Amidst funding cuts, burnout and the increasing urgency of crisis situations, generative AI has emerged as a highly attractive tool in the humanitarian sector.
This World Humanitarian Day, it’s worth remembering that the transformative potential of generative AI has two sides. As such, both individuals and organisations must take the time to fully consider the broad ethical dimensions of its use.
At its core, artificial intelligence (AI) is a pattern-recognition tool. Generative AI (GenAI) is a form of AI that predicts what is likely to come next in a pattern to generate new content. The most popular GenAI tools include ChatGPT, Claude and CoPilot.
Like many other sectors, GenAI is already widely used in the humanitarian sphere. A 2025 report, which surveyed 2,539 humanitarian workers from 144 countries and territories, found that 69% of humanitarian workers use GenAI. Common tasks include developing reports and proposals, writing emails, and translation. Notably, however, only 22% of humanitarians reported formal AI policies in their organisation.
Exploring the benefits
Efficiency
Such high usage among humanitarians makes sense. In a workforce that faces frequent burnout, GenAI can take on everyday tasks, like summarising meetings and writing emails, to free-up time for more important work. It can also streamline more labour-intensive tasks such as writing proposals and reports.
Using GenAI may even lead to quicker, more effective responses to crisis situations. Fast-tracked data analyses, needs assessments and funding proposals can significantly reduce the time between a crisis and a response. And 38% of humanitarians felt that AI use led to better decision-making. Whether this translates into better impacts, however, is unclear.
Supporting locally-led action
The theme for this year’s World Humanitarian Day is ‘Strengthening Global Solidarity and Empowering Local Communities’ . This links very well to one of the most compelling benefits of GenAI: its availability to local humanitarians and potential for creating a more equal humanitarian landscape.
Popular GenAI tools are available via a web-browser or app, are often low or no-cost, and support a range of languages (to various degrees). These features allow local organisations with limited resources to access the tools, where they have the potential for even more transformative effects.
They can be used to develop advocacy resources to amplify local causes. They can find funding opportunities. They can simplify and translate complex proposal requirements that are often only available in English. And they can help write proposals and reports, which also typically need to be in English.
From the other end of the scale, GenAI tools can also be used by international organisations and funders to streamline their search for partner organisations and support locally-led action.
Its harms and potential risks
Despite the potential that AI holds, it’s also responsible for significant harm and comes with high levels of risk. Given the highly complex and sensitive contexts humanitarians operate within, this should be of particular concern.
Data protection and sovereignty
Data leaks are not a new phenomenon. Just this year, a cyber-attack on the World Food Program exposed the personal information of some 600,000 households in Gaza.
The risks of this happening again are significantly amplified by the use of GenAI. Traditional security controls are not designed to address the unique threats posed by these tools. For example, prompt injections can make a GenAI tool ignore its guardrails and access and share information it shouldn’t be able to.
Furthermore, specialised AI data centres, which store account and prompt information from various GenAI tools, are located in only 16% of the world’s nations—the vast majority in the US, China and the European Union. This raises significant privacy concerns for affected populations in the Global South, as what can be done with the stored data is under the jurisdiction of the host country.
Crucially, the best way to minimise potential data-related harm is to anonymise all data input.
Inaccuracies and biases
As the outputs of GenAI are the result of little more than pattern recognition and prediction, these tools frequently produce both false and biased responses.
This is most noticeable in ‘hallucinations’: outputs that look statistically credible but are, nevertheless, false. Without caution, such responses can lead to misinformation, flawed data analysis and inaccuracies in reporting.
Potentially less noticeable is the bias a GenAI tool can replicate from its training data. This can lead to decisions that reinforce social inequalities and discrimination. Furthermore, Popular GenAI models are trained on data that largely originates from the United States and is almost entirely English. This can lead to culturally inappropriate responses. In a sector that already has an issue with top-town, Western-led systems of thinking, such biases could be particularly problematic.
Entrenching top-down systems
While recognising the benefits these tools offer local humanitarians, they may also entrench the top-down, Western-led thinking that the sector is currently grappling with. As GenAI improves efficiency, the potential for job losses and reduced opportunities for local employees increases. Additionally, the ability of these tools to help with proposals and reports discourages change to the inherent inaccessibility of these requirements. An alternative humanitarian system that supports more flexible approaches is possible; AI use may be pushing the sector further away from that alternative.
Impacts on cognition and creativity
Considering the highly complex situations in which humanitarians work, it’s important to acknowledge the emerging research into the effects of AI use on the brain.
The most prominent of this research looks into overreliance on AI tools. These findings have shown that ‘cognitive offloading’, in which users delegate mental tasks such as problem-solving and analytical reasoning, may reduce critical thinking skills.
Equally interesting are the results of a study looking into the effects of GenAI on creativity. While AI use may increase the creativity of an individual, on a collective scale, creativity may be reduced as the ideas it produces remain relatively consistent.
The potential impacts of these findings are particularly relevant for AI use in needs assessments, data analysis and program design, where critical thinking, creativity and diversity of thought are hugely important.
Environmental harms
The notion that AI is harmful for the environment is relatively well known. It relies on critical minerals and rare elements and is already responsible for concerning levels of water usage, greenhouse gas emissions, land clearing and hazardous electronic waste.
But its impacts are rising fast.
In 2030, its electricity use is predicted to nearly triple, water consumption could match the annual domestic needs of 1.3 billion people and land use could expand to roughly twice the size of Jakarta.
And contrary to some debates, day-to-day use is the main cause, responsible for 80-90% of total energy demand.
Of course, similar, and much worse, environmental impacts can be attributed to other technologies and innovations. This does not negate the harms of AI. Conversely, it situates it in a context in which the planet, and humanity, cannot afford any additional burdens.
And humanitarians must also recognise that it’s the Global South that is first impacted by these burdens, which is already facing the brunt of climate change.
What does all this mean?
The question of GenAI is undoubtedly one of the most difficult dilemmas humanity is currently facing. The way forward presents various options.
Lean into GenAI use and count on the development of governing mechanisms.
Refuse to adopt it as long as the risks and harms remain so high.
Or find a middle ground by exploring alternatives, such as AI for the common good, open-source models, green AI powered by renewables, or companies headquartered in the Global South.
For humanitarians who choose to adopt GenAI, in any form, conditions must be implemented to ensure it does not compromise the humanitarian principles or risk harm to the people they serve. Organisations must develop formal AI policies with strict guardrails. All employees approved to use AI, and even those who are not, should receive comprehensive training on the risks of AI and its responsible use. And, finally, all AI use should involve direct engagement with impacted communities, and personal data should never be shared without informed consent.
This World Humanitarian Day, it’s time to stop using AI without considering both sides of its transformative potential. It’s time humanitarians make an informed decision.
Images:
Anton Maksimov: https://unsplash.com/photos/green-pink-and-blue-lights-hotybva3hqi
Katrin Bolovtsova: https://www.pexels.com/photo/a-golden-balance-scale-beside-a-laptop-6077797/
