Switzerland is the only established democracy in which the people regularly vote on specific issues -- quarterly, at municipal, cantonal and federal level. This system presupposes that voters understand the proposals. This is precisely where AI can make a contribution that does not replace the system but strengthens it.
Many ballot proposals are complex. The OASI reform proposal in 2024 comprised dozens of pages of legal text with cross-references to other federal laws. The official voting booklet is helpful but limited: it summarises the arguments without providing a comprehensive impact analysis.
AI-powered impact assessments could simulate every ballot proposal: What does it cost? Who is affected? Which scenarios are likely? The result would not be a recommendation -- but a comprehensible presentation of the possible consequences.
Specifically:
Such systems must be transparent. The algorithm, its data sources and its assumptions must be openly accessible. No black box that is blindly trusted must be allowed to emerge. Voters continue to decide for themselves -- but better informed than ever before [1].
An algorithm that decides which news 300 million people see makes more political decisions every day than any parliamentarian. The sorting of news feeds, the prioritisation of search results, the recommendation logic of social media -- all of this shapes public opinion [2].
All AI systems that prioritise, rank or filter content in the public information space must disclose their criteria. This means:
The EU AI Act, in force since August 2024, takes first steps in this direction with risk-based transparency obligations [3]. Switzerland can go further -- with an approach that relies on disclosure rather than prohibitions.
The second fundamental rights question concerns surveillance. In China, biometric real-time surveillance in public spaces is a reality [4]. The EU AI Act largely prohibits it, but with exceptions for law enforcement.
For Switzerland, the following should apply: No exceptions. The ban on AI-powered biometric real-time surveillance in public spaces belongs in the Federal Constitution. Every instance of biometric surveillance requires a court order -- like a house search.
Switzerland, with the revised Data Protection Act (revDPA) that has been in force since September 2023, has already created a progressive framework [5]. A constitutional article on biometric surveillance would be the logical complement.
Switzerland's direct democracy is not a relic -- it is a protective mechanism against the concentration of power that AI enables. If every change to AI regulation is subject to optional or mandatory referendum, control remains with the people.
This distinguishes Switzerland from countries where governments can enact AI regulation by decree without consulting parliament or the people. Swiss democracy is slower -- but it is more resilient against abuse.
There is also a risk. If AI systems demonstrably make "better" decisions than the electorate, pressure will emerge to replace democracy with algorithms. "AI knows better" could become the most dangerous sentence in political history.
There is only one defence against this: education (see Education and AI Literacy). Voters who understand how AI works -- and where it fails -- will not blindly rely on it.
AI can strengthen Swiss democracy if it is used as a tool for more informed decisions -- not as a substitute for human judgement. Transparency, education and constitutional safeguards are the three pillars on which this model rests.
[1] Swiss Federal Chancellery: Voting explanations of the Federal Council (voting booklet).
[2] Zuboff, Shoshana: The Age of Surveillance Capitalism. PublicAffairs, 2019.
[3] EU AI Act, Regulation EU 2024/1689, in force since 1 August 2024.
[5] Swiss Data Protection Act (revDPA), in force since 1 September 2023.