Co-initiator of Public AI Switzerland, Anna Sotnikova presented the National AI Dialogue, which invites the public to shape Apertus, the Swiss open model.

1,600 Voices and Counting
Participants upvote questions, propose their own and see others' answers. Retention exceeded 60%, far beyond expectations. Early results show strong support for Swiss AI, data stored in Switzerland and shared governance. A large majority wants a tool rather than a companion.

A Case for AI Literacy
Bias appeared in only about 17% of concerns, and around 40% saw harmful outputs as the user's responsibility. "That's a signal for us that there should be AI literacy."

Designing for Agency
Unlike a fixed survey, the dialogue lets people add their own answer options, submit questions and post public quotes. The hardest part was writing questions for everyone: some participants found them too technical. Coverage is uneven, strong in Zurich and German-speaking Switzerland and weaker in Romandy, so the team is reaching out canton by canton. A native speaker translated the dialogue into Romansh by hand.

Constitutions Need Concrete Cases
Like most models, Apertus has a constitution that is still being shaped. "When you write a constitution, AI responses should be fair, transparent. But what does that mean in practice?" Results are never applied by majority vote. They are reported as distributions to policymakers or used for soft-label training.

Fair Data Has a Price
Training only on properly licensed data, and removing those who opt out, shrinks the corpus and costs performance. Her answer is task-specific models: what matters is identifying the use cases the Swiss public cares about.

Results feed the Apertus constitution, developer feedback and a report for the Geneva AI Summit.