2025. “Does the United Nations Need Agents? Testing the role of AI agent generated personas in humanitarian action.” Working Paper, UNU CPR, New York.
This paper illustrates a case study with two AI agent generated persona systems, one called “Ask Amina” and the other “Ask Abdalla.” The first is designed to create an accurate digital representation of a refugee living in a camp in Chad. The second creates a digital replica of a combatant leader in the Rapid Support Forces (RSF), a group active in the southeastern part of Sudan from which many refugees are fleeing. Both systems combine digital avatars with large language models (LLMs), retrieval-augmented generation (RAG) and AI agent curated knowledge bases specifically designed to maximize the representativeness of each persona.
The effectiveness of Amina’s representation is measured by comparing “her” responses to survey questions against those provided by actual members of the refugee population. A sample of a negotiation with Abdalla, as well as a conversation between Amina and Abdalla, are qualitatively assessed. The experiment methodology and results are described in the case study section of this paper with further details in the appendix. Both personas can be experienced and tested at askamina.ai. Readers are invited to interact with either persona on any desired subject.
The primary objective of Ask Amina is to enable humanitarian and relief workers to ask questions about refugee experiences, needs, and sentiments, and receive responses that closely mirror real refugee perspectives. The objective of Ask Abdalla is to enable diplomats, and mediators to practice their skills with a persona that responds in ways consistent with a real combatant’s known behavioural patterns. The two personas are also put in conversation with each other as a preliminary test of a virtual AI-based community dialogue simulation.