original sources and explanations are provided?
AI role	What will AI do—and not do?
Data	What is collected, inferred, retained, or published?
Summarisation	How will disagreement and rare concerns be preserved?
Authenticity	How will coordinated synthetic submissions be handled proportionately?
Review	Who checks the AI output and can reject it?
Feedback	How will organisers explain what changed and why?
Challenge	How can participants correct or appeal?
Stop conditions	What would pause or end the AI-supported process?

Present your design as a short written analysis, audio response, or presentation.

Reflection questions

1. Where is the real decision?

2. What can participants change?

3. Does the process provide access, participation, and influence?

4. What might disappear in an AI summary?

5. Could frequent responses hide a rare but serious concern?

6. How are personal data and political opinions protected?

7. Could coordinated synthetic submissions distort the process?

8. Who checks, explains, corrects, and can stop the AI-supported process?

9. How will organisers report what changed and why?

10. Which institutions should support citizens?

Key vocabulary

Civic participation:
Ways people contribute to public discussion, community action, and public decisions.

Agency:
The ability to understand choices, act, question a process, and use another route.

Accountability:
Responsibility for explaining decisions, correcting errors, and responding to consequences.

Political advertising:
Paid or sponsored communication intended to influence political views, participation, or decisions under applicable rules.

Microtargeting:
Using data to deliver tailored messages to particular groups or individuals.

Astroturfing:
Coordinated activity designed to look like independent public support or opposition.

Contestability:
The ability to question, challenge, correct, or appeal an outcome.

Equity:
Identifying different barriers and providing the support needed for meaningful participation.

VOICE:
Voice, Openness, Information, Contestability, and Equity.

Summary

In Week 19, we learned that AI can make public information and participation more accessible through translation, search, summaries, and accessible formats.

The same tools can:

• omit minority concerns;

• distort disagreement;

• expose sensitive information;

• enable hidden political targeting;

• flood consultations with synthetic responses;

• make public decisions harder to inspect.

Democratic participation requires more than access. People need real influence, agency, and an accountable response.

Use VOICE:

• Voice;

• Openness;

• Information;

• Contestability;

• Equity.

A democratic process shows where the decision is, documents AI’s role, protects personal information, preserves disagreement, supports several participation channels, and explains what changed.

Show where the decision is, protect every route to participation, preserve disagreement, and require an accountable response.

Official guidance

• European Commission: Political-advertising transparency and targeting

• European Commission: EU AI Act

• European Commission: Navigating the AI Act

• European Commission: Digital Services Act

• European Data Protection Board: Personal data in elections

Lesson 19 Interactive Quiz: AI, Democracy, and Civic Participation

Choose one answer for each question. Then select Check my answers. This practice quiz does not collect names or scores.

1. A municipality collects thousands of comments but has already made the final decision. What is the main democratic problem?

A. The consultation did not use enough AI features B. People were invited to participate without a clear route to influence C. A consultation must always follow the most common opinion D. Public decisions should never include consultation

2. Which design makes a municipal AI chatbot most accountable?

A. Present every answer as a final official decision B. Hide source documents to make answers shorter C. Remove human contact because the chatbot is available all day D. Link to current official sources, show dates and limits, and provide a human route

3. An AI summary omits an accessibility problem because only four participants mentioned it. What should reviewers do?

A. Restore and assess the concern because importance is not determined only by frequency B. Ignore it because consultations are simple majority votes C. Delete all rare comments before human review D. Count the four responses as zero because they form a small group

4. Which record would make an AI-supported consultation summary most auditable?

A. Only the polished final summary B. Only the total number of responses C. The tool, prompts, categories, duplicate method, human checks, omissions, and limitations D. A statement that the vendor is responsible for every error

5. Consultation comments contain names, political opinions, and detailed personal stories. What is the safest approach?

A. Publish every original response to maximise openness B. Use approved processing, minimise identifiers, limit access, and define retention C. Assume that removing names always makes the comments anonymous D. Upload them to any public AI tool because the issue is civic

6. Thousands of nearly identical comments arrive within minutes. What is the most democratic response?

A. Treat every message as a different independent citizen without checking B. Delete every repeated message and ignore its argument C. Require intrusive identity checks from every participant D. Investigate coordination, report duplicate handling, preserve genuine arguments, and avoid unnecessary surveillance

7. You receive a targeted paid political advertisement. Which information is most useful for democratic scrutiny?

A. Who paid and authorised it, the audience criteria, and why it reached you B. Only the advertisement’s background colour C. Only whether an AI image was used D. No information because all political communication is identical

8. A local consultation is available only through a complex online form in one language. Which improvement best applies the Equity part of VOICE?

A. Require everyone to buy a newer device B. Keep one identical channel because equal treatment means no support C. Add accessible web, language support, telephone, paper, meetings, and assisted participation D. Let the AI system decide which groups do not need to participate

9. A vendor provides an AI system used to support an important public decision. Who remains democratically accountable?

A. Only the vendor because it wrote the software B. The responsible public authority, with meaningful human responsibility and adequate vendor information C. The affected citizen, who must detect every system error D. Nobody, because complex systems cannot be explained

10. Which final step best closes the democratic participation loop?

A. Collect comments and provide no further information B. Publish only the number of participants C. Let the AI summary become the final decision automatically D. Explain what was heard, what changed, what did not, why, what follows, and how to correct errors

Reflection: Apply VOICE to one fictional consultation: Voice, Openness, Information, Contestability, and Equity.