Hacking Democracy / ETH Zurich, September 7, 2026Joshua C. Yang

Building
Digital
Democracy
in Practice

Build tools that give people a meaningful say.

Three illustrated people bringing different perspectives to a shared discussion board

A little about me

Joshua C. Yang

Postdoc · ETH Zurich
PhD · ETH Zurich
Computational Social Science team

PhD thesis: Designing computational methods for digital democracy

Aarau

Participatory budgeting
Helped design the city’s first programme

Winterthur

Kultur Komitee
Process design over three years

Taiwan

vTaiwan
Active member

Switzerland

Murmi
App design for the Swiss AI Futures citizens’ assembly

Collective decisions

What can democracy look like?

Where do you already make decisions together?

Friends comparing food choices — Dinner with friends

People watching a song performance and expressing preferences — Eurovision voting

Illustrated Switzerland with people and shared places — Public decisions

Shared budgets · Workplace decisions · Citizens’ assemblies · Referendums

Begin with a working idea of democracy: people having a meaningful say in decisions that affect them. These everyday and civic examples invite questions about participation, power and shared decisions. A group choosing dinner is not a complete democratic political system; it lets us examine the same design questions on a small scale.

Collective decisions range from choosing dinner with friends and managing a shared-house budget to elections, citizens’ assemblies and public funding. Ask which settings people recognize. The examples are invitations to notice processes they already know.

Everyday coordination is not identical to public democracy: rights, authority and accountability differ. But these settings let us examine how a group shares information, deals with disagreement and chooses what to do. Use the range to widen possible project contexts.

Invite two quick examples from the room: where have you made a decision with other people this week? Start with dinner or a favourite song, then move to public decisions. Eurovision and dinner illustrate aggregation and coordination, not equivalence to political democracy. Avoid explaining current Eurovision rules here. The Swiss image is a conceptual illustration, not a geographically precise map.

Timing cue: leave this scene by 2:30.

Collective intelligence

How groups use what people know

A group’s knowledge is distributed. The process determines what becomes visible and usable.

Different knowledge

Invite experiences that others may not have.

Independent views

Collect initial views before people follow the crowd.

Structured exchange

Compare reasons, preserve disagreement and revise.

Agreement, accuracy and democratic legitimacy are different goals.

Collective intelligence asks how a group uses knowledge that no one person has alone. Democratic theory also asks who gets a say and who has the right to decide.

Imagine a neighbourhood deciding how to use a shared space. Different people know different needs. Discussion lets them compare those needs and consider changes to the proposals. A decision rule then determines the outcome.

Landemore argues that including different experiences and perspectives can improve public judgment. That is a reason to pay attention to participation and discussion, not a guarantee that every group will make a better decision. Equality, legitimate authority and the ability to challenge a result also matter.

These are design questions, not universal guarantees. Lorenz et al. (2011) show that social influence can reduce diversity in estimation tasks; Woolley et al. (2010) associate group performance with social sensitivity and turn-taking. See the reading list. Discussion may also reveal information unavailable in isolated judgments.

Timing cue: leave this scene by 3:30.

Research directions

Four research directions

Computational social choice

How should we count votes, share a budget or select representatives?

Deliberative democracy

How can discussion include different views and inform public decisions?

Human–AI interaction

How do people use, check and correct help from AI?

Complexity science & collective intelligence

How do interactions and networks shape what groups learn and decide?

These are overlapping fields, not a sequence of progress. The same budgeting tool can raise a social-choice question about allocation, a design question about understanding, and an ethics question about who is included.

Keep the field names because students can use them to find courses, researchers and papers. The questions underneath explain what each field studies in ordinary language.

Current work includes AI-assisted discussion, allocation rules with real-world constraints and ways to evaluate models of human behavior. The reading guide includes primary papers and explains what each finding does and does not show. Students can choose any of these research directions.

Computational social choice studies the properties of formal decision rules. Complexity science and collective intelligence study how group-level learning and behavior emerge from interactions, networks and feedback. Computational social science offers methods used across these directions; it is not presented as another kind of voting-rule design.

Timing cue: leave this scene by 4:30.

Democracy as an open-source project

Rules are inherited, adapted and changed along different paths.

United Kingdom

Parliamentary oversight

Parliamentary institutions influenced the US framers. The Parliament Acts of 1911 and 1949 later limited the Lords’ ability to block certain bills.

United States · 1787

Checks & balances

British and state precedents informed a federal constitution with a president and Congress. Its two-chamber federal model then influenced Switzerland and Australia.

Switzerland · 1848

Direct democracy

A federal parliament drawing on the US model. Later changes: optional law referendum (1874), initiative for partial constitutional revision (1891), and the first proportional National Council election (1919).

Australia · 1901

Compulsory voting

British parliamentary government combined with a US-inspired federal Senate. Later changes: House preferential voting (1918), compulsory federal voting (1924), and the first proportional Senate election (1949).

Selected influences and reforms · US Senate · Swiss Parliament · AEC

Democracy as an open-source project is a useful analogy: institutional rules can be inspected, borrowed and revised. The branching structure follows Motivation 1 in Self-Introduction.pptx. Follow the British line into the US branch, then the US branch into Switzerland and Australia. Australia also drew directly on British parliamentary government, as its entry label explains.

The short country labels pick out one feature of each branch, not an exclusive national invention. Parliamentary oversight: scrutinize government and require public explanations. Checks and balances: divide authority so each branch can constrain the others. Direct democracy: let citizens challenge laws and propose constitutional changes. Compulsory voting: address low participation; the 1924 reform aimed to increase turnout. These are design purposes, not guarantees that a system achieves them.

Feature-label sources: UK Parliament, https://www.parliament.uk/about/how/role/scrutiny/ ; US government branches, https://www.usa.gov/branches-of-government ; Swiss direct democracy, https://www.aboutswitzerland.eda.admin.ch/en/direct-democracy ; Australian Electoral Commission, https://www.aec.gov.au/About_AEC/Publications/backgrounders/compulsory-voting.htm .

These are selected institutional influences, not a complete ancestry or a literal software fork. The US framers drew on British institutions and state constitutions while rejecting monarchy. Their 1787 federal compromise created a population-based House and equal state representation in the Senate. The US Senate documents these precedents at https://www.senate.gov/about/origins-foundations/senate-and-constitution.htm . Switzerland’s two chambers drew on the American model in 1848: https://www.parlament.ch/fr/über-das-parlament/Pages/Ablauf-Wahlen-1848.aspx . Australia combined Westminster government with a US-inspired federal Senate in 1901; see the Australian Parliament source in the reading guide.

Continue along the country branches to show different paths of change. UK Parliament Acts in 1911 and 1949 limited the Lords’ power over certain bills. Switzerland added the optional legislative referendum in 1874 and the initiative for partial constitutional revision in 1891. Proportional National Council elections were approved in 1918 and first held in 1919. Australia introduced House preferential voting in 1918 and compulsory federal voting in 1924; Senate proportional representation was legislated in 1948 and first used in 1949. Dates are attached to the relevant changes. The diagram is schematic, not a shared time scale.

The Swiss constitutional referendum and initiative for total revision already existed in 1848. The 1891 label here refers to partial revision. The UK changes do not remove every Lords veto. The Swiss electoral-history source is https://www.parlament.ch/fr/über-das-parlament/Pages/Vom-Majorz-zum-Proporz.aspx ; the Australian dates are documented at https://www.aec.gov.au/Elections/history-of-electoral-reform.htm .

Different paths do not imply inevitable progress or equal access to political power. These systems excluded many people; Swiss women gained federal voting rights only in 1971. Ask what a changed rule allows people to do and whose voice it changes. Later, the voting example will isolate one design choice; institutions have many such choices.

Timing cue: leave this scene by 5:45.

The process

Five parts of a collective decision

People define the participants and purpose at a board

Framing

Define the purpose and participants

Ideas

Bring knowledge and possibilities together

Deliberation

Compare reasons and perspectives

Voting

Turn preferences into a decision

Accountability

Explain, follow through and learn

A shared starting point. Adapt it to the process; revisit earlier steps.

Kultur Komitee · Winterthur

What is Kultur Komitee?

Kultur Komitee participants seated in a circle, with the Winterthur programme on the screen — Kultur Komitee, Winterthur · a residents’ funding assembly

A residents’ assembly that funds art and cultural projects in Winterthur.

I’ve worked on its voting and funding process for three years.

Over CHF 2 million

distributed to art projects in Winterthur across those three years.

Kultur Komitee is a residents’ budget assembly in Winterthur. Participants consider art and cultural proposals, discuss them, assess them and decide how the available funding should be allocated.

I have worked on its voting and aggregation across the 2024, 2025 and 2026 editions, including interfaces, allocation rules and the explanation of results. Fynn Bachmann and I have worked on the calculations and aggregation. Across my three years of involvement, we have distributed over CHF 2 million to art projects in Winterthur.

We can now look closely at what this work involves. The next five sections use actual fieldwork and tools to examine the choices within the process.

Sources: Joshua C. Yang confirmed the updated three-year total of over CHF 2 million on 6 September 2026. This supersedes the older CHF 1.2 million statement in the repository project page. The total spans the three-year involvement; the following published 2025 case figures describe a specific deployment and should not be multiplied to derive this total. Role and editions: https://joshuacyang.com/projects/kultur_komitee/ . Photograph: Joshua’s supplied images/IMG_2826.JPG; no individual event date is inferred from the image.

Timing cue: leave this scene by 8:00.

Step 1 of 5

Framing

What are we deciding, and who should take part?

Citizen initiativesHelp people turn a concern into a clear public question.

Citizens’ assembliesMap who is affected and whose experience is missing.

Step 1: Framing. Establish who has a say, what is being decided and what authority the group has. Introduce Kultur Komitee on the following fieldwork page.

Ask: What can the group decide, and who takes part? Give the audience a moment to consider it, then continue to the example.

Conceptual process illustration supplied by Joshua: images/process-01-framing.png. Repeated from the overview to identify this stage.

These are possible design directions, not prescribed hackathon solutions. Start by researching the actual procedure, people and constraints. Citizen initiatives: Help people turn a concern into a clear public question. Citizens’ assemblies: Map who is affected and whose experience is missing.

Timing cue: leave this scene by 8:40.

1 / FramingKultur Komitee · Winterthur

Who decides, and what can they decide?

Kultur Komitee, Winterthur · participants gathered for a plenary session — Kultur Komitee participants gathered for a group session

Kultur Komitee · 2025

Who decides?38 residents

Selected at random

What can they decide?CHF 380,000

Funding for local culture

Step 1: define the decision. In the published 2025 deployment, 38 randomly selected residents allocated CHF 380,000 to cultural projects. These are figures from that study, not a headcount inferred from the photo.

The participants, budget and scope are already design choices. Ask who should have a say, what the group can decide, and which constraints have been set beforehand. These choices shape all the work that follows.

Compare this with Swiss direct democracy next: an initiative proposes a change, while a referendum challenges legislation. The fieldwork photos were supplied by Joshua; the deployment is documented at https://joshuacyang.com/pubs/komitee-equal-shares/ .

Timing cue: leave this scene by 9:20.

What happens
before your
tool exists?

Follow an existing decision from start to finish. Where do people already have a say—and where does their influence stop?

People defining who takes part and what the group can decide

Pause after the framing case. A participation tool needs a real connection to a decision. Research the institution: what is genuinely open, who has authority, and what commitment exists to respond? A polished consultation can still leave participants with no influence. For a project, investigate this connection before choosing a technology. This is a design principle, not a claim that every process must give participants final authority.

Timing cue: leave this scene by 9:50.

Ask students to describe what happens today without mentioning an app: who starts the process, who does the work, who decides, and what happens afterwards? Existing practices may contain hard-won knowledge. Build from that understanding; a small improvement can be useful without replacing the whole process.

Step 2 of 5

Ideas

What do people know, need or propose?

Participatory budgetingHelp residents develop and compare feasible proposals.

Neighbourhood planningCollect local knowledge and explore alternative designs.

Step 2: Ideas. Gather experiences, evidence, needs and possible courses of action. A formal proposal is one way to contribute; other processes begin with questions, stories or shared observations. The next photograph shows the project information people worked with in Kultur Komitee.

Ask: What information do people need to compare the options? Give the audience a moment to consider it, then continue to the example.

Conceptual process illustration supplied by Joshua: images/process-02-ideas.png. Repeated from the overview to identify this stage.

These are possible design directions, not prescribed hackathon solutions. Start by researching the actual procedure, people and constraints. Participatory budgeting: Help residents develop and compare feasible proposals. Neighbourhood planning: Collect local knowledge and explore alternative designs.

Timing cue: leave this scene by 10:30.

2 / IdeasKultur Komitee · Winterthur

Project information for participants

Project briefs, assessment sheets and annotations on a working table

Residents compare project descriptions, requested amounts and the criteria used to assess them.

For discussion if time allows: What information would you need before comparing two proposals?

Step 2: Ideas. This example focuses on: What information makes the options clear and comparable?

Point to the briefs and assessment sheets. Give students a few seconds to suggest what they would need: the purpose, cost, intended beneficiaries, feasibility and uncertainties. These are reflection prompts, not a claim that every field appears on every photographed sheet.

Proposal preparation includes work by applicants and organizers before the group meets. During review, residents may discover missing or ambiguous information. A simple comparable brief or missing-information checklist can improve an existing process.

Notice the ordinary materials: paper, notes and pens. Ask what currently takes time or causes misunderstanding before deciding that a new platform is necessary.

Photograph supplied by Joshua in the conference workspace. Deployment source: https://joshuacyang.com/pubs/komitee-equal-shares/ .

Timing cue: leave this scene by 11:05.

Step 3 of 5

Deliberation

How can people compare reasons and perspectives?

Public consultationsCompare arguments and surface unresolved disagreements.

Multilingual assembliesSupport translation, source checking and fair summaries.

Step 3: Deliberation. Discussion lets people compare reasons and bring concerns into the decision. Continue to the table discussion, then the other deliberation examples.

Ask: How do people compare reasons and raise concerns? Give the audience a moment to consider it, then continue to the example.

Conceptual process illustration supplied by Joshua: images/process-03-deliberation.png. Repeated from the overview to identify this stage.

These are possible design directions, not prescribed hackathon solutions. Start by researching the actual procedure, people and constraints. Public consultations: Compare arguments and surface unresolved disagreements. Multilingual assemblies: Support translation, source checking and fair summaries.

Timing cue: leave this scene by 11:45.

3 / DeliberationKultur Komitee · Winterthur

Small-group discussion and assessment

Residents discussing project material around a table

Residents discuss how each proposal could contribute to the cultural goals the group has agreed on.

For discussion if time allows: How would you keep a concern from being lost when the group disagrees?

Step 3: Deliberation. This example focuses on: How can people compare reasons and raise concerns?

Allow a short response, then distinguish sharing preferences from deliberation: participants explain reasons, ask questions and assess expected contributions against shared criteria.

Komitee combines individual preferences with these small-group evaluations. The paper documents the two roles; a photograph cannot show whether every person had equal influence or whether the discussion improved the eventual decision.

This is where distributed knowledge can enter collective judgment. Facilitation, useful evidence and a record of unresolved concerns matter. Later, Murmi and the summary exercise show digital tools for related tasks.

Photograph supplied by Joshua in the conference workspace. Deployment source: https://joshuacyang.com/pubs/komitee-equal-shares/ .

Timing cue: leave this scene by 12:20.

3 / DeliberationCitizens’ assemblies · Ireland

Citizens’ assemblies study an issue before making recommendations

01 Select participants by lottery
02 Study evidence and discuss in groups
03 Make recommendations; receive a response

Possible tools: a clear evidence summary, a record of open questions, or a page tracking the public response.

Irish Citizens’ Assembly · process and responsibilities

Step 3: Deliberation. This example focuses on: How can people compare reasons and raise concerns?

Irish citizens’ assemblies use a selection process intended to reflect the public, evidence sessions and facilitated discussion before recommendations. Recommendations require political consideration and a response; they are not automatically binding decisions.

Civic technology can help people find and question evidence, record unresolved questions or follow the response to recommendations. Selection is also a research problem: Flanigan et al. study fairness of selection probabilities subject to demographic quotas.

For students, an evidence pack can be a sufficient project. Pick one question that an assembly would examine, link competing claims to their sources and test whether a reader can explain the remaining disagreement.

Timing cue: leave this scene by 13:05.

3 / Deliberation

Murmi helps people check a discussion summary

People can review the main themes and point out what is missing.

People share experiences and concerns.

Their comments provide the material for a group summary.

Contribute → organize themes → participant review. People can check and correct the shared interpretation.

Workshop interface; controls illustrate the process.

Open Murmi ↗

Murmi in use on a phone at Swiss AI Futures. The participant is reviewing a statement about AI and translation. — Murmi in use at Swiss AI Futures, 2026.

The embedded Murmi website opens at “See Murmi at work.” Scroll within the page and use its Vote, Record, Clusters and Common ground carousel controls to show the real interface. This public showcase contains screenshots; it is not a populated live session. Offline view retains the workshop photograph and three-part explanation. Do not join the historical session shown in the photograph.

Step 3: Deliberation. This example focuses on: How can people compare reasons and raise concerns?

Use the three controls to explain the procedure: participants contribute; the system organizes themes; participants review and correct the shared interpretation. The photo is authentic, while these controls are a guided explanation rather than a replay of the historical session.

The central design problem is whether people recognize their views and can correct a misrepresentation. A related academic example is Tessler et al.’s study of AI-mediated group statements, but that study does not establish Murmi’s effects.

A small contribution could expose the original contributions behind one theme, collect a correction or retain a question that still needs discussion. Compare the summary with its sources and ask participants what is missing.

Timing cue: leave this scene by 14:35.

3 / DeliberationIllustrative discussion

What makes an AI summary fair?

A “A garden would give us a place to meet.”

B “The music event would bring people together.”

C “I would use a workshop space every week.”

D “Whichever we choose, I need step-free access.”

Draft summary

People want more places to meet and do things together.

Generative social choice: choose statements that represent different groups.

Step 3: Deliberation. This example focuses on: How can people compare reasons and raise concerns?

Compare these four invented contributions from a community discussion with the draft summary. Give the audience ten seconds to find an omission, then reveal the step-free-access requirement.

The summary captures shared interest in social activities but loses a concrete condition on participation. Frequency is not the only measure of importance. A team could build a source-linked correction action or a list of requirements that must remain visible.

This is a teaching example, not output from Murmi. Evaluate a prototype against the original contributions: which claims lack support, which requirements are omitted and can someone correct the error?

Generative social choice treats summary construction as a representation problem: select a short set of generated statements that covers different groups’ views. Boehmer, Fish and Procaccia, ICML 2025, study proportional representation under a text-length budget. Their guarantees depend on a formal preference model and query assumptions. https://proceedings.mlr.press/v267/boehmer25a.html

The access example illustrates an omission, not that algorithm or a proportionality guarantee. Ask separately: is the summary faithful, are different groups represented, and are essential rights or access needs preserved? Participants must be able to inspect and correct it.

Timing cue: leave this scene by 15:50.

Whose
perspective
did your
system lose?

A neat summary can hide a real disagreement. What should stay visible, even when it complicates the result?

Three people discussing different perspectives

Pause after the summary example. The purpose of a democratic tool is not to manufacture agreement. A useful outcome can be a clearer tradeoff, an unresolved concern, or a better option. Ask what would help participants understand a view they do not share. Preserve space for dissent and correction when summarizing. This is the design aspiration illustrated by the preceding case.

Timing cue: leave this scene by 16:20.

Connect this question to both democracy and system design. Which perspectives survive an interface, ranking or summary? Who can notice an omission and correct it? Disagreement may expose a missing requirement or a tradeoff rather than an error to be removed.

Step 4 of 5

Voting

How do preferences become a collective decision?

ReferendumsLet people explore how ballot wording changes the choice.

Shared budgetsCompare voting rules and which groups each outcome serves.

Step 4: Voting. A procedure combines individual and group inputs into a result. A ballot is one form of input; the Komitee example also includes group assessments and a budget rule.

Ask: How are different views turned into a result? Give the audience a moment to consider it, then continue to the example.

Conceptual process illustration supplied by Joshua: images/process-04-voting.png. Repeated from the overview to identify this stage.

These are possible design directions, not prescribed hackathon solutions. Start by researching the actual procedure, people and constraints. Referendums: Let people explore how ballot wording changes the choice. Shared budgets: Compare voting rules and which groups each outcome serves.

Timing cue: leave this scene by 17:00.

Democracy as a design resource

Equal for whom?

Each person
Give individuals equal weight.

Each community
Ensure distinct groups have a voice.

Who should your system represent—and how?

Two people considering individual contributions together and distinct groups of contributions

Inspired by slides 56–58 of Self-Introduction.pptx. Democratic institutions prompt a distinction between representing people and representing groups. Neither framing by itself settles what is fair. Ask which people and communities a system affects, who defines the groups, and how influence should be distributed.

Translate that question into system design: an average user rating and representation across language communities may prioritise different things. This is a conceptual comparison, not a claim that all democracies or fairness mechanisms fall into two exclusive types.

Timing cue: leave this scene by 17:45.

4 / VotingIllustrative group decision

The same votes can give different winners

Same nine voters.

4Garden › Music › Workshop

3Music › Workshop › Garden

2Workshop › Music › Garden

Only first choices count.

Garden4 votes

Music3 votes

Workshop2 votes

Garden wins.

Plurality: Garden wins, 4 / 3 / 2 votes.
Borda: Music wins, 8 / 12 / 7 points.

Start with this decision before introducing theory or history. Nine people rank a garden, a music event and a workshop. These are invented teaching ballots. Ask the audience which project should win.

With plurality, count only first choices: garden receives four, music three and workshop two. Ask what changes if second choices count. Switch to Borda: first gets two points, second one and third zero. Music wins with twelve points, ahead of garden on eight and workshop on seven.

Nobody changed their vote. Only the aggregation rule changed. Use this to introduce the individual assessments and Equal Shares example that follow. Different rules express different priorities; no rule is universally best.

Timing cue: leave this scene by 19:30.

4 / VotingStadtidee · Aarau · 2023

Aarau’s first participatory budget

Residents decide what their city funds

CHF 50,00033 proposals · about 1,800 voters

Residents voted with 10 points.
The Method of Equal Shares allocated the budget.

17 projects funded
across all city districts.

My role: co-designing the vote and how results were explained with the city team.

Residents distributing tokens among an imagined neighbourhood’s shared amenities

Stadtidee Aarau · project and implementation

Aarau launched its first participatory budgeting programme in 2023. Residents voted on 33 proposals for a CHF 50,000 budget; about 1,800 residents participated. Josh worked with Sebrina, Jasmin and Lea from the city on implementation, from the voting design to communicating outcomes. Preliminary lab experiments informed the setup.

The city chose a 10-point ballot and the Method of Equal Shares. The project page reports 17 funded projects across all districts, compared with 7, generally more expensive projects under a most-votes-first comparison. This is a comparison of allocation rules on the case, not a separate randomised trial or a guarantee that more projects always means a better result.

Connect this example to the previous voting-rule slide: a ballot collects preferences; an allocation rule also has to account for costs and representation within a limited budget. Explaining the method before voting and publishing the district breakdown afterwards are part of the implementation. The illustration is conceptual, not an actual screenshot of Aarau’s ballot.

Source: https://joshuacyang.com/projects/aarau_pb/

Timing cue: leave this scene by 20:15.

Conceptual illustration generated for this presentation in the established figurative gouache style; not a documentary image or quantitative result. Adapted from Self-Introduction.pptx.

4 / VotingKultur Komitee · Winterthur

Design the vote, the rule and the explanation

Three people expressing preferences, allocating resources and explaining the result

Express preferences

What can people tell you?

Allocate resources

Which rule translates votes into funding?

Explain the result

Can people understand and question it?

For discussion if time allows: What should happen when popular projects leave less money for other priorities?

Step 4: Voting. This example focuses on: How are different views turned into a result?

The photograph shows individual assessment of the displayed proposals; do not describe a specific ballot being cast or infer anyone’s preference from the image. The published procedure separately establishes that participants cast individual point votes.

Ask for one principle students would want the allocation to respect: proportional representation, shared impact, feasibility or support for smaller projects. There can be tensions among these aims.

Komitee Equal Shares combines personal preferences and shared group assessments. The next slide uses the Komitee handout to isolate one part: how individual points and equal virtual budgets can produce an allocation. The handout is a simplified invented example, and does not show the group assessment component.

Photograph supplied by Joshua in the conference workspace. Deployment source: https://joshuacyang.com/pubs/komitee-equal-shares/ .

Timing cue: leave this scene by 20:45.

This diagram separates interface design, allocation and explanation, following slides 10–11 of Self-Introduction.pptx. Each offers a distinct intervention. In Kultur Komitee, individual votes and group assessments contribute to the allocation. The generated illustration depicts concepts, not actual data or a specific ballot.

4 / Voting

How votes become a funding result

Komitee’s Method of Equal Shares handout: three participants’ points, equal budgets and contributions to six project proposals. An English walkthrough appears alongside. — Open the original handout ↗

Invented example · 3 people · CHF 1,000 each

Each person assigns seven points

Up to two per project. The handout puts projects in order of total points.

Votes and project costs in the handout’s funding order
Project	Cost / CHF	Anna	Miro	Jasmin	Points

Points express preferences. Budgets determine which contributions can still be made.

Four projects share CHF 3,000

Two projects are skipped when supporters cannot cover the contributions shown.

Final funding result in the handout
Project	Points	Funding / CHF

All three budgets are spent. Jasmin’s remaining CHF 600 funds Bilder Buch, which has two points.

Creative Club: Anna 400 + Miro 200 + Jasmin 400 = CHF 1,000. Auto Ausstellung is skipped: Anna has CHF 200 left but would need CHF 600.

Step 4: Voting. The source is the Method of Equal Shares handout used by Kultur Komitee and supplied by Joshua. Its caption explicitly says this is a simplified example with invented data. Use the original graphic as the overview and the English walkthrough to explain the arithmetic.

Votes: Anna, Miro and Jasmin each assign seven points, with at most two per project, and each starts with a virtual CHF 1,000. The handout orders projects by total points. Retain its printed order for ties; the sheet does not establish a general tie-breaking rule.

First project: Creative Club costs CHF 1,000 and has five points. In this example, each point pays CHF 200. Anna’s two points contribute CHF 400; Miro’s one contributes CHF 200; Jasmin’s two contribute CHF 400. Remaining budgets are CHF 600, 800 and 600.

Budget limit: Experimental Event is next, costing CHF 800. Anna and Miro each contribute CHF 400, leaving CHF 200, 400 and 600. Auto Ausstellung has four points and costs CHF 1,200: Anna would need CHF 600, Miro CHF 300 and Jasmin CHF 300. Anna cannot pay, so the handout skips it and deducts nothing.

Result: Dance Disco is funded with CHF 200 from Anna and CHF 400 from Miro. Folklore Fest is skipped because Miro and Jasmin cannot cover their contributions. Bilder Buch is funded with Jasmin’s remaining CHF 600. The four funded projects cost CHF 1,000 + 800 + 600 + 600 = 3,000. Each person spends CHF 1,000. Ask the audience to notice how remaining budgets affect the result: the total point count alone does not decide everything.

This is a walkthrough of the handout’s simplified procedure, not a general implementation of MES or a reconstruction of a real allocation. It illustrates individual points and equal budgets; the group impact assessments in Komitee Equal Shares are not represented here. The 2025 deployment involved 38 citizens allocating CHF 380,000 to 43 projects; the receipt later in the talk uses separate 2026 data.

Source: images/KKW_Infokarten_Grafiken_platziert_MES.pdf, page 1, Method of Equal Shares. The complete original is bundled and linked beside the rendering. Formal foundation: https://proceedings.neurips.cc/paper_files/paper/2021/hash/69f8ea31de0c00502b2ae571fbab1f95-Abstract.html . Field study: https://doi.org/10.1145/3772363.3798958 .

Transition to step 5: after computing an allocation, people need to see how it was reached. This explanation and the voting receipt make the calculation inspectable.

Timing cue: leave this scene by 21:45.

Step 5 of 5

Accountability

What happens next, and what can we learn?

Public commitmentsTrack what was promised, funded and delivered.

Community decisionsHelp participants question a result and request a review.

Step 5: Accountability. Computing a result is followed by explaining it and tracking what happens. A receipt is one example of an explanation; implementation and responsibility also matter.

Ask: Can people understand the result and see what happens next? Give the audience a moment to consider it, then continue to the example.

Conceptual process illustration supplied by Joshua: images/process-05-accountability.png. Repeated from the overview to identify this stage.

These are possible design directions, not prescribed hackathon solutions. Start by researching the actual procedure, people and constraints. Public commitments: Track what was promised, funded and delivered. Community decisions: Help participants question a result and request a review.

Timing cue: leave this scene by 22:25.

5 / AccountabilityKultur Komitee · Winterthur

Presenting the funding result

The allocation table presented back to the room

The 2025 allocation funded 43 projects. Receipts connect contributions to the projects that received funding.

For discussion if time allows: What would you want to check if your preferred project was not funded?

Step 5: Accountability. This example focuses on: Can people understand the result and see what happens next?

Point to the results table. A collective decision must be communicated to participants and applicants. Ask what they would want to inspect: the inputs, rule, available money, explanation, or route to raise a concern.

The photograph documents presentation of an allocation table. It does not establish that funded projects were delivered or that there was a formal appeal procedure. Following implementation is a further accountability task; a progress record is a possible hackathon contribution.

We are now at step 5: explain the result and follow up. Return briefly to the other four steps to show how the project information, discussion and votes led here. The next demo uses a separate published KK26 receipt from 2026.

Photograph supplied by Joshua in the conference workspace. Deployment source: https://joshuacyang.com/pubs/komitee-equal-shares/ .

Timing cue: leave this scene by 22:55.

5 / Accountability

A receipt shows how your share helped fund each project

Choose a project to see the contribution and the full cost.

Förderpool Kultur

This participant's share contributed CHF 250 to the project.

Project cost: CHF 4,000.

Open the original receipts ↗

KULTUR KOMITEE
VOTING RECEIPT

Participant DNW / group BLAU

Total allocatedCHF 7,497.00

4 supported projects funded.
Amounts are contributions from a virtual share.

Presentation view of published receipt data.

Yang et al. · Digital Government 2024

The embedded website opens at the published 2026 personal receipts. Use EN to switch to English, scroll through receipts, or inspect the Projects and Table pages. The website controls remain inside the frame. Offline view keeps the four verified contributions for participant DNW. Use the presentation’s bottom arrow to continue after interacting inside the page.

Step 5: Accountability. This example focuses on: Can people understand the result and see what happens next?

This presentation view uses the published 2026 DNW receipt. Its four contributions total CHF 7,497. Select a project to inspect the participant’s contribution alongside the project’s cost. A recorded vote and an allocated amount are distinct.

A readable receipt can support a question about the allocation. It is not by itself a complete fairness or security audit. Research on digital voting interfaces asks how design affects perceived fairness, legitimacy and understanding.

Accountability continues after allocation. Decidim’s accountability component links results to implementation progress. Another small project could connect an approved proposal to its responsible body, milestones and evidence of delivery.

Timing cue: leave this scene by 24:25.

Who gets to
change
the rules?

Let democracy inform system design: give affected people a voice, a way to challenge decisions, and a role in revising the rules.

Residents jointly adjusting the balance between algorithmic allocation and human deliberation

Connect the receipt to a broader principle: people should be able to inspect how an outcome was reached, raise a problem and learn what happens next. Transparency becomes useful when someone can act on it. A focused improvement can matter; its value depends on whether it helps people do something meaningful in the process. The earlier receipt comparison remains a possible evaluation example, not the required hackathon task.

Timing cue: leave this scene by 25:10.

Democracy can inspire the governance of a system as well as provide a use case for it. Who sets its objectives, who can contest an output, and who can revise the rules? Consider voice, accountability and revisability as design questions. Democratic legitimacy is not achieved merely by adding a vote button.

Conceptual illustration generated for this presentation in the established figurative gouache style; not a documentary image or quantitative result. Adapted from Self-Introduction.pptx.

+1 / AI agents & simulation

Explore possible interactions

What could we learn before bringing people together?

Deliberation agents

Rehearse a discussion. Vary the evidence, facilitation or decision rules.

Personas & digital twins

Explore responses grounded in consenting participants’ accounts.

Human-controlled assistants

Help someone explore arguments, then let them edit and approve their position.

An illustrated person and a wooden mannequin compare preference cards

Use simulations to develop questions. People retain voice and decision authority.

This is an optional research direction across the five stages, not a sixth mandatory step. Deliberation agents simulate interactions; personas model perspectives; a digital twin suggests a model tied to a particular person. Treat all as imperfect models.

Park et al., LLM Agents Grounded in Self-Reports Enable General-Purpose Simulation of Individuals, arXiv:2411.10109, revised June 2026, tests agents grounded in interviews and surveys. Predictive performance is not consent or authority to speak for someone. https://arxiv.org/abs/2411.10109

Human-controlled assistants are a proposed building direction here. Require a person to inspect, edit, revoke and approve any representation. Test simulations against held-out human observations and examine failures across groups. Bisbee et al. (2024), Synthetic Replacements for Human Survey Data?, cautions that matching averages can hide distorted variation.

Timing cue: leave this scene by 26:25.

Participation connected to institutions

Shape the places we share

OmaStadi ↗

Helsinki

People allocate resources among shared projects

Residents propose and choose local projects.

31,103 voters · €10m budget2026 round

VelObserver ↗

Switzerland

Conceptual illustration of people and civic participation

Cyclists rate routes. Shared evidence makes gaps in the network visible.

Supported by Prototype FundPublic route ratings across Swiss cities

OmaStadi remains the international reference Josh knows. Helsinki recorded 31,103 voters, a €10 million allocation and 54 selected proposals in 2026. https://paatokset.hel.fi/en/case/hel-2026-004390?decision=b7f23e87-e7f8-4607-9900-90776b1b6764

VelObserver is the bicycle project in the Prototype Fund Switzerland catalogue, round 2. Cyclists rate route segments; the platform aggregates subjective assessments and aims to track improvements. Its public site lists Zurich, Basel, Bern, Gossau, Aarau, Winterthur and Limmattal. The Bern view displayed 161 contributors and 9,012 ratings when checked; these are local cumulative counts, not nationwide reach. https://prototypefund.opendata.ch/en/projects/ and https://velobserver.ch/

The Swiss map is an illustration, not a route map or deployment screenshot. A useful project could connect residents’ experience to an existing planning or monitoring process. Ratings alone do not guarantee infrastructure changes.

Timing cue: leave this scene by 27:25.

Swiss public-interest tech

Make public information easier to use

@gora ↗

Carbon Copy · Zurich

Explore an issue, compare perspectives and discuss before forming a view.

Used in Swiss participation projectsSwiss Bevölkerungsrat 2025 · provider case study

ZüriCityGPT ↗

Independent project · Zurich

Ask a question about city information and check the original source.

Publicly accessible toolBuilt by Liip · not an official city service

Carbon Copy is the Zurich-based association behind @gora. The current product combines learning material, different perspectives, polls and discussion. Its field notes document the Swiss Bevölkerungsrat 2025, Swiss AI Futures and school projects. These are provider-reported deployments, not independent evidence of effectiveness or a large unique-user count. https://www.atgora.org/ and https://www.atgora.org/field-notes

ZüriCityGPT is the likely Zurich GPT reference. Liip’s public OSS version answers questions using Zurich website content and open-source models. It explicitly states it is not associated with the City of Zurich. This is access-to-information public tech, not a binding democratic decision mechanism. Check answers against official pages; no verified adoption figure is claimed. https://oss.zuericitygpt.ch/ and https://github.com/liip/SwissGovernmentChatbots

Only VelObserver is identified here as Prototype Fund-supported. Do not imply that @gora or ZüriCityGPT received that funding. The illustrations are conceptual, not product screenshots. Sources checked 6 September 2026.

Timing cue: leave this scene by 28:10.

For your own project

Democracy is still being built!

Let’s improve how it works, one real process at a time.

Would you or others actually use it?

What existing process gets better?

What must you learn before building?

People passing, discussing and revising proposals in a continuing loop — Express → discuss → revise → decide → revisit

Papers, project demos and process examples ↗

joshuacyang.com/projects

Joshua’s projects

End with possibility: democratic institutions and everyday collective processes can change. You do not need to redesign the whole system to contribute. Start from curiosity, learn with others and make something worth trying. The direction is theirs to discover; an uncomplicated idea can make a meaningful contribution.

Choose a project or process you want to understand. Find out how it works, who is involved and what is already known. Investigate the context before committing to a solution.

Talk to the people involved, read relevant research and explore previous attempts. Ask what you still need to learn and which assumptions you should check.

Use that research to consider different possibilities and decide what you want to build. The projects in this talk are examples to learn from. Your direction can differ from theirs.

Invite one or two brief reflections if there is time. The linked projects and papers can support further research. Finish by 28:45 and leave 1:15 for questions or variation.

Conceptual illustration supplied by Joshua: images/civic-building-a-prototype.png. The image represents exploring and making an idea; it does not prescribe a particular project or solution.

Bring the three principles together: meaningful influence, useful disagreement and decisions people can question. Invite students to research a real process, learn with those involved, and choose an improvement worth testing. The ambition is a meaningful change in what people can do, even when the tool itself is small.

Timing cue: leave this scene by 29:30.

Ask for an honest usefulness test: would you use this in a process you belong to, or do the intended participants see a reason to use it? What work would it save or enable, and what extra work would it create? Enthusiasm for a demo is not evidence of lasting usefulness. Research existing tools and practices, talk to the people involved, and try a small version in context before expanding it.

Conceptual illustration generated for this presentation in the established figurative gouache style; not a documentary image or quantitative result. Adapted from Self-Introduction.pptx.

BuildingDigitalDemocracyin Practice

Joshua C. Yang

What can democracy look like?

How groups use what people know

Different knowledge

Independent views

Structured exchange

Four research directions

Computational social choice

Deliberative democracy

Human–AI interaction

Complexity science & collective intelligence

Democracy as an open-source project

United Kingdom

United States · 1787

Switzerland · 1848

Australia · 1901

Five parts of a collective decision

Framing

Ideas

Deliberation

Voting

Accountability

What is Kultur Komitee?

Framing

Who decides, and what can they decide?

What happensbefore yourtool exists?

Ideas

Project information for participants

Deliberation

Small-group discussion and assessment

Citizens’ assemblies study an issue before making recommendations

Murmi

Murmi helps people check a discussion summary

People share experiences and concerns.

What makes an AI summary fair?

Draft summary

Whoseperspectivedid yoursystem lose?

Voting

Equal for whom?

The same votes can give different winners

Same nine voters.

Aarau’s first participatory budget

Design the vote, the rule and the explanation

Express preferences

Allocate resources

Explain the result

How votes become a funding result

Each person assigns seven points

Creative Club costs CHF 1,000

Auto Ausstellung is skipped

Four projects share CHF 3,000

Accountability

Presenting the funding result

Kultur Komitee voting receipts

A receipt shows how your share helped fund each project

Förderpool Kultur

Who gets tochangethe rules?

Explore possible interactions

Deliberation agents

Personas & digital twins

Human-controlled assistants

Shape the places we share

OmaStadi ↗

VelObserver ↗

Make public information easier to use

@gora ↗

ZüriCityGPT ↗

Democracy is still being built!

Would you or others actually use it?

What existing process gets better?

What must you learn before building?

Process examples, project demos and readings

Democracy as an open-source project: historical sources

Existing contexts and possible contributions

Research directions and selected papers

Project demos

Building
Digital
Democracy
in Practice

What happens
before your
tool exists?

Whose
perspective
did your
system lose?

Who gets to
change
the rules?