Today, on World Humanitarian Day, we honour the people who dedicate their lives to alleviating suffering and protecting human dignity. However, as crises grow more complex, the humanitarian system is being reshaped by one of the most transformative technologies of our time: artificial intelligence (AI).
Across the world, development and humanitarian agencies are experimenting with AI to target assistance, handle grievances, monitor programs, report progress, and even make real-time decisions. Whether its predicting poverty, improving agricultural yields, or optimizing social protection systems, AI is increasingly embedded in global development practice.
Yet for many low- and middle-income countries (LMICs), AI presents a paradox. It offers immense promise, but without robust data and technology infrastructures, it risks deepening inequalities and misrepresenting ground realities.
AI is only as good as the data it learns from. But many LMICs grapple with: Fragmented administrative records; low-quality or outdated surveys; minimal digitization of services; and a lack of interoperability across databases.
When AI systems built on high-income country data, or thin LMIC datasets, are deployed, the risks are clear: Poverty models using satellite imagery may miss context-specific signs of deprivation in rural Africa or South Asia; health tools trained on electronic records may exclude populations relying on paper-based systems or informal providers; agricultural prediction models may ignore indigenous knowledge and climate variability unique to local contexts.
In these cases, AI could obscure realities rather than reveal them—skewing resource allocation and inadvertently marginalizing vulnerable communities.
Despite these risks, agencies are finding innovative ways to deploy AI in humanitarian and development contexts. The World Bank, for instance, has applied artificial intelligence and big data in low- and middle-income countries by using satellite imagery and machine learning to map poverty in Nigeria, enabling more effective targeting of social protection programmes. It has also analyzed over a million crowdsourced tweets to create a 30,000-crash dataset in Nairobi, which is now being used to inform road safety interventions.
UNICEF has similarly embraced AI for social good. In Thailand, it has developed an air quality monitoring model that integrates sparse ground sensor readings with satellite data, while in Tanzania, it has funded innovations such as the Elsa Health Assistant, an AI-powered clinical decision support tool that equips frontline providers to deliver specialist-level paediatric care.
Mercy Corps has also made notable strides by launching Methods Matcher, a generative AI tool that offers aid workers instant, evidence-based guidance during crises. Since its rollout in late 2024, the tool has already been deployed in more than 40 countries. These examples show AI’s potential for poverty reduction, healthcare, education, and crisis response. But they also remind us: context matters.
One pressing concern is the risk of LMICs becoming passive consumers of AI solutions designed elsewhere. If tools are imported without local adaptation, countries may rely on opaque, “black box” systems that lack transparency, accountability, or cultural sensitivity. Instead of fostering resilience, this could reinforce digital dependency.
For artificial intelligence to truly strengthen humanitarian action and development, low- and middle-income countries must not only adopt the technology but also play an active role in shaping it. This requires deliberate strategies that place local needs and perspectives at the centre. First, countries must invest in data infrastructure by modernizing civil registration systems, digitizing public services, and promoting open-data standards while ensuring strong protections for privacy. Equally important is building local capacity—training data scientists, AI engineers, and policymakers who can drive innovation from within.
Developing context-aware AI is another key step, one that involves co-creating tools with local communities so that indigenous knowledge and lived realities are reflected in both design and application. Strong ethical governance is also essential, ensuring that fairness, transparency, and accountability guide AI systems, while giving affected communities a meaningful voice in decision-making. Beyond national borders, South–South collaboration can accelerate progress by enabling countries to share expertise and resources across regions. Finally, hybrid approaches are needed—blending AI-generated insights with surveys, fieldwork, and participatory research to create richer, more reliable solutions. In conclusion, AI has the power to scale impact, unlock new insights, and make humanitarian work more efficient. But without inclusive design and strong local foundations, it risks painting distorted pictures and reinforcing inequality.
On this World Humanitarian Day, the question is: Will LMICs join the AI revolution as active architects of their futures or as passive recipients of imported technologies?
The humanitarian imperative demands the former.
*Haastrup is a UK-based scholar and an experienced data scientist and development specialist with over 15 years of progressive experience in data analysis, artificial intelligence, monitoring and evaluation, and global development.
