A recent discussion on the use of AI-generated visuals in the humanitarian sector highlights concerns about charities creating synthetic images of poverty and suffering. While ethical risks persist, the conversation also opens possibilities for using AI to produce more empowering representations. One overlooked opportunity is the potential for organisations to collaborate directly with communities to co-create imagery. Participatory photography projects, such as WaterAid’s initiative in Sierra Leone, already demonstrate how communities can lead their own storytelling. Extending this approach with AI could allow local groups to train models on images rooted in their own cultural and social contexts, challenging long-standing Western narratives.
This participatory approach raises new ethical questions about who selects the communities involved and who provides the training. Yet organisations like Fairpicture and EveryDay Projects show that these barriers can be addressed. Evidence also suggests that AI image generation can improve when training data becomes more inclusive. Researchers found in 2023 that Midjourney failed to invert stereotypical global health visuals, but running the same prompts one year later showed measurable progress, underscoring the importance of diverse datasets. If local organisations had access to AI tools, they could generate large volumes of authentic visuals that reshape how AI systems portray the Global South.
Current AI imagery often reinforces harmful stereotypes because models are predominantly trained on historical Western visuals. But this can be changed. If major tech companies funded community organisations to produce thousands of culturally grounded images each week, they could create a steady flow of counter-narratives that influence global datasets. The scale is achievable—just a fraction of the billions invested in AI infrastructure could meaningfully address representational bias. Ultimately, the core issue is not whether AI imagery is good or bad, but who controls the tools and whose perspectives define the outputs. For ethical and accurate representation, communities themselves should have that control.
