Researchers working at the intersection of computational social science, political science, and data analytics have developed models that use social media activity, economic indicators, and historical event data to predict civil unrest — including riots, protests, and political violence — with claimed accuracy that has attracted both interest and skepticism from governments and academics alike.

The systems analyze patterns in Twitter and Facebook posts, news reports, economic data, and historical records of social unrest to identify the conditions under which violence tends to emerge. Spikes in particular types of language, unusually high volumes of discussion around certain topics, and specific combinations of economic and political indicators can, the researchers argue, generate probabilistic forecasts of where and when civil unrest is likely to occur.

EMBERS (Early Model Based Event Recognition using Surrogates), one of the most publicized of these systems, was developed with funding from IARPA, the U.S. intelligence community's research arm, and claimed to have predicted civil unrest events in Latin America with meaningful lead time.

The systems raise important questions that go beyond technical accuracy. Predictive policing systems — which use similar analytical approaches at the local law enforcement level — have been criticized for encoding and amplifying existing biases in policing data, leading to the over-surveillance of communities that were already over-policed. Riot prediction models deployed by governments could be used to preemptively suppress legitimate political activity rather than to prepare constructive social interventions.

Civil liberties advocates note the chilling effect on political expression if citizens know that their social media activity feeds systems being used to assess their likelihood of participating in protests.

The tension between the genuine utility of early warning systems and the potential for their misuse in the service of political suppression is, researchers acknowledge, not resolved by the technology itself.