An international team of experts including WorldPop Director, Professor Andy Tatem advise that while Artificial Intelligence (AI) can revolutionise how we predict hunger, it requires strict human oversight to prevent life-threatening errors. In a commentary, published in Nature Food, they warn that replacing expert analysis with unvetted algorithms could undermine humanitarian efforts in low- and middle-income countries.
Food security early warning systems, such as the Famine Early Warning Systems Network (FEWS NET), guide the distribution of billions of dollars in food aid and cash transfers. With humanitarian funding facing deep cuts, there is a growing pressure to use AI to reduce costs. However, the researchers argue that “timely” and “accurate” data alone are not enough; domain expertise is essential to interpret complex crises.
AI and machine learning (ML) already assist in monitoring earth system hazards like droughts and floods. In conflict-prone areas, these technologies help with geospatial mapping and detecting building footprints from high-resolution satellite imagery. Yet, data in these regions is often sparse or intentionally modified by conflict actors. Without a human analyst to validate these reports, automated systems are prone to bias and misinformation.
Professor Tatem comments: “AI is a powerful tool for streamlining data collation, but it is no substitute for human judgement. We must use these technologies to augment analyst capabilities, ensuring that every life-saving forecast remains transparent and accountable”.
The study highlights the success of using Large Language Models (LLMs) to digitise agricultural reports, saving hours of manual work. To maintain trust, the team recommends that all AI-informed systems use de-identified or aggregated data and adhere to robust governance frameworks. Models must be versioned and accredited by independent auditors to ensure they are suitable for informing preparedness decisions.
Ultimately, AI should be a tool for the analyst, not a direct producer of unreviewed forecasts. Further investment in data infrastructure is necessary to modernise early warning systems while maintaining the accountability required for humanitarian action.
Co-authors of the commentary include experts from the Department of Geographical Sciences at University of Maryland, the International Organization for Migration, the International Food Policy Research Institute, the Peace Research Institute Oslo, ACLED, Microsoft’s AI for Good Lab, FEWS NET, the European Commission, the National Oceanic and Atmospheric Administration, American Institutes for Research, Bioversity International and others.
