Berlin, Germany – VIEWS researchers Paola Vesco, Chandler Williams, Sonja Häffner, and Simon Polichinel von der Maase participated in the Symposium on Crisis Early Warning, held at the German Federal Foreign Office in Berlin on September 17–18, 2025. Co-organized with the Center for Crisis Early Warning (CCEW) at University of the Bundeswehr Munich, the symposium focused on “The Potential of Data, AI, and Interdisciplinary Analysis in Situational Awareness and Decision Making”.

Paola Vesco chaired a panel on “Advancing Methodological Approaches in Crisis Early Warning“, featuring contributions from Thomas Chadefaux, Chandler Williams, and a desk officer from the German Intelligence Agency. The session highlighted innovations in analytical frameworks for crisis forecasting, including Williams’ recent work on strengthening real-time conflict monitoring through nowcasting. His study leverages UCDP’s monthly candidate event data and an ensemble of machine learning models to correct for reporting delays and biases, producing more accurate short-term fatality estimates. By anticipating gaps in early reporting, this approach enhances the reliability of near-term forecasts—offering humanitarian actors and policymakers more timely and actionable insights.

Sonja Häffner led a workshop session with Christian Oswald (Center for Crisis Early Warning, Munich) titled “Pimp My Dataset – Enhancing Existing Event Datasets with Large Language Models”. They presented research on how Large Language Models (LLMs), combined with human expertise, can be used to enrich existing conflict event datasets. Häffner demonstrated the concept with recent work using synthetic text data (generated with LLMs) and an active learning framework to develop a corpus capturing targeted attacks on education in armed conflict. This latter forms part of the EdAttack Project, led by Gudrun Østby at PRIO, to which several VIEWS researchers are contributing.

Simon Polichinel von der Maase attended as an expert on AI methodologies. His contribution focused on the responsible and effective integration of AI into early warning systems, emphasizing five core principles: the importance of specialized models, integration of user feedback, embedding forecasts into decision mechanisms, continuous collaboration with end users, and caution in the application of generic AI tools.

The symposium brought together researchers, policymakers, and practitioners to explore the potential of data, AI, and interdisciplinary analysis in situational awareness and decision-making. The interactive format fostered dialogue between practitioners and researchers on the promises and pitfalls of emerging data sources and predictive methodologies; ethical considerations; and the importance of collaborative, transparent approaches to conflict early warning.