Send message to Authors

Short abstract:

The contribution discusses the outcomes of the empirical study that combines survey and AI audit methods to investigate how Swiss citizens use AI-driven search engines to find information about popular votes and whether specific groups of citizens are more likely to be exposed to AI bias.

Long abstract:

AI-driven platforms, such as search engines, play an increasingly important role in how citizens find and consume politics-related information in liberal democracies. The functionality of these platforms is affected by various system factors, including the degree to which they use contextual information or randomise their outputs. However, despite the importance of system factors, it is also integral to account for the agency of use who interact with AI, especially as these interactions have major implications for the quality of information provided by AI. To better understand how user agency can affect the role of AI in the context of democratic decision-making, we empirically investigate how Swiss citizens use AI-driven web search engines, Google and Bing, to find information about federal-level popular votes in the spring of 2024. Using a large-scale survey, we first investigate how the choice of search queries used by a representative sample of Swiss citizens to find information about votes is shaped by their political views, education, and attitudes toward a voted issue. Then, we use the survey data to conduct a virtual agent-based AI audit and simulate search behaviour of Swiss citizens to systematically investigate whether certain groups of citizens are more likely to be exposed to AI bias (e.g. in terms of selection of information about votes being disproportionately influenced by opinions of specific political parties). The findings of the study will contribute to better understanding of the interaction between system- and user-side factors of AI systems and their implications for democratic decision-making.