The Science of “Civil Unrest Prediction” Meets Social Media

• In Event: STS and Computational Knowledge Production in Policing and Criminal Justice - II

Abstract

Social media are lauded for enabling and organizing social movements. Yet in contrast to these hopeful imaginaries are efforts to surveil social media at scale and in near-real time using advanced techniques from data science (Tufekci, 2017). Recently, technologies to monitor suspicious individuals and groups culminated in government-funded research that used social media data to predict US protests critical of President Trump in 2016. That project is part of a research area called “civil unrest prediction” in a scientific literature dedicated to forecasting strikes and protests across the globe (e.g. Indonesia, Brazil and Australia). Outside government, human rights organizations and industries such as tourism and supply chain management are the potential consumers of these techniques. In this paper we critically review the recent history of civil unrest prediction products and the data science that supports them. As a multidisciplinary team of co-authors versed in both data science and STS, we characterize the state of the art, concluding that a key feature of civil unrest prediction software is a series of foundational assumptions about what and who is “risky.” Opportunities for agents of the state and industry to act are overwhelmingly emphasized in contrast to the potential risks and harms for those surveilled. Using the framework of infrastructure studies, we explain civil unrest prediction as classification infrastructure (Bowker & Star, 1999), trace the riskiness classification decisions from their assumption to their detailed quantification and execution, then reveal the potential for normatively worrisome consequences not previously discussed.

Authors

• Gabriel Grill, University of Michigan

• Christian Sandvig, University of Michigan