The second day at the workshop of the Bots Building Bridges project in Bielefeld starts with Dennis Frieß, whose interest is in AI and deliberation. He notes the importance of online publics for deliberation; however, they often do not live up to the ideal norms of deliberation, due to a lack of equality, rationality, and reciprocity as well as problems with incivility.
There have been various attempts to counter this, through law enforcement, moderation, civil society initiatives, and other means; most recently, there is considerable interest in using artificial intelligence to support high-quality, deliberative online discussions.
What problems are there with online discussions, then; how might they be addressed by AI? Key issues include input conditions (acces, equality, anonymity), throughput issues (rationality, civility, reciprocity), and outcome questions (what happens afterwards). Most literature tends to focus on the throughput stage.
Key dimensions of deliberation at this stage include equality, civility, rationality, and reciprocity. Dennis’s project conducted a literature review to explore relevant articles on these dimensions, covering some 171 articles in total, from 2022 and before. This found a strong focus on rationality, in particular; more than half of all articles addressed this. Civility was a ore minor concern, and reciprocity and equality were rarely addressed at all.
Research on rationality very predominantly used various argument mining approaches (at the level of whole discussions, or of individual comments). Civility research mostly addressed questions like the detection of incivility and hate speech, and on occasion explore automated moderation systems as well as AI assistants that encouraged participants to revise their comments for more civility. Reciprocity research, to the extent that it exists at all, often encouraged engagement between users with different viewpoints. Equality research is underdeveloped and shows no clear patterns.
Current applications of AI mostly focus on rationality and civility, then; however, the use of AI can also reinforce inequalities in participation, especially if they focus mainly on formal argumentation – this can discriminate against certain types of participants or privilege particular types of argument (e.g. fact-based rational argumentation) while excluding others (e.g. storytelling).
The study is limited by its end date of 2022, though: there has been considerable further development with the explosion in AI tools, and new research in the field needs to be assessed as well. However, there is also immense public skepticism towards AI at this point, including as a moderator of online discussions, so the situation has not necessarily improved much in the meantime.
