MRMCD 2017

How much do we live in different worlds, when we search for political topics online? Germany has an upcoming election, and nobody really knows much about the knowledge spheres internet users find themselves in. This is in contrast to those spheres in our well-established newspaper and magazine landscape. To alleviate this gap of knowledge, Algorithm Watch has set up the "Datenspende Bundestagswahl 2017", a plugin for firefox and chrome which automatically searches for 16 keywords and submits the data to a central server. The keywords consists of the full name of the most important politicians and the seven largest parties. The harvested data is public and can be analyzed by anyone. The project was funded by a consortium of German media authorities and is supported by SpiegelOnline as our media partner. (datenspende.algorithmwatch.org)

In this talk I will give you some first insights into the variance of search results different people get. While we do not know anything about our users - for data protection reasons - we can see that for some keywords, there are clusters of people getting different results while for others the results are very homogeneous. In all cases analyzed so far, the amount of variance is less than expected based on a full personalization of search engine results and more varied than expected based on a mere regionalization.

The main result of the project, however, is that is a proof-of-concept showing that society is actually able to audit socially relevant algorithms by combining the efforts of individuals - both as data donors and data analysts. This principle has to be transferred to audit other social platforms that are even more likely to influence public opinion than search engines, especially social networks. Today, however, most of them do not offer the technical tools for society to be able to do that. This has to be changed.