# Civic technology: processes, projects and research Reviewed 6 September 2026. The talk has 28 scenes and 28 planned minutes, including interaction, plus two minutes of buffer. ## Framing The opening voting-rule comparison demonstrates how procedure can change a result. One branching history diagram connects inherited institutions and reforms, followed by everyday and civic examples, collective intelligence and parallel research fields. A short Kultur Komitee introduction explains the setting and Joshua’s role before the first step. The main section follows five steps: define the decision, prepare proposals, discuss the options, make the decision, and explain and follow up. Projects illustrate each step, with a matching number and label on every example. The closing asks students to explore a project or process, research its context and consider different possibilities before deciding what to build. Collective intelligence asks how groups combine knowledge and capabilities. Democratic theory also concerns inclusion, legitimate authority and the right to contest a result. See [Landemore’s chapter on inclusive deliberation](https://doi.org/10.23943/princeton/9780691155654.003.0004). ## Democracy as an open-source project The opening adapts **Motivation 1, slide 3 of Joshua’s `Self-Introduction.pptx`**: “Democracy is alive and can be designed (better).” Its raster branching diagram is redrawn in native HTML/SVG with the talk’s existing fonts, warm colors and paper texture. One diagram combines institutional influences with reforms along each country’s path. The dates identify events; horizontal spacing is schematic. British institutional precedents inform the US branch; the US federal model informs the Swiss and Australian branches. Australia’s entry also states its direct British parliamentary influence. The original file is unchanged. The analogy is that people can inspect, borrow and revise institutional rules. This is not a complete family tree, a literal software-development process or an inevitable progression toward better democracy. Arrows represent selected documented influences. Political rights and the power to change rules have historically been unequal; reforms involve interests, legitimacy and tradeoffs. - The 1787 constitutional compromise combined a House apportioned by population with two senators per state. [US Senate](https://www.senate.gov/about/origins-foundations/senate-and-constitution.htm) - The 1848 Swiss Constitution established a bicameral federal parliament, using the United States as a model. [Swiss Parliament: 1848](https://www.parlament.ch/de/%C3%BCber-das-parlament/Seiten/Ablauf-Wahlen-1848.aspx) - Australia combines parliamentary government responsible to the elected house, drawn from Westminster, with a federal Senate inspired by the United States. [Australian Parliament](https://peo.gov.au/understand-our-parliament/your-questions-on-notice/questions/what-is-the-washminster-system) - The Australian Constitution took effect with federation on 1 January 1901. [Australian Constitution](https://peo.gov.au/understand-our-parliament/how-parliament-works/the-australian-constitution/the-australian-constitution-in-focus) - The Parliament Acts of 1911 and 1949 permit certain public bills to pass without Lords consent. The 1949 Act reduced the delay for eligible non-money bills from two years to one. Money bills and other categories have different provisions; the Lords did not lose every veto. [UK Parliament](https://commonslibrary.parliament.uk/research-briefings/sn00675/) - The optional federal legislative referendum dates to 1874. The mandatory constitutional referendum already existed in 1848. [Swiss Parliament: referendums](https://www.parlament.ch/de/%C3%BCber-das-parlament/Seiten/faktenblatt-referenden.aspx) - The federal popular initiative for partial constitutional revision dates to 1891. The initiative for total revision predates this, going back to 1848. [Swiss Parliament: initiatives](https://www.parlament.ch/de/%C3%BCber-das-parlament/Seiten/faktenblatt-volksinitiativen.aspx) - Preferential voting for the House of Representatives was introduced in 1918; compulsory voting at federal elections in 1924. Senate proportional representation was enacted in 1948 and first used in 1949, not at federation in 1901. [Australian Electoral Commission](https://www.aec.gov.au/Elections/history-of-electoral-reform.htm) - Political participation was historically restricted. Swiss women gained federal voting rights in 1971. [Swiss Federal Archives](https://www.bar.admin.ch/en/the-swiss-state-and-its-citizens-after-1848) The bridge to building is concrete: examine an existing process, identify a difficult task or rule, and assess what a change would do for the people involved. The opening voting-rule demonstration holds ballots fixed to isolate the effect of changing a rule. - The US framers drew on British institutions and state constitutions. The connection represents selected institutional precedents, not a literal fork of modern Westminster government. [US Senate: British and state precedents](https://www.senate.gov/about/origins-foundations/senate-and-constitution.htm) - Proportional election of the National Council was approved in 1918 and first used in 1919. [Swiss Parliament: proportional National Council elections](https://www.parlament.ch/fr/%C3%BCber-das-parlament/Pages/Vom-Majorz-zum-Proporz.aspx) ## Kultur Komitee: setting and Joshua’s role Kultur Komitee is a residents’ funding assembly for art and cultural projects in Winterthur. Joshua has worked on voting and aggregation across the 2024, 2025 and 2026 editions, alongside Fynn Bachmann on calculations and aggregation. Joshua confirmed **over CHF 2 million distributed across these three years** on 6 September 2026; this supersedes the older CHF 1.2 million repository figure. It is not derived by multiplying the 2025 deployment figure. The introductory photograph is the supplied `images/IMG_2826.JPG`, copied unchanged. ## Illustrated topic pages Five large illustrated topic pages introduce the steps before the real fieldwork examples. Each contains a step number, its name and one question. They use the supplied artwork unchanged and retain its proportions. The topic pages share time with the following photographs, so the talk remains 28 planned minutes. The photographs keep their evidence and captions; repeated questions move into the notes. The overview and collective-intelligence page use less projected text. The closing keeps three open prompts beside the prototype illustration, with no suggested project domains or solutions. The illustrations are conceptual; the fieldwork photographs and published sources establish the actual project context. Original artwork files remain unchanged. ## Fieldwork examples within the five steps Five illustrated topic pages mark the sections after the introductory context. Each is followed by a Kultur Komitee fieldwork photograph supplied by Joshua from `images/`. Related projects and civic cases follow the relevant photograph. Each photo asks the audience to consider a task in a real budget assembly: | Step | What the group needs to do | Audience reflection | | ---------------------- | ------------------------------------------------------------------------- | ------------------------------------------------------------------- | | 1. Define the decision | Establish the decision, participants and available resources. | Who should take part, and what should they be able to decide? | | 2. Prepare proposals | Provide project descriptions, costs and information people can compare. | What would you need before comparing two proposals? | | 3. Discuss the options | Examine reasons and assess projects against shared criteria. | How would you preserve a minority concern? | | 4. Make the decision | Combine individual preferences and group evaluations under a budget rule. | How should the rule balance popular projects with other priorities? | | 5. Explain & follow up | Communicate the allocation and make its consequences inspectable. | What would you check if your preferred project was not funded? | The case reports 38 randomly selected residents allocating CHF 380,000 to 43 projects in 2025. These figures come from the [Komitee study](https://joshuacyang.com/pubs/komitee-equal-shares/), not a count inferred from a photo. The photographs show a plenary, project material, table discussion, individual assessment and a projected allocation table. They do not establish equal influence or completed implementation. The five steps are a teaching walkthrough, with work that can overlap or repeat. The later receipt demonstration uses separate published 2026 KK26 data. The five-step photo navigation remains visible in the PDF, so every step has its own page. Original photographs are preserved; web copies are resized and compressed, with provenance and checksums recorded in `assets/media/provenance.json`. ## Aggregation: the Komitee handout Step 4 now uses the supplied `images/KKW_Infokarten_Grafiken_platziert_MES.pdf`, **Method of Equal Shares**, page 1. The slide retains the complete source graphic and offers four views: votes, the first funded project, a budget limit, and the final allocation. The original PDF is bundled unchanged; an English walkthrough makes the numbers readable on the projector. The handout explicitly labels its data as invented and simplified. Three participants each have seven points (at most two per project) and CHF 1,000. The walkthrough follows its printed order, including ties, and charges each supporter in proportion to their points. It does not implement a general MES algorithm or reproduce the 2025/2026 results. The group impact assessments of Komitee Equal Shares are not shown in this example. | Project, in the handout’s order | Cost / CHF | Anna | Miro | Jasmin | Points | Result | | ------------------------------- | ---------- | ---- | ---- | ------ | ------ | --------------------------------------------------------- | | Creative Club | 1,000 | 2 | 1 | 2 | 5 | Funded: 400 + 200 + 400 | | Experimental Event | 800 | 2 | 2 | 0 | 4 | Funded: 400 + 400 + 0 | | Auto Ausstellung | 1,200 | 2 | 1 | 1 | 4 | Skipped: Anna has 200 but needs 600 | | Dance Disco | 600 | 1 | 2 | 0 | 3 | Funded: 200 + 400 + 0 | | Folklore Fest | 1,800 | 0 | 1 | 2 | 3 | Skipped: Miro and Jasmin cannot cover their contributions | | Bilder Buch | 600 | 0 | 0 | 2 | 2 | Funded: 0 + 0 + 600 | The four funded projects total CHF 3,000, and each person spends CHF 1,000. Total points alone do not determine the outcome: remaining budgets affect which supporters can still fund projects. This connects the assessment and voting stage to the next step, where participants need an explanation of the result. Source: [Original Komitee handout](assets/data/komitee-aggregation-handout.pdf). [Transcribed data](assets/data/aggregation.json) records the page and source checksum. In the conference workspace, the original lives in `images/`; these relative links resolve in the bundled web directory. Formal background remains [Peters et al., NeurIPS 2021](https://proceedings.neurips.cc/paper_files/paper/2021/hash/69f8ea31de0c00502b2ae571fbab1f95-Abstract.html) and the [Komitee field study](https://doi.org/10.1145/3772363.3798958). ## Existing processes and prototype ideas ### Swiss direct democracy **Process:** A federal popular initiative proposes a constitutional change; an optional referendum challenges legislation. **A concrete difficulty:** A voter or organizer needs to understand the text and the published procedural status. **Possible prototype:** A comparison of the proposal with the status quo, or a tracker using official records. **Evaluation:** Can a reader find the relevant passage and explain what is being proposed? Sources: [Popular initiatives](https://www.ch.ch/en/votes-and-elections/popular-initiative/), [Referendums](https://www.ch.ch/en/votes-and-elections/referendum/). ### ETH student representation **Process:** Fachvereine and VSETH represent students in departmental and ETH-wide bodies; VSETH’s members’ council governs its own budget, elections and statutes. **A concrete difficulty:** A student needs to identify who can act and what has already been discussed. **Possible prototype:** An agenda brief or issue record connecting a concern, evidence, the responsible body and its response. **Evaluation:** Can another student identify the responsible body and the present status? Sources: [VSETH politics](https://vseth.ethz.ch/politik/), [VSETH structure](https://vseth.ethz.ch/struktur/). ### Citizens’ assemblies and consultations **Process:** Participants examine evidence, discuss alternatives and produce recommendations or consultation outputs. **A concrete difficulty:** An important concern can disappear when discussions are summarized. **Possible prototype:** A source-linked correction or unresolved-question record. **Evaluation:** Compare the account with original contributions; count unsupported claims and missing requirements. Sources: [Irish citizens’ assemblies](https://citizensassembly.ie/about/faq/), [vTaiwan](https://info.vtaiwan.tw/), [Murmi project](https://joshuacyang.com/projects/murmi/). ### Participatory budgeting **Process:** Develop proposals, assess feasibility and allocate a limited budget using a defined procedure. **A concrete difficulty:** A proposer needs to meet requirements; a participant needs to understand how the allocation was produced. **Possible prototype:** A feasibility checklist, comparison of rules, or readable allocation receipt. **Evaluation:** Hold ballots and costs fixed for a rule comparison; ask a participant to explain an allocation. Sources: [Decide Madrid process](https://decide.madrid.es/mas-informacion/dm/presupuestos), [Komitee Equal Shares](https://joshuacyang.com/pubs/komitee-equal-shares/), [Public KK26 receipts](https://equalshares.kulturkomitee.win/). ### Accountability after a decision **Process:** Connect an adopted proposal to responsibility, implementation and public reporting. **A concrete difficulty:** A resident can see that a proposal was approved but cannot tell what happened next. **Possible prototype:** A proposal-to-implementation record using existing milestones and evidence. **Evaluation:** Can a reader identify the current status, responsible body and supporting record? Sources: [Decidim accountability component](https://docs.decidim.org/en/develop/admin/components/accountability.html). ## Parallel research directions This is a selected reading guide, not a chronology or exhaustive map. Recent topics include AI-mediated deliberation, allocation under feasibility constraints, and grounded or inspectable agents. ### Computational social choice Voting rules, proportional allocation and fair selection of representatives. - [Peters, Pierczyński & Skowron (2021). Proportional Participatory Budgeting with Additive Utilities](https://proceedings.neurips.cc/paper_files/paper/2021/hash/69f8ea31de0c00502b2ae571fbab1f95-Abstract.html). _NeurIPS_. **Peer-reviewed mathematical paper.** Develops proportionality axioms and allocation rules for projects with different costs and additive utilities; a foundation for the Method of Equal Shares. **Limit:** A guarantee depends on its preference model and precise axiom. It does not make every outcome fair under every definition. - [Flanigan, Gölz, Gupta, Hennig & Procaccia (2021). Fair algorithms for selecting citizens’ assemblies](https://www.nature.com/articles/s41586-021-03788-6). _Nature_. **Peer-reviewed algorithms and deployment study.** Optimizes selection probabilities among volunteers while meeting demographic quotas; evaluates fairness on data from ten assemblies. **Limit:** Fairness within the volunteer pool does not solve unequal willingness or ability to volunteer. - [Skowron (2026). A General Theory of Proportionality with Additive Utilities](https://arxiv.org/abs/2602.08504v1). _arXiv:2602.08504_. **Preprint · February 2026.** Proposes proportional rules for scored preferences under general feasibility constraints, and proportional rankings. **Limit:** A theoretical proposal, not a deployment evaluation. Constraints and the definition of proportionality must be specified. ### Deliberative democracy Inclusion, reason-giving, public judgment and legitimate authority. - [Woolley et al. (2010). Evidence for a Collective Intelligence Factor in the Performance of Human Groups](https://doi.org/10.1126/science.1193147). _Science_. **Peer-reviewed empirical study.** Two studies with 699 people found a shared factor in small groups’ performance across tasks, associated with social sensitivity and conversational turn-taking. **Limit:** These associations do not show that equal speaking time causes better decisions, or measure democratic legitimacy. - [Lorenz, Rauhut, Schweitzer & Helbing (2011). How social influence can undermine the wisdom of crowd effect](https://pmc.ncbi.nlm.nih.gov/articles/PMC3107299/). _PNAS_. **Peer-reviewed laboratory experiment.** In numerical estimation tasks with 144 participants, seeing others’ estimates reduced diversity and increased confidence without a corresponding improvement in collective accuracy. **Limit:** An estimation experiment is not a general verdict against discussion; task and interaction structure matter. ### Human–AI interaction How people use, question and correct AI assistance during participation. - [Tessler et al. (2024). AI can help humans find common ground in democratic deliberation](https://doi.org/10.1126/science.adq2852). _Science_. **Peer-reviewed experiments.** The Habermas Machine uses opinions and critiques to revise group statements. Across experiments with 5,734 participants, AI statements were preferred to those from human mediators. **Limit:** Preference for a statement is not proof of factual accuracy, legitimate authority, or lasting political agreement. - [Yang, Hausladen, Peters, Pournaras, Hänggli Fricker & Helbing (2024). Designing Digital Voting Systems for Citizens: Achieving Fairness and Legitimacy in Participatory Budgeting](https://doi.org/10.1145/3665332). _Digital Government: Research and Practice_. **Peer-reviewed empirical article.** Studies how voting-system design shapes citizens’ perceptions of fairness and legitimacy in participatory budgeting. **Limit:** Perceived fairness and actual understanding are different outcomes; evaluate them separately. ### Complexity science & collective intelligence - [Woolley et al. (2010). Evidence for a Collective Intelligence Factor in the Performance of Human Groups](https://doi.org/10.1126/science.1193147). _Science_. **Peer-reviewed empirical study.** Two studies with 699 people found a shared factor in small groups’ performance across tasks, associated with social sensitivity and conversational turn-taking. **Limit:** These associations do not show that equal speaking time causes better decisions, or measure democratic legitimacy. - [Lorenz, Rauhut, Schweitzer & Helbing (2011). How social influence can undermine the wisdom of crowd effect](https://pmc.ncbi.nlm.nih.gov/articles/PMC3107299/). _PNAS_. **Peer-reviewed laboratory experiment.** In numerical estimation tasks with 144 participants, seeing others’ estimates reduced diversity and increased confidence without a corresponding improvement in collective accuracy. **Limit:** An estimation experiment is not a general verdict against discussion; task and interaction structure matter. The following agent studies are optional further reading; the talk does not include a Belief Engine demonstration. How collective learning and judgment emerge from interactions, networks and feedback. This differs from computational social choice’s study of formal decision rules. Computational social science provides methods that can serve several of these research directions. - [Park et al. (2026 revision). LLM Agents Grounded in Self-Reports Enable General-Purpose Simulation of Individuals](https://arxiv.org/abs/2411.10109v3). _arXiv:2411.10109v3_. **Preprint · June 2026 revision of a 2024 paper.** Builds agents from interviews and/or surveys of 1,052 Americans and evaluates predictions on held-out responses against people’s own test–retest consistency. **Limit:** Predictive fidelity on studied tasks does not confer consent or political authority. The earlier title was Generative Agent Simulations of 1,000 People. - [Yang, Flechtner, Dailisan & Bakker (2026). Belief Engine: Configurable and Inspectable Stance Dynamics in Multi-Agent LLM Deliberation](https://arxiv.org/abs/2605.15343). _arXiv:2605.15343_. **Preprint · May 2026.** Maintains explicit proposition-level stances with configurable evidence uptake and prior anchoring, recording the evidence associated with updates. **Limit:** Evidence scoring and update rules are modeling choices. An inspectable trace is not by itself a faithful simulation of a citizen. ## Evidence and evaluation The rule comparison uses nine invented ballots. The Polis schematic uses invented groups and is not a clustering analysis of vTaiwan. The summary exercise and agent trace are illustrations. Swiss AI Futures photographs are real public project media. The Kultur Komitee examples use user-provided fieldwork photographs. The 2025 Komitee deployment figures and 2026 KK26 receipt are separate evidence. A few usability sessions can reveal difficulties. They do not establish a population effect. A theorem gives a guarantee under assumptions; an experiment compares outcomes under a study design; a deployment documents a setting. Keep those claims distinct. For a hackathon, identify a person’s task, the existing workflow and a specific difficulty. Build one support tool and compare it with what that person uses today. An uncomplicated idea can be useful when it fits the task. ## Project pages in the presentation - [Murmi](https://www.murmi.org/#in-action): the public interface showcase. - [Kultur Komitee voting receipts](https://equalshares.kulturkomitee.win/): personal receipts, projects and results. The funding illustration uses the supplied Equal Shares PDF and its local walkthrough. Live embeds require internet access; Offline view and the PDF keep the local explanations. ## Features highlighted in the historical branches The short labels identify one feature of each branch, not an exclusive national invention or a complete description of the system. - **United Kingdom — Parliamentary oversight:** scrutinizing government and requiring public explanations. [UK Parliament](https://www.parliament.uk/about/how/role/scrutiny/). - **United States — Checks & balances:** dividing authority so branches can constrain one another. [US government](https://www.usa.gov/branches-of-government). - **Switzerland — Direct democracy:** citizen routes to challenge laws and propose constitutional changes. [Swiss government](https://www.aboutswitzerland.eda.admin.ch/en/direct-democracy). - **Australia — Compulsory voting:** addressing low participation; the 1924 reform aimed to increase turnout. [Australian Electoral Commission](https://www.aec.gov.au/About_AEC/Publications/backgrounders/compulsory-voting.htm). These are design purposes, not guarantees of outcomes. The dated reforms remain on the branches. ## Additional readings connected to the four directions Representation, consent, power and responsibility in civic systems. - [Selbst, boyd, Friedler, Venkatasubramanian & Vertesi (2019). Fairness and Abstraction in Sociotechnical Systems](https://doi.org/10.1145/3287560.3287598). \*FAT\*\*. **Peer-reviewed conceptual paper.** Explains how isolating a technical model from its social setting can undermine fairness goals. **Limit:** A conceptual analysis of system design, not an empirical guarantee that a particular civic tool is fair. - [Bisbee, Clinton, Dorff, Kenkel & Larson (2024). Synthetic Replacements for Human Survey Data? The Perils of Large Language Models](https://doi.org/10.1017/pan.2024.5). _Political Analysis_. **Peer-reviewed empirical audit.** Synthetic responses could match survey averages while understating variation and changing statistical relationships. Results were sensitive to prompts and collection dates. **Limit:** The study tests particular models and tasks; use its evaluation questions rather than assuming identical failures for every later model. How digital tools fit institutions, workflows and public accountability. - [Small, Bjorkegren, Erkkilä, Shaw & Megill (2021). Polis: Scaling Deliberation by Mapping High Dimensional Opinion Spaces](https://www.e-revistes.uji.es/index.php/recerca/article/view/5516). _Recerca_. **Peer-reviewed system and case study.** Describes a system that maps agreement patterns and surfaces statements supported across opinion groups, with real consultation cases. **Limit:** A useful map of participants is not a representative estimate of an entire population or evidence that consensus is always desirable. - [Yang, Scheurer & Helbing (2026). Komitee Equal Shares: A Hybrid Algorithm for Priceable Participatory Funding](https://doi.org/10.1145/3772363.3798958). _CHI Extended Abstracts_. **Published conference extended abstract · field deployment.** Combines personal preferences with group evaluations and allocation receipts. In Winterthur in 2025, 38 citizens allocated CHF 380,000 to 43 projects. **Limit:** One deployment, with 32 survey respondents. Simulated comparisons do not isolate every component’s causal contribution. ## Revised presentation flow A short introduction to Joshua follows the title. Everyday decisions lead to collective intelligence and four research fields. The historical lineage and five-part framework follow. Kultur Komitee is introduced on slide 8. The framework is Framing, Ideas, Deliberation, Voting and Accountability. Swiss AI Futures and the initiative/referendum comparison are not projected. The aggregation example uses the supplied handout; the live dashboard is not embedded. Murmi and voting receipts remain embedded. ## September 6 revision: stages and AI research 28 scenes / 28 minutes plus 2 minutes buffer. ETH introduction follows Josh’s supplied biography. Each stage offers two possible use cases. The projected Polis example is removed. Fair summaries introduce [generative social choice](https://proceedings.mlr.press/v267/boehmer25a.html). A separate +1 section covers AI agents, personas, digital twins and [Belief Engine](https://arxiv.org/abs/2605.15343); simulations inform research while people retain decision authority. ## Emerging projects and illustration update 29 scenes, 28 minutes plus buffer. The penultimate scene pairs Talk to the City, Habermas Machine and Alignment Assemblies with their public-input, group-statement and AI-governance problems. Project links and limitations are in slide notes. Five supplied illustrations now appear in the introduction, deliberation, voting, agent simulation and Belief Engine sections. Transparent paper-set assets are copied without modification; provenance is recorded. ## Current showcase: civic tech in use The penultimate slide now covers OmaStadi, Rahvaalgatus and FixMyStreet, replacing the earlier three AI projects. See [evidence and limitations](civic-tech-in-use.md). Current adoption is the selection criterion; these are established tools, not newly launched products. ## Latest slide revision 28 scenes, 28 minutes plus buffer. The dedicated Belief Engine slide is removed; the paper remains optional reading. The civic-tech slide uses three conceptual illustrations with brief captions and dated adoption figures.