Authors
Mykola Makhortykh, Tobias Rohrbach, Maryna Sydorova, Ani Baghumyan
Abstract
In cooperation with the BAKOM, the project will audit the performance of Google text and image search algorithms in the context of 2023 elections in Switzerland. Specifically, the project will investigate how Google algorithms retrieve textual/visual information about individual candidates participating in the elections. Using a combination of automated/manual web domain labelling and image recognition, this project tracks the presence/absence of algorithmic bias in pre-election web searches of Swiss candidates along three main dimensions (see questions 1-3 below). In an experimental extension, the project will additionally assess the impact of bias in search engine output on evaluations of political candidates (question 4). Specifically, it asks:
- To what extent do textual and visual search results about candidates differ according to their party and gender?
- How do search results about candidates change over the course of the electoral campaign?
- How prevalent is the use of political ads on Google during elections and how do sponsored ads differ between individual candidates (in terms of who sponsors them and who is being targeted)
- How do gender and party differences in search engine output affect political evaluations of candidates?
