An Unsupervised Approach to Extracting Knowledge from the Relationships Between Blame Attribution on Twitter

Matija Franklin, Trisevgeni Papakonstantinou, Tianshu Chen, Carlos Fernandez-Basso, David A. Lagnado

Abstract: The present paper suggests and examines a technique for analyzing individuals’ public blame attributions on online platforms, with a specific focus on the blame attributions expressed in Twitter ‘Tweets’ in response to various incidents involving Artificial Intelligence (AI). Twitter was chosen as it offers an ’Academic Research product track, which provides researchers with free historical data of discourse which took place on the platform. AI Incidents were chosen as they are a contemporary topic that is often talked about on Twitter, and currently, the focus of many academic papers due to the nature of “The responsibility gap” present when an AI does something that might be blameworthy [44]. Online experiments have been used to investigate these issues in recent years. However, this paper suggests a more ecologically valid approach that can reproduce findings from this field of research. The benefit of the approach is the potential discovery of novel factors that haven’t been traditionally manipulated or measured in experimental settings. The outlined method can also be applied to different online platforms, and research topics within the field of causal attribution [6].