Working Paper

Does the United Nations need agents?

Testing the role of AI agent generated personas in humanitarian action.

A new working paper published by UNU-CPR, authored by Eduardo Albrecht, explores the potential of AI agent-generated personas to support humanitarian, peacebuilding and development work in crisis-affected contexts.

The study introduces two AI personas: Ask Amina, a simulated refugee from Chad’s Metche camp, and Ask Abdalla, a fictional Rapid Support Forces (RSF) commander in Sudan. These personas are powered by large language models and Retrieval-Augmented Generation (RAG) systems, drawing on knowledge curated by “anthropologist agents”—AI systems that autonomously collect and organize cultural and contextual data.

Amina achieved 80 per cent accuracy when responding to real-world survey questions on refugee experiences, while Abdalla effectively modeled negotiation dynamics typical of conflict actors. A simulated dialogue between the two revealed common ground on key issues like education, women’s safety and humanitarian needs - demonstrating how such tools can inform planning, training and early warning systems.

However, the paper emphasizes that these technologies pose ethical challenges. Risks include data misuse, biased outputs, misrepresentation of vulnerable populations and a potential erosion of human agency. These concerns are particularly relevant in high-stakes humanitarian operations.

To mitigate these risks, the paper proposes five key recommendations:

  1. Transparency: explaining how each response is generated, including data sources and reasoning.
  2. Explainability: responses must be interpretable by non-technical users.
  3. Pilot-first approach: gradual adoption that allows for testing and refinement before scaling.
  4. Integration: AI personas should integrate existing forecasting and planning tools.
  5. Community participation: affected communities must co-govern persona development, with the authority to approve or modify outputs.

Used responsibly, the Working Paper argues, AI-generated personas can supplement traditional data collection methods, improve access to underrepresented perspectives, and enhance decision-making in fragile settings. But they must be guided by robust ethical frameworks and rooted in community engagement to ensure they serve, rather than displace, the voices of those most affected.

Read 'Does the United Nations need agents?' here

Suggested citation: Albrecht Eduardo. Does the United Nations need agents? : UNU-CPR, 2025.