Disinformation in social media makes it more difficult to form opinions in the run-up to elections. What challenges arise when AI systems are also used?
Christoph Bieber: AI systems can be used to accompany and shape discussions on digital platforms. This sometimes creates even more confusing communication situations in social media than they already are. It is therefore important that participants in online conversations know whether they have a human or non-human counterpart. At the same time, AI systems can also provide greater clarity when used, for example, to detect and suppress misinformation. In addition, AI systems expand the spectrum of misinformation and disinformation - for example, through multimedia "deep fakes" when audio and/or video statements of well-known personalities are artificially created - i.e.: faked. In this way, misleading campaign messages could be developed and disseminated, which would make proper voter information more difficult.
Large amounts of data are generated in digital election communication. How can these be used for the parties?
Christoph Bieber: On the part of the parties, such evaluations can in principle contribute to better and more targeted use of available resources in the election campaign. The aim here is not to directly influence citizens, but rather to organize the election campaign in such a way that voters can be addressed in a better, more targeted way. The automated evaluation of social media communication opens up possibilities for large-scale issue monitoring in the election campaign. Recommendation algorithms could be used to improve intra-party group communication, while AI-assisted image recognition can help detect negative campaign themes from political rivals and facilitate responses to them. Such evaluations develop particular value in a long-term perspective. After all, the more data that is available and can be processed within a party, the better the starting conditions for future election campaigns. However, the parties are still a long way from such "learning systems," and not only in Germany.
Does AI also promise to benefit voters?
Christoph Bieber: Potential also arises from a civic perspective: AI-supported processes can be used to pre-sort and sift through the diverse information materials of political actors according to their own interests. Such improved engagement with election programs - and election promises or actual voting behavior in parliament - would be an update of popular voter information tools. Common "voting advice applications" such as Wahl-O-mat, Wahlkompass, FollowtheVote or WahlSwiper already work with specially developed recommendation systems. However, users can only personalize them to a very limited extent and within these applications. A kind of "digital twin" that knows the online usage behavior (and possibly also the political preferences) of "its" user could filter out the content that matches the user's personal attitude or thematic interests from the almost unmanageable range of information on the election. In this way, a democratic "diversity requirement" could be realized if algorithmic recommendations helped Internet users to take a "broader" look at the political landscape.
The white paper "AI Systems and the Individual Choice Decision" of Plattform Lernende Systeme is available for download (in German).
This interview is released for editorial use (if the source is named © Plattform Lernende Systeme).