This topic builds on both Building Evidence: Data Science in the Age of AI and Listening to Politics at Scale. It applies the same questions about representation, sampling, classification and inference to an especially difficult and politically contested problem. Misinformation is not a single observable property of a post. An algorithm might examine language, compare claims with evidence, score the history of a source, trace patterns of diffusion or look for coordinated behaviour. Each method detects something different, and none gives us a simple mechanical test for truth.

The difficult cases expose the politics inside the technical problem. Who defines the ground truth? Is a sincerely mistaken claim the same as deliberate deception? How should a system handle satire, contested interpretations or knowledge that is still changing? A detector can also become a political instrument if its errors fall unevenly or if nobody can challenge its decisions.

I have worked on research involving bots, online communities and the spread of contested information. I can help you compare content-, source-, bot- and coordination-based methods, then design and test a practical detection process on difficult cases. The aim is to understand what a detector is really detecting, evaluate its errors and identify where human judgement and political power enter the system.

Core reading

• Francesco Pierri et al., “One Year of COVID-19 Vaccine Misinformation on Twitter: Longitudinal Study” (2023).
• Salvador Lopez-Joya et al., “Dissecting a Social Bot Powered by Generative AI: Anatomy, New Trends and Challenges” (2025).