Popular Votes and Algorithms in Switzerland: Intransparent Priorisation of Political Information

Whitepaper English

Whitepaper German

Whitepaper French

AI-driven systems – such as web search engines and chatbots – are increasingly used by Swiss citizens to find information on political issues, for example in the context of popular votes and elections. This creates both new possibilities, but also risks for the functionality of Swiss democracy. A key risk is related to the systems' tendency to individualize the information delivery by providing different results to different users on the basis of location, the language of the user input, or the formulation of the query. In the case of a multilingual direct democracy, such treatment can result in information inequalities between regions and languages that have to be taken seriously not least also in the light of other known risks such as societal polarization or the spread of misinformation.

To illustrate the risks caused by such AI-driven information inequalities, this whitepaper presents as a case study the recent popular vote on the «climate protection law» in June 2023. An analysis of data from Google – the search engine used by 86% of the Swiss population – shows that depending on the language (i.e. German, Italian, or French) and the formulation of the search query, Google prioritizes different information sources and different viewpoints on the vote. Queries in different languages resulted in unequal visibility of information sources in the top results, including websites of Swiss political parties advocating for or against the law. This has consequences in two dimensions:

• It may result in advantaging or disadvantaging specific parties or politicians in relation to particular political issues – without, however, any transparency regarding how these AI-driven systems prioritize information for potential voters. This can facilitate the manipulation of the Swiss public sphere by foreign and domestic actors.
• It may interfere with Swiss citizens receiving a fair representation of the range of political positions on specifi c political issues. Particularly in the context of Switzerland’s direct democracy, such fair representation is crucial for participating in political decision-making, in particular as AI-driven systems increasingly become a prominent source of information.

By providing unequal access to information about politics, AI-driven systems can infl uence the ability of different groups of citizens to participate in popular votes and elections and, thus, challenge the functionality of Swiss democracy.

• they can exacerbate information disparities among citizens;
• they can further restrict the diversity of media and political opinions;
• they can influence the agenda-setting power of parties and politicians;
• they can can expose users to erroneous or misleading information.

The authors of the whitepaper therefore propose:

• Identifying pathways for the long-term integration of AI-driven systems into the functionality of Swiss democracy and defining a common framework for AI-driven systems in the context of political decision-making, involving all relevant stakeholders.
• Defining Key Performance Indicators (KPIs) to measure the impact of AI-driven systems in the context of Swiss politics, which support the framework.
• Establishing infrastructure for consistent monitoring of the impact of AI-driven systems in the political decision-making process in Switzerland. This creates transparency in society about the influence of AI-driven systems on the formation of citizens' opinions and thus also promotes public awareness of this issue.
• Conducting more empirical studies on how AI-driven systems address political issues in Switzerland and how different groups of Swiss citizens use these systems. There is a need for further research to understand how citizens use AI-driven systems, especially concerning different demographic characteristics such as age or place of residence.

The basis for this whitepaper consisted of stakeholder interviews and a "Franxini Fireside Chat" with representatives from academia, politics, administration, media, and civil society as part of the Innovation Hub project "AI-driven information inequalities in Switzerland in the context of popular votes" by the Reatch association. The writing of the whitepaper and the development of concrete solutions were the responsibility and decision-making authority of the participating authors under the leadership of Mykola Makhortykh.