Democracy must become better informed, not more digital.
AI opens two paths: deeper participation through Citizen Assemblies and AI policy augmentation, or algorithmic concentration and manipulation. Which path prevails depends on whether democracies act proactively now or chase events reactively. Switzerland, with its direct democracy, is the natural test laboratory.
Where do things stand today, and where do they break?
Democracies today face double pressure. First: platforms such as X, Facebook, and TikTok have become de facto forums for political opinion formation, algorithmically optimised for engagement rather than deliberation. Polarisation is rising empirically (Pew Research, V-Dem Index). Second: complex topics (AI regulation, climate policy, digital sovereignty) exceed the cognitive bandwidth of classical parliamentary processes. Citizen Assemblies (Ireland, France, UK) are pilot responses, but not yet mainstream. AI-augmented policy is discussed conceptually (Stanford Center for AI Policy) but is not institutionally embedded. Switzerland's direct democracy provides an institutional corrective that protects against capture by vested interests. It does not protect against being overwhelmed by complexity.
If AI over the next 5–10 years Information environment dominates (Foundation Models as the default source of information, AI-generated content accounting for the majority), the question becomes acute, where citizens form their perception of reality. Legacy media is losing reach, platforms are optimising for attention, and AI models carry structural bias. Unless democracies actively address this gap, the result may be an erosion of trust severe enough to constitute a structural break — as seen in Poland and Hungary between 2015 and 2025. At the same time, the complexity of concrete policy (AI standards, sovereign compute, genetics, climate) has grown to the point where parliamentary majorities can rarely reach coherent decisions without structured preparation.
Paths out of the crisis
The stars show how realistic a path is: 1 star = barely feasible, 5 stars = highly feasible.
Institutionalising Citizen Assemblies
Randomly selected citizens work alongside experts over weeks or months on complex issues, producing structured recommendations.
+ Robustly deliberative; broadly representative.
Time commitment, lack of binding structure.
Trade-off: Depth vs. speed.
Feasibility: Ireland 2016 (abortion question), France 2019 (Gilets Jaunes crisis and the subsequent Citizens' Convention on Climate), and the UK 2020 Climate Assembly serve as models. Switzerland already has a precursor element in the Vernehmlassung consultation process. This instrument is politically broad-based and culturally resonant: it is the most direct lever.
AI-augmented Policy / Sense-Making
AI systems aggregate data, identify conflicting objectives, and simulate policy consequences. A tool for policymakers and citizens alike.
+ Scales with complexity.
Black-box risk, bias.
Trade-off: Efficiency vs. explainability.
Feasibility: Stanford CISAC and Anthropic Constitutional AI as proof-of-concept. Crucially: methodologically rigorous, transparent, with ultimate human accountability. Switzerland's tradition of data protection places it in a strong position.
Algorithmic Accountability Act
Binding requirements on explainability, audit and liability for algorithmic systems used in public administration and on digital platforms.
+ Guards against black-box governance.
Implementation costs; potential brake on innovation.
Trade-off: Security vs. speed.
Feasibility: The EU AI Act (in force since 2024), US state-level legislation (New York City, California), and the Algorithmic Accountability Bill (under discussion at the federal level in the US) set the frame. Switzerland urgently needs comparable legislation. It is the institutional foundation for everything else.
Liquid democracy / delegation-based voting rights
Citizens can delegate their voting rights on specific topics to experts or representatives. This is more flexible than parliamentary representation.
+ High participation, subject-matter expertise.
Concentration risks, technical complexity.
Trade-off: Flexibility vs. stability.
Feasibility: Pirate Party experiments (Germany, Iceland) and Berkeley studies provide the theoretical foundation. Practical scaling has largely failed to date. The concept is appealing, but for the Swiss system — where direct democracy is already participatory — it is more additive than transformative.
Public-service AI as a commons
Foundation models and AI tools for public administration and public advisory services as a digital commons: open source, privacy-preserving.
+ Democratises AI access in public administration.
Compute costs, maintenance burden.
Trade-off: Sovereignty vs. efficiency.
Feasibility: EU EuroHPC, Mistral and BLOOM point the way. Switzerland, with CSCS Lugano, has the compute base. Realistic for public administration and education; difficult to sustain as a universal solution.
Trigger-based phases
Trigger: A federal Citizens' Assembly on AI regulation is convened.
2026–2028. First federal Citizens' Assembly comprising 100–150 randomly selected citizens, eight to ten weeks of deliberation, and clear recommendations to parliament. In parallel: a Federal Council preliminary dispatch on algorithmic accountability.
Trigger: Swiss Algorithmic Accountability Act enters into force.
2028–2032. Binding explainability, audit and liability requirements for algorithmic systems. Citizens' assemblies become a standard prerequisite for complex ballot measures. A public-service AI platform hosted by CSCS/ETH/EPFL.
Trigger: AI tools for analysing public consultations and simulating policy impacts embedded in the federal administration.
2032–2036. The federal administration uses AI for sense-making, policy simulation and stakeholder mapping. Consultation processes are deliberatively enhanced. An EU–Swiss public-sector AI alliance takes shape. Swiss GovTech emerges as a European champion.
Trigger: More than eight EU member states adopt the Swiss model of Citizen Assemblies, Algorithmic Accountability and Public-Service AI.
2036 and beyond. Swiss democratic innovation finds resonance across Europe. Democratic erosion has been halted; deliberative elements have entered the mainstream. AI governance is transparent and auditable.
Real-world pilot programmes
Where the ideas behind this pillar are already being tested. "Learn more" opens the explanation without leaving the page.
What we are watching
Citizen Assemblies as an institutionalised component. Threshold 2–3 marks the point of normalisation.
Institutional foundation for AI governance. Threshold: legislation in force, audit practice in place.
Track in the index →AI as the standard analytical tool. The 70% threshold marks the point of standardisation.
Democratic substance is maintained. Threshold: no structural erosion — ideally improvement through deliberative elements.
Our assessment
At Globalance, we regard Switzerland as the natural testing ground for democratic innovation. in the economy ahead. Direct democracy is not the problem but part of the solution: when it is extended with deliberative elements and complemented by AI-enhanced sense-making tools. What is not needed: technocratic substitution of democratic decisions by AI. What is needed: better preparation of democratic decisions through structured deliberation and transparent AI tools.
Citizen assemblies should be institutionalised, as a binding prior step for popular initiatives of substantial complexity (AI regulation, climate policy, social policy). The Swiss consultation procedure is a precursor; what is missing are the representative citizen sample and the structured deliberation processIreland and France have demonstrated that this works. National citizens' assemblies addressing two to three key topics per legislative term are both realistic and effective.
AI-augmented Policy does not mean that AI takes political decisions. It means that AI helps Anticipating consequences, making trade-offs visible, Aggregating dataStanford's CISAC, Anthropic's Constitutional AI, and the EU's AI4Public initiatives demonstrate what is possible. The Swiss approach: AI tools for the federal administration and the analysis of consultation responses, with a clear obligation to explain decisions and ultimate human accountability.
A Swiss algorithmic accountability law is long overdue. The EU has set the pace with the AI Act; Switzerland is lagging behind. We need binding requirements for explainability, auditing and liability for algorithmic systems used in public administration, the judiciary and platforms. This is not hostile to innovation — it is a prerequisite for deploying AI in areas subject to democratic accountability. A Federal Council message in 2027 is a realistic target.
Liquid Democracy we consider intellectually elegant, but not of decisive importance to the Swiss context. Direct democracy is already participatory; liquid democracy would function as a complement, though probably not as the primary mechanism. Pilot experiments in cantons or municipalities are of interest, but should not displace the focus on Citizen Assemblies and algorithmic accountability.
Public-service AI as a commons is the infrastructural foundation. EU EuroHPC, Mistral, and the Swiss CSCS in Lugano are the right approaches. What is missing is a systematic Swiss strategy to bring federal and cantonal administrations onto sovereign, transparent AI stacks. A European alliance for AI in public administration, aligned with EuroHPC and with Swiss participation through the CSCS and ETH, would be a conceivable way forward.
What this means for investors: First, rules on algorithms and their explainability are coming; the only open question is when. Companies that build auditable systems today carry the smaller regulatory risk. Second, the public sector is becoming a major customer for trustworthy AI. Providers with European data locations and open standards are better placed for that. Third, stable, predictable institutions are a competitive advantage — including for capital; political polarisation raises the risk premium. Fourth, disinformation is a business risk for platforms whose revenues depend on reach.
Why now: AI disruption is accelerating, and so is democratic erosion. Switzerland has a window of three to five years to position itself as a European model for democratic innovation in the economy ahead. Those who shape this window set the rules. Those who do not will inherit them from Brussels or Washington.
The strongest objection concerns the legitimacy of citizens' assemblies themselves. A randomly selected assembly has no mandate; it has chance. If its recommendation effectively shapes the ballot proposal, influence shifts from elected bodies to a process that nobody can vote out. There is a further issue: whoever frames the question and selects the experts determines the corridor of possible answers. The process is therefore no more neutral than parliament, only less visible in how it is steered.
What we do not yet know
- How do we bind citizen assembly recommendations to parliamentary processes without undermining direct democracy?
- How do we prevent AI-augmented policy from becoming a technocratic substitute for democratic decision-making?
- Should the public-service AI platform be operated by the state, a non-profit, or a public–private alliance?
- How do we regulate foundation models that are developed internationally yet shape decisions within individual jurisdictions?
Connected to: 07 · Data & AI, 04 · Labour, 10 · Meaning
Scenarios: Geopolitics & Governance · Society & Psychology · Technology & AI
