Publications

AI- and machine learning–based social risk forecasting and explanation are applied across fragile and climate-affected contexts to predict change in conflict (DRC), population movement (Horn of Africa), and crime (a SIDS), using large multimodal datasets (satellite imagery, NLP on news/social media, economic and climate indicators, geospatial data). Three proof-of-concept models achieve out-of-sample accuracies of 63–76% (conflict) and 70–74% (population change), and generate crime risk proxies where official statistics are scarce. Key associated factors include language about sensitive topics and political actors, economic pressures and prices, climate stress, and social perceptions. Results illustrate how social science–informed AI complements conventional analytics and supports policy and operations for risk monitoring, early action, and targeted resource allocation.