In the “Off the Record” commentary format, we offer our employees and volunteers the opportunity to share their experiences from deployments and work with CADUS, as well as their thoughts on humanitarian response, politics and society. The comments do not necessarily have to reflect the opinion of the CADUS organization.

I’ve spent enough nights under mosquito nets and sat through long, often circular UN coordination meetings to know this: if we want to reach people in crisis faster, with less waste, and without burning out the sector, we can’t keep doing things the way we always have. We must innovate. Clinging to the old ways won’t cut it.

That’s why AI has been on my mind a lot. Not the dramatic sci-fi “robots take over” version. I mean the gritty, unglamorous, sometimes glitchy AI that could actually help us get food to a flood-hit village before the road washes out, or spot a disease outbreak before the clinic is full.

At the AI for Good conference this year, the message was clear: humanitarian needs are exploding, funding is shrinking, and we need every tool we can get. AI is already here, it’s just not always where you’d expect.

Where AI is Already Helping

Forget the hype. Some of the best humanitarian AI work isn’t flashy at all, here are a few grounded examples:

Seeing the invisible: Microsoft’s Precision Populations project, in collaboration with Planet Labs, uses AI and high-frequency satellite imagery to map where people actually live in places that haven’t had a census in over a decade. When a cyclone hits, that means we’re not guessing where to send help.

Preventing blindness: A simple phone camera plus an AI model can help a nurse in a rural clinic spot retinopathy of prematurity in newborns, the biggest cause of preventable childhood blindness.

Accelerating humanitarian decision-making: Mercy Corps’ generative AI tool called Methods Matcher, helps aid workers to quickly access evidence-based answers from an archive of past projects and research. From gauging cash aid values during inflation to thinking through nutrition strategies, the tool supports smarter, quicker decisions in the field.

These aren’t hypothetical future scenarios; they’re real, operational today, and saving both money and lives. I’ve used AI-powered diagnostic tools in outbreaks, not perfect, but sometimes the line between “we caught this in time” and “we’re already too late.”

But It’s Not All Sunshine and Algorithms

Humanitarian contexts are messy. People are displaced, vulnerable, traumatised, have lost homes, families and are often targeted. Drop an AI tool into that mix without thinking, and you can do serious harm. Here are some pressing risks:

Data protection: In a war zone, the wrong data leak can literally get someone killed. Switzerland’s Humanitarian Data and Trust Initiative is one step in the direction of protecting sensitive data in war zones.

Language bias: Most AI is trained in English, leaving many languages and dialects poorly served. So good luck if your community speaks Tigrinya or Rohingya.

Digital divides: The ITU estimates that 2.6 billion people still lack access to the internet. No internet means no AI benefit, no matter how good the model is.

“Parachute AI”: Tools built in shiny offices, far from the field, dropped in with little local input, and gone six months later. If communities aren’t part of the design from the start, AI won’t just fail, it could backfire. We need to move from parachute AI to partnership AI, where affected communities, local NGOs, and national institutions are co-designers, not just “data points”.

Partnerships and Sustainability

AI won’t transform humanitarian action if every project remains a one-off pilot. We have too many proofs-of-concept and not enough long-term systems. If we want AI to stick, we need to fund the infrastructure, the data systems, the training, the maintenance, not just the experiment. The UN’s Complex Risk Analytics Fund is showing how pooling resources across agencies can make that happen. Sexy? Not really. Useful? Absolutely.

When Private Sector Partnerships Work

Done right, private-humanitarian partnerships can be magic. Infosys, for instance, partnered with the NGO Street Child in Ukraine in Ukraine to adapt its e-learning platform for war-displaced students and teachers. They tailored content in Ukrainian, added cyber-security modules, and even built digital learning hubs near the front line. It’s not just aid, it’s resilience.

Five Rules I Try to Live By When It Comes to AI

1. 1. Start with the need, not the tech: If it doesn’t solve a real problem, it’s just digital clutter.
   2. Ethics first: Privacy, consent, and transparency aren’t optional.
   3. Invest in people: Especially local talent. A model that needs a Silicon Valley engineer to run it will die in the field.
   4. Share openly: What works, what doesn’t, and yes, even the failures.
   5. Think big, start small: Prove it in one camp before you roll it out to a whole country.

Why This Matters

AI isn’t going to “save” humanitarian aid but it can help us spend our scarce resources and the trust of the people we serve more wisely. If we get it right, AI could mean faster, fairer, more effective aid. If we get it wrong, we risk turning a tool for humanity into just another way to leave people behind. And in this line of work, leaving people behind isn’t an option.

Dr. Hisham Abdulaziz is a medical doctor and humanitarian worker, currently working as Country Manager for Cadus. He holds a Master’s degree in International Health and is a guest lecturer at Charité – Universitätsmedizin Berlin. Over the past decade, he has worked with Médecins Sans Frontières, the World Health Organization, and Ada Health where he was involved in designing, developing, and testing medical AI models. His work bridges the fields of health, humanitarian response, and artificial intelligence, with a passion for leveraging technology to improve care in crisis settings .