In July, the International Telecommunications Unit (ITU) in partnership with the United Nations will host the inaugural AI for Good summit in Geneva, Switzerland. Convening under the theme AI’s potential to serve humanity, the summit aims to platform conversations on using AI and frontier systems to solve global challenges, developing AI standards to expand opportunities for innovation and digital transformation globally, and facilitate exchanges on effective approaches to AI governance. On a systems level, this convening is a strong coordinator for shared strategies that will guide countries on diffusion of AI across economies to ensure no country is left behind. Yet, the countries from the global majority who carry the highest burden of humanity’s pressing challenges are not worried about AI’s transformative potential but fairness and accountability in the rush for adoption.
An impact report released by AI for Good in 2025 details the multilayered wins of AI and frontier systems including reshaping how communities deliver healthcare, influencing learning for indigenous societies, optimizing early warning systems for disastrous events and strengthening climate resilience. This transformative impact is steering progress towards the achievement of sustainable development goals. But beyond the systems change, mediating decision-making and transforming service delivery, sustainable development is about people and communities who experience development challenges on a daily basis. Sustainable development goals are founded on the principle of decentralising agency to priority populations who are often excluded from decision-making spaces by setting priorities for forgotten crises, expanding reach to communities left out by system failure, and driving people-centered change. Therefore, when the development sector moves too quickly to adopt AI and other frontier systems, the bigger question becomes – are these systems widening the distance between the suffering majority and the development sector? What stereotypes are being (re)produced in the process?
Bilateral donors such as the UK government have adopted a careful approach on the discourse of AI in the development sector. In a recently released White Paper on International Development, detailing an overview of the UK’s approach to global development, the UK government firmly recognises the importance of “innovation and digital transformation in driving progress.” A review by the Oxford Policy Management critiques the UK’s neutral approach, noting that the UK is failing to “fully appreciate the potentially transformative potential of AI and its ability to significantly affect development efforts.” Yet from my perspective, I think that the UK’s careful servings on AI is recognition that immense positive impact by intelligent systems can co-exist with equal negative implications by the same frontier systems. Hence, this should prompt careful considerations before large scale adoption, especially by the development sector which works closely with marginalised communities.
The first consideration is thinking about the data used in training AI. It’s not a secret that one of the critiques surrounding the development sector is its success in leveraging pressing challenges such as poverty, disease burden, and disasters to prolong the lifeline of its operations. AI systems are built and trained within this ecosystem where the poor, hungry, and struggling are the global majority, and the savioristic solution provider is the global North. Case in point, Europe-based researcher, Arsenii Alenichev, found that AI generated tools produced stereotypical images of sickness and poverty. When AI generation tools were prompted to produce images of patients undergoing HIV treatment, Arsenii’s research found that all images generated showed black characters as patients undergoing HIV treatment whereas a prompt on traditional African doctor produced images of white characters. This noir is a product of a system that equates blackness with sickness and poverty, and whiteness as ‘messaic.’ Hence, in making decisions about AI adoption, development actors must scrutinize who and how the intelligent system under consideration was trained. This is because the impact of technology is dependent on who controls it, which in turn sets a clear hierarchy of accountability.
The second consideration is defining who development work is meant for. The people are at the center of development work. In the event that AI mediates development outcomes, who is gaining? Who is being left out and why? University of East Anglia researcher, David Girling, and co-author, Deborah Adesina Wonders just released a report exploring the use of AI imagery in development communications. On the one hand, development organisations argue that it is cheaper to pay for image generation intelligent systems than to hire image curators to take pictures of development beneficiaries. On the other hand, the public receives the AI generated imagery with a lot of skepticism. Rather than focusing on the development cause, the report found that the use of AI imagery shifted discussions towards mundane topics such as the quality of the generated output. This points towards a possible dilution of development priorities, curating a spectacle out of the lived realities of many people from the global majority. Additionally, continued use of synthetic multi-visual products might blur differentiation lines between briefcase charities and real development organizations, leading to erosion of trust, and eventual spillover effects on fundraising goals.
The third and final consideration is the tradeoffs. What is at stake and its long-term implications? Development organizations such as Amnesty International, have been criticised for using AI imagery and videos in advocacy materials. Whereas Amnesty International has defended its intentions citing privacy and preservation of dignity for survivors, such decisions inspire questions on credibility and validation of advocacy priorities. I recently had a conversation with the founder of Duckrabbit and award-winning filmmaker Benjamin Chesterton on using AI in the development sector. Benjamin was a lead voice in calling out Amnesty International’s use of fake images in advocacy materials. According to Benjamin, people’s stories and realities must be handled with dignity. If the argument is using AI to protect one’s identity, Benjamin notes that it should be a sign of “wrong timing” for that particular story. But again the development sector has been accused of holding a tight leash on narratives, seeking to closely control timelines to suit a bigger agenda. This introduces a delicate navigation of ethics, delivering sustainable impact and creating an illusion of doing good.
In summary, AI use in the development sector is not a question of whether its right or wrong but rather transplantation of AI frontiers into development contexts they were never designed for. Development practitioners have a responsibility to safeguard vulnerable communities by thinking critically before adopting a new AI system. There’s also an opportunity to advocate for and develop locally-built AI models which are context-specific and indigenous-led. This is doable because open-source pilot models such as InkubaLM (African languages) and Te Hiku Media (Te Reo Māori) are testimonials that local and participatory AI is possible.