Predictive Security and Early Warning: A Theoretical Critique of Conflict Forecasting, Anticipatory Intervention and the Politics of Prediction in Fragile States

Abstract

Governments, regional organisations and international agencies are investing in machine-learning systems that forecast where violence will erupt, drawing on event data, economic indicators, climate signals and news text, in order to intervene before it does. This theoretical paper offers a critique of conflict forecasting as both a scientific and a political practice. Drawing on conflict theory and the early-warning literature, on recent research on machine-learning conflict prediction and its evaluation, on the measurement scholarship on conflict event data, on the anticipatory-action literature, and on critical security studies, it distinguishes three claims that forecasting makes, that violence is predictable, that prediction enables prevention, and that prevention is what predicted violence will be met with, and examines the conditions under which each holds. It argues that the first holds well for the continuation of violence but weakly for its onset; that the second depends on institutional response capacity that forecasting cannot supply; and that the third is contingent on the interests of those who hold the forecast, who may respond with repression, abandonment or pre-emption rather than prevention. The paper derives seven propositions about forecast accuracy under data sparsity and reporting bias, about the institutional conditions under which early warning produces early action, and about the consequences of prediction for the communities predicted. The Sahel and Nigeria's North-West and North-East are the anchoring contexts. The paper closes with implications for regional early-warning systems, national security agencies, humanitarian actors and the research agenda of African peace and security studies. It concludes that a forecast is only as valuable, for those it concerns, as the institution that receives it and the purpose for which it is used, so that investment in local response capacity matters more for the security of the people forecast than further gains in the precision of the models.