En February, the Artificial Intelligence (AI) Action Summit, in Paris, brought together more than a hundred countries to make a declaration “inclusive and sustainable artificial intelligence”. Nine months later, the question remains: when AI is applied to the most vulnerable contexts (refugee camps, indigenous territories, disaster zones), who determines the design, uses and limitations of AI?
The answer shows the gap between the rhetoric of inclusion and the reality of power. AI, which impacts society’s most vulnerable groups, is often designed by others for purposes far removed from their needs.
The example of the Bidi Bidi camp in Uganda illustrates this asymmetry. In 2018, the World Food Program implemented Primes there, a biometric system for aid distribution. For South Sudan’s 270,000 refugees, the choice is simple: hand over their biometric data or not provide food. Some have had their aid suspended due to registration errors, while others have provided their fingerprints without knowing how their data will be used or how to appeal a decision. The system operates in a black box: opaque to those it serves, transparent to those who control it.
There is no local consultation
What is striking is not the technical failure, but rather the absence of structural participation. Designed by Northern experts, the system was never the subject of local consultation. When problems arise, it is the refugees who suffer the consequences without being able to speak out.
This logic is reproduced elsewhere. In the Philippines, more than 15 million indigenous communities are witnessing increased use of AI in their territories: surveillance aims to identify “terrorist group”automated mining, biometrics for access to public services. However, a 1997 law on the rights of indigenous peoples requires free, prior and informed consent (FPIC) before undertaking any project on ancestral lands. AI systems are implemented as if these rights do not exist.
You have 65.95% of this article left to read. The remainder is provided to customers.
