How the UN System Uses AI
20 September 2024 · 48th HLCM Session
How the UN System Uses AI: 716 Projects, 5 Focus Areas
A comprehensive case study based on the UN Chief Executives Board report on the operational use of Artificial Intelligence across the United Nations System.
50+ UN Entities
Generative AI · ML · NLP
UN Chief Executives Board for Coordination (CEB)
Report on the Operational Use of AI in the UN System – High-Level Committee on Management, Task Force on AI (HLCM TF-AI)
This case study synthesizes the official UN report. The original document was prepared by the HLCM Task Force co-led by IFAD and WFP, with input from over 30 UN entities.
The Big Picture: AI Across the UN System
The United Nations System is rapidly integrating Artificial Intelligence into both its external programmatic delivery and internal core operations. From predictive analytics for humanitarian response to automated language translation for multilingual communication, AI is transforming how UN agencies work and deliver on their mandates.
However, the decentralized nature of the UN System presents challenges. Without coordination, agencies risk duplicating efforts and missing shared opportunities. The High-Level Committee on Management Task Force on AI (HLCM TF-AI) was established in October 2023 to address this gap, with two core objectives: develop normative guidance on AI use, and identify mechanisms for pooling technical capacity and knowledge sharing.
📌 Key finding: Nearly 60% of AI projects report collaborations with at least one other stakeholder, and 30% involve partnerships within the UN System itself.
The 5 Focus Areas of UN AI Adoption
The HLCM TF-AI structured its analysis around five focus areas, based on working meetings and bilateral consultations with over 30 UN entities.
Knowledge sharing, joint efforts, resources, and projects supporting the SDGs.
Mapping of applications by data type (text, multimodal, geospatial, etc.) and area.
Open-source movement, GitHub, Hugging Face, and UN-led inner-source platforms.
RAG architecture, chatbots, fine-tuning, and UN-specific implementations.
Workforce impact, governance structures, and organizational strategies.
AI Use Cases: Text Dominates, But Multimodal Is Rising
Of 363 AI use cases and solutions catalogued, 48% focus on text, documents, and language – including chatbots, translation, and knowledge management. Mixed/multimodal applications account for 23%, while quantitative, GIS, audio, and image processing make up the remainder.
64% of projects are custom solutions for thematic work (e.g., climate, health, education), while 25% address functional areas (HR, finance, procurement). This reflects the UN’s priority on mission-driven AI rather than generic automation.
Top technologies, tools, and vendors mentioned in UN AI projects:
Python (60)
Microsoft (44)
OpenAI (30)
Google (30)
AWS (20)
Generative AI: 244 Projects and Growing
Of 716 total AI projects, 244 (34%) are Generative AI projects. Among these, 183 are use cases or solutions, and 64 are specifically Generative AI chatbots. The report notes that “2023 will go down in history as the year that generative AI took the world by storm.”
Retrieval-Augmented Generation (RAG) has become the dominant architecture for UN chatbots, allowing models to ground answers in organizational knowledge bases. Case studies include ITCILO’s AnswerMate (internal governance chatbot) and the UNIFY HR AI Chatbot (inter-agency HR policy assistant across 13 UN organizations).
Frameworks for Responsible AI
Two practical frameworks emerged from the report:
- PRISM Framework (DTN GenAI CoP) – evaluates AI use cases across Mission Impact, Efficiency, Risk, Nonfinancial Value, and Technical/Internal/External Readiness.
- WIPO Risk Assessment Framework – assesses commercial AI offerings across data collection, security, residency, retention, access management, ethics, and integration.
Open Source & Inner-Source: The UN’s Digital Public Goods
The UN System has over 60 organizational accounts on GitHub with 2,700+ repositories. On Hugging Face, UNHCR’s Hate Speech Detection Model and UNDP’s SDGi Corpus (text classification for SDGs) are leading examples of AI-specific digital public goods.
The DTN Open-Source Software Community of Practice is advancing five key initiatives: a software catalogue, common policy framework, UN open-source license, code-hosting platform (GitLab), and cross-organizational capacity building.
Knowledge Sharing: The UN’s AI Communities
Co-led by UNICC & UNAIDS. Guidance on Generative AI tools, vendor selection, and PRISM framework.
Co-led by UNESCO & ITU. Principles for Ethical AI, White Paper on AI Governance.
Informal community with 400+ members. Technical exchanges on prompt engineering, fine-tuning, agents.
40+ UN partners. Global summit, neural network (35k+ members), and focus groups on AI for health, disaster management, agriculture.
How UN Organizations Are Structuring AI
The report highlights diverse approaches:
- IFAD: Bottom-up via Omnidata platform (60+ use cases), now shifting to corporate AI architecture.
- UNDP: AI Working Group with senior leadership; sandbox model for 2-week prototyping sprints.
- UNHCR: Data Innovation Fund; 8 Generative AI use cases in production; refugee-led innovation fund.
- UNICEF: Phased approach – administrative functions first, then programmatic, with children as a longer-term priority.
- WFP: Responsible AI Task Force; AI sandbox environment with cloud partners (CERN, Google, Microsoft).
Key Recommendations for the UN System
The HLCM TF-AI proposes 7 specific recommendations:
- Extend scope of AI communities of practice to cover all AI subsets (ML, Generative AI, etc.)
- Standardize and centralize cataloguing of AI projects across the UN System
- Support a common framework for evaluating and prioritizing AI use cases (PRISM)
- Develop a UN project-hosting platform for code, datasets, and models
- Create a system-wide risk assessment framework for commercial AI offerings
- Examine workforce implications (reskilling, talent pools, job roles)
- Strengthen collective bargaining with AI vendors and explore partnerships for equitable access
What M&E Professionals Can Learn
The UN’s communities of practice (DTN, UNIN, IAWG-AI) show that inter-agency knowledge sharing reduces duplication and accelerates innovation.
PRISM and WIPO’s risk assessment provide practical, replicable tools for any organization evaluating AI use cases or vendors.
244 UN projects prove that Generative AI is not experimental – it is being deployed in HR, knowledge management, and programmatic work.
The UN’s approach to controlled sharing (inner-source) within trusted partnerships offers a model for sensitive M&E data.
Key Takeaway for Development Professionals
The UN System is not waiting for AI to mature. It is actively deploying 716+ AI projects across every SDG, guided by ethical principles and practical frameworks. The future of multilateral action is AI-augmented – and the HLCM report provides a roadmap for any organization seeking to adopt AI responsibly at scale.
Original Source & Further Reading
Primary Source: United Nations Chief Executives Board for Coordination (CEB). (2024, September 20). Report on the Operational Use of AI in the UN System. High-Level Committee on Management, 48th Session. CEB/2024/HLCM/28/Add.2.
The report was co-led by IFAD and WFP, with inputs from FAO, ILO, IMF, ITU, UNDP, UNESCO, UNFPA, UNHCR, UNICEF, WIPO, and over 30 other UN entities.
Advance your skills in AI-enhanced evaluation
Learn how to apply the UN’s AI frameworks in your own M&E practice.
