AI & Humanitarian Technology Blog

Last updated: 2026-09-05

Practical insights on artificial intelligence, machine learning, large language models (LLMs), cybersecurity, and digital transformation from 20+ years of implementing these technologies in UN humanitarian operations. From RAG applications and multi-agent AI systems to predictive analytics for refugee resettlement, these articles bridge the gap between advanced AI and real-world humanitarian impact.

All Articles (51)

  • ISO/IEC 27090 in Ten Minutes: What the New AI Security Standard Asks of You (2026-10-19)
    The first international standard on security threats to AI systems, explained in ten minutes: what it is, what it is not, the twelve threats, how it fits ISO/IEC 27001, and where to start.
  • Open-Weight vs Open-Source LLMs, Fine-Tuning and RAG: A Plain-English Guide (2026-10-05)
    Weights are the model, open weight is about who holds it, fine-tuning changes it, and RAG changes what it reads. A plain-English guide to all four, with examples.
  • Open Source and AI in Digital Transformation: A Practical Playbook for Organizations (2026-10-04)
    A practical playbook for combining open source and AI: alternatives by technology layer, seven use cases, real institutional case studies, and a 90-day plan.
  • Open Source AI in Healthcare: A Practical Guide to Models, Use Cases and Where to Start (2026-10-01)
    Open weight models let health organizations run AI inside their own security boundary. A guide to MedGemma, Meditron, Llama and DeepSeek, real use cases, and how to start safely.
  • Three Lessons from the OIOS Digital Transformation Evaluation for the AI Era (2026-09-18)
    The OIOS evaluation found 1,217 digital initiatives but uneven strategy, interoperability and measurement. Three lessons for leaders entering the AI era.
  • AI Ethics in Humanitarian Action: When an Algorithm Influences a Human Life (2026-09-14)
    AI ethics in humanitarian action means converting human rights, fairness, privacy and accountability into controls. This practical guide examines the risks, the safeguards and UNHCR CareLine as a real case study.
  • AI Fraud Detection in the GenAI Era: Both Sides Got Faster (2026-09-09)
    Generative AI upgraded fraud detection and fraud itself in the same eighteen months. Synthetic identities, deepfake social engineering and forged documents on one side; better anomaly detection and faster triage on the other.
  • AI Agents in the Workforce: What Actually Changes at Work (2026-09-09)
    When agents start doing real work across HR, finance, IT and operations, the technology is the easy part. Accountability, identity, access review and the documents that assume a human signatory all have to be redesigned.
  • Before You Build an AI Agent: 7 Decisions Leaders Must Make (2026-09-09)
    Seven decisions that determine whether an agent project succeeds, all of them made before any code exists: the outcome, the autonomy level, the tools, the data access, where a human approves, how it is evaluated, and who owns it.
  • Enterprise AI Agent Platform: From Demo to Production (2026-09-09)
    A demo agent needs a model and a prompt. A production platform needs orchestration, memory, identity, observability, cost control and recovery. This is what sits between the two, and why most of the work is not the model.
  • How Modern AI Applications Actually Work: A Clear Guide (2026-09-09)
    Most explanations of AI stop at the model, which is roughly as useful as explaining a bank by describing a calculator. This traces one request through validation, retrieval, the model, tools, memory and guardrails, then follows an enterprise example end to end.
  • AI Explained Simply: 28 Questions Answered at Three Levels (2026-09-05)
    Twenty-eight common questions about AI, each answered three times over: a picture you will not forget, where the idea already touches your day, and the mechanics underneath. Read only as deep as you need.
  • AI and Localisation Pull in Opposite Directions (2026-09-04)
    The sector spent a decade committing to shift capability toward local actors. AI accumulates wherever the data, compute and engineers already are, which is almost never the local organisation. Nobody is choosing this, and it will keep happening unless it is designed against.
  • How to Tell a Last-Mile AI System Is Failing (2026-09-04)
    Users who cannot go elsewhere do not complain, so the usual failure signals never arrive. Which proxies to instrument before launch, why every one of them is gameable, and the single control that actually detects confident wrong answers.
  • What Sovereign AI Cannot Buy (2026-09-04)
    Sovereign AI is sold as independence. No country achieves that, including the largest. What is achievable is managed dependency, and it is cheaper than the version being marketed.
  • When Not to Build an AI Agent: The Reliability Arithmetic (2026-09-04)
    Step reliability multiplies. A component that is right 95% of the time produces a system that is right 60% of the time over ten steps. Most things being built as agents would be more reliable as workflows, and the test for which is one question.
  • Does the UN Regulate AI? Two Systems That Share a Name (2026-09-04)
    One system is member states negotiating non-binding norms. The other is each organisation governing what its own staff may deploy. They share a name, and confusing them sends people looking for rules in the wrong document.
  • RAG or Fine-Tuning: The Governance Questions That Decide It (2026-09-04)
    The choice is usually argued on accuracy and cost. In an institution it is settled earlier, by who can correct a wrong answer, whether personal data can be deleted afterwards, and what an auditor can be shown.
  • How LLMs Actually Work: Weights, Training and RAG in Plain Language (2026-09-04)
    What a weight actually is, what training does to it, and what happens inside the model between your question and its answer. Plain language, no mathematics.
  • How to Run AI Locally Without a GPU: A Beginner's Guide (2026-09-04)
    You do not need a graphics card or a cloud subscription to run a language model. This is the whole setup on an ordinary Windows machine: the llamafile runner, a quantized GGUF model, and one batch file. It runs offline, and it runs from a USB drive.
  • From Learning to Earning: What Peru Is Testing With an AI Career Coach (2026-08-20)
    A World Bank programme in Lima is testing whether a large language model can deliver the personalised half of career guidance at scale. It launched in March 2026 across 110 schools and is being evaluated as a randomised controlled trial. The results are not published yet, and the connectivity constraint is unsolved.
  • The Walkthrough Stops Working: IT Audit Methodology for AI Systems (2026-08-19)
    IT audit methodology assumes a system can be read. AI systems cannot be, in the sense the methodology means. Three things change: design testing moves upstream to the process that produced the model, sampling gives way to full population testing, and opinions acquire a shelf life.
  • Who Audits the Agent? AI Agents and the Future of Internal Audit (2026-08-19)
    AI agents hold credentials and act without waiting for approval, which makes them participants in the control environment rather than tools. What that changes for audit evidence, identity governance, and the audit programme itself.
  • Why AI Pilots Fail in Government and UN Agencies (2026-08-17)
    Most AI pilots do not fail because the model is weak. They fail because nobody designed ownership, data readiness, language testing, governance, procurement control, and the path from demo to production.
  • A Practical AI Roadmap for National NGOs (2026-08-14)
    National NGOs should start with AI in five stages: pick one repetitive, text-heavy, low-risk problem; put data protection in order before any pilot; run one 90-day pilot with a measured baseline and a named reviewer; write a one-page AI policy; and build capability through AI champions and a shared prompt library. Never enter beneficiary-identifying data into public AI tools, and keep AI away from decisions about people.
  • Building AI Systems Where Infrastructure Fails: Lessons from 20 Years in the UN (2026-05-09)
    AI systems work where infrastructure fails when they are designed around the constraint rather than the ideal. An interactive voice response system that required no internet, smartphone, or literacy served 700,000 refugees across five countries and cut registration time by 75 percent. Data governance precedes AI governance: 63,000 missing contact records were identified before deployment.
  • Seven Decision Rights for AI Governance in UN Agencies (2026-05-09)
    AI governance in the United Nations works as practical decision-making about who controls data, who audits algorithms, and who is accountable when automated systems affect displaced people. Data governance comes first: at UNHCR Jordan it identified 63,000 missing contacts. Reviewers are assigned by role, not name, and frameworks reference UNESCO, OECD, EU AI Act, and NIST guidance.
  • Agentic AI Explained Through a Humanitarian Workflow (2026-05-09)
    Agentic AI systems break goals into steps, select tools, execute workflows, and adjust based on outcomes, unlike traditional AI that returns one output per input. In humanitarian operations, an agentic system can receive a voice message, transcribe it, classify its urgency, route it, and log it. Human-in-the-loop oversight is not optional in these contexts.
  • The Next Great Divergence: Asymmetries in Artificial Intelligence Preparedness and the Political Economy of the Arab World (2026-01-05)
    This paper examines the widening AI preparedness gap in the Arab world, identifying a "Next Great Divergence" between GCC frontier economies and lagging states, proposing a "Two-Track, One-Region" policy framework for regional integration.
  • Why I Am Using McKinsey's Vizro (And Why You Should Too) (2025-12-03)
    Vizro, McKinsey's open-source low-code dashboard framework, removes the friction of building dashboards: output looks professional immediately, Pydantic validation adds engineering discipline, and modular filters work across charts without callback rewriting. Shahzad Asghar built a fully interactive dashboard with filtering in about 15 minutes, after two decades of dashboard work in resource-constrained environments.
  • Open Source in Humanitarian Response: Case Studies, a Decision Matrix, and What LLMs Are Changing (2025-09-23)
    How humanitarian organizations can evaluate open source tools, digital public goods and open-weight AI through real examples and a practical decision matrix.
  • Understanding Synthetic Data: A Privacy-Preserving Solution for Sensitive Information (2025-04-06)
    Synthetic data is artificially generated information derived from real datasets that cannot be linked back to specific individuals, preserving statistical properties while protecting privacy. It keeps data useful where k-anonymization causes significant data loss by redacting outliers. Generation follows four steps: prepare the data, select attributes, synthesize artificial records, and verify utility.
  • The Model Context Protocol (MCP): The USB-C for AI Applications (2025-04-02)
    The Model Context Protocol (MCP) is a standard that connects AI applications to external data sources and tools the way USB-C unified device connectivity. Instead of custom connectors for every API or database, MCP clients send structured requests to MCP servers, reducing development time, maintenance cost, and long-term technical debt while supporting enterprise security requirements.
  • Artificial Intelligence’s Transformative Impact on Industry and Economy (2025-04-01)
    After years of working with AI in humanitarian operations, I began to explore how artificial intelligence is transforming not just how we deliver aid, but how entire industries, labor markets, and global economies are evolving. This article examines that shift through real-world examples in manufacturing, healthcare, finance, logistics, and governance, offering an analytical look at where we are, and what’s next.
  • AI's Role in Safeguarding Vulnerable Populations and Developing Inclusive Digital Frameworks (2025-03-25)
    Exploring how AI can protect vulnerable populations through inclusive digital frameworks, with case studies on refugee support, disability inclusion.
  • Debunking Data Protection Myths in the Age of AI for Humanitarian Work (2025-03-03)
    Common data protection beliefs fail under scrutiny in humanitarian AI work: AI cannot guarantee full security, anonymized data can be re-identified (a 2019 study re-identified 99.98 percent of Americans from 15 demographic points), regulatory compliance is only a baseline, AI does not replace human oversight, and algorithms reproduce the biases in their training data.
  • How ChatGPT Works: Inside AI Language Models (2025-02-07)
    ChatGPT works through a Generative Pre-trained Transformer, a neural network with millions to billions of parameters. It is pre-trained on massive internet text, then refined with Reinforcement Learning from Human Feedback. It breaks input into tokens, weighs context with self-attention, and predicts responses one token at a time from statistical patterns, which makes hallucinations and inherited biases possible.
  • Digital Transformation Strategy for Humanitarian Organizations (2025-02-01)
    A practical framework for changing humanitarian operating models through mission outcomes, shared accountability, data governance, secure architecture, progressive modernization and field-ready delivery.
  • Using AI to Categorize Sensitive Refugee Call Transcripts: A Step-by-Step Approach (2025-02-01)
    Learn how we built an AI system to categorize sensitive refugee call transcripts using advanced techniques like Knowledge Distillation and multilingual model training.
  • Frequently Asked Questions: AI in the UN System (2025-01-10)
    A comprehensive FAQ guide about how Artificial Intelligence is being implemented and managed across the United Nations system, including challenges, frameworks, and future recommendations.
  • Operational Use of AI in the UN System (2025-01-10)
    The United Nations system runs more than 700 AI projects across over 50 UN entities, but adoption is decentralized and inefficient, so it needs a unified, integrated, and coherent approach. About 64 percent of projects target custom thematic solutions, text and language processing dominates at 48 percent, and generative AI chatbots using retrieval-augmented generation are expanding fastest.
  • The Data-to-Decision Journey: A Practical Guide for Humanitarian Work (2025-01-08)
    In humanitarian work, raw data becomes action through five interconnected stages: data collection from sources such as refugee surveys and registration systems, data cleaning and preparation, data analysis, insight generation, and decision-making. Skipping any step can lead to flawed outcomes. Evidence-backed decisions produced results such as targeted immunization campaigns and mobile clinics for elderly refugees.
  • Learn Generative AI and LLMs in 7 Weeks (2025-01-07)
    A comprehensive 7-week program to master Generative AI, LLMs, and AI agents. From foundations to advanced applications, this course provides hands-on experience and practical skills.
  • OpenAI Swarm for Multi-Agent Collaboration: A Guide for UN IT Teams (2025-01-06)
    OpenAI Swarm provides a robust framework for coordinating multiple AI agents, enabling efficient task specialization, seamless collaboration, and scalable solutions. In the United Nations context, this technology offers transformative potential for enhancing data analysis, streamlining humanitarian operations, and improving user interactions across diverse domains
  • My Journey Using AI and Technologies and Key Takeaways (2025-01-04)
    Reflecting on two decades of using AI and technology for humanitarian impact, sharing key lessons and insights from transformative projects in refugee assistance and digital innovation.
  • AI for Good: How Technology is Shaping the Future of Humanitarian Work (2025-01-02)
    Discover how AI is transforming humanitarian work through real-world applications and projects implemented in Jordan, focusing on refugee support and crisis response.
  • How AI is Transforming Global Development: Insights from My UN Experience (2024-12-18)
    Explore how artificial intelligence is transforming global development through UN initiatives, from poverty alleviation to healthcare and climate change mitigation.
  • Secure and Anonymized Reporting System for Humanitarian Emergencies (2024-11-09)
    A comprehensive proposal for a secure and anonymized reporting system for humanitarian emergencies, using WhatsApp voice messaging, AI transcription, and data analytics.
  • The Role of Artificial Intelligence in Achieving the UN Sustainable Development Goals (SDGs) (2024-11-08)
    Discover how artificial intelligence is accelerating progress towards the UN Sustainable Development Goals, from fighting hunger to climate action.
  • Shahzad Asghar: Artificial Intelligence in UN Operations (2024-07-02)
    How I have applied artificial intelligence and data analysis to humanitarian operations in the UN system, and what those projects produced.
  • Poverty Mapping: Technical Methods and Applications (2024-03-24)
    A technical examination of poverty mapping methodologies, including geospatial analysis, satellite imagery, machine learning, and remote sensing.