Quis custodiet ipsos custodes: Why we need to monitor AI-driven information inequalities in Switzerland in the context of popular votes

Today, our ability to stay informed about important things happening in our society (e.g. elections or healthcare crises) is dependent on information retrieval systems powered by artificial intelligence (AI). Ranging from search engines (e.g. Google) to personalised news apps (e.g. My NZZ), these systems retrieve information from their databases and rank it based on user input (e.g. search queries) and information they learn about the user (e.g. from which part of Switzerland the user searches for information). By deciding what sources to prioritise and what to downplay, AI-driven systems become information watchers who influence our decisions on whether to make a vaccination or support a popular vote. However, by doing so, these systems can not only help solve the current crises (e.g. the COVID crisis; [1]) but also cause future ones.

Despite the importance of information retrieval systems, we often do not know how they make decisions and how different the selection of information they provide us is compared with our fellow citizens. Such lack of transparency is worrisome concerning that earlier studies demonstrated how these systems can make errors (e.g. by prioritising factually incorrect information; [2]) or promote systematically distorted representation of social reality (e.g. by giving more visibility to information sources with certain political leaning; [3]). Consequently, AI behind these systems can undermine our fundamental right to receive information guaranteed by the European Charter of Human Rights [4] and challenge democratic decision-making by amplifying information inequalities which may already exist between different groups of citizens.[5]

There are several reasons why current problems with AI-driven systems can cause future crises in Switzerland. While the ability of citizens to be properly informed is essential for all types of democracies, direct democracy with the frequent rounds of popular votes is particularly dependent on citizens being able to make well-informed decisions about ongoing societal developments. Simultaneously, the multilingual composition of the Swiss Confederation increases the likelihood of information inequalities between language communities in the context of important societal developments, in particular as earlier research (e.g. [1],[6]) showed that the choice of language can result in dramatic changes in the quality of the information provided by information retrieval systems. Under these circumstances, the possible malperformance of these systems, in particular, its systematic forms (e.g. the consistent prioritisation of less reliable information sources for citizens searching in a particular language or from a specific canton) can impair the functionality of Swiss democracy and limit the possibilities of individuals for participation in the democratic decision-making.

To address this potential future crisis, it is important to increase awareness of Swiss citizens about the impact of AI-driven information retrieval systems on how they are informed about societal developments as well as provide tools for academics and the general public to monitor whether these systems create information inequalities. Because there are many different systems and many issues which they can impact, we propose to start by focusing on one specific type of system - i.e. web search engines such as Google and Bing - and its performance in the context of popular votes which are an essential component of the Swiss democratic system. For this aim, our team of communication and computer scientists from the University of Bern and the University of Zurich intends to work on computational tools for monitoring how search engines most commonly used in Switzerland (i.e. Google and Bing; [7]) select information about federal and cantonal votes and whether the quality of information varies depending on the language in which individuals search for the information about the vote and the individuals’ location. In addition to collecting and analysing data to identify whether search engines create information inequalities between different groups of Swiss citizens, our team aims to find partners to 1) make the Swiss public awareness of the importance of such monitoring; 2) find ways to make the tools sustainable and easy to use for non-academics; and 3) produce design recommendations to improve the performance of search engines and prevent them from amplifying information inequalities.

The implementation of the project will require collaboration with a number of stakeholders. To increase the awareness of Swiss citizens, it is essential to cooperate with non-governmental organisations (e.g. Algorithm Watch Switzerland) and journalists in analysing our data and discussing the implications of our findings. We also want to make policy-makers in the federal government and the parliament more aware of the potential impact of AI-driven information inequalities on Swiss democracy and stimulate the discussion about possible regulation of the growing use of AI-driven information retrieval systems in Switzerland as well as look for possibilities to attract funding to make monitoring infrastructure available and usable for the public. Finally, we want to advance our contacts with technical companies (e.g. Google) responsible for information retrieval systems to increase their awareness about the societal implications of these systems and look for ways to prevent their malperformance.

Dieser Text ist im Rahmen des Reatch Ideenwettbewerbs entstanden. Um die Krisen von morgen zu meistern, brauchen wir als Gesellschaft Vorstellungsvielfalt als Rüstzeug. Dafür notwendig ist ein gemeinsames Verständnis, aber auch konkrete Ideen und Ansätze, wie Probleme angegangen und gelöst werden können. Dieses Ziel verfolgt der Reatch Ideenwettbewerb. Im Rahmen davon, haben wir nach Ideen gesucht, wie die Wissenschaften zusammen mit Politik, Wirtschaft oder Gesellschaft die Krisen von morgen meistern können. Denn eine enge und konstruktive Zusammenarbeit zwischen den verschiedenen Akteuren ist entscheidend, damit wissenschaftliche Erkenntnisse auch Eingang in politische und gesellschaftliche Lösungen finden. Der Ideenwettbewerb wurde u.a. unterstützt durch die Stiftung Mercator Schweiz, die Gebert Rüf Stiftung und den Schweizerischen Nationalfonds (SNF).

Die Beiträge auf dem Reatch-Blog geben die persönliche Meinung der Autor*innen wieder und entsprechen nicht zwingend derjenigen von Reatch oder seiner Mitglieder.