We derive a conceptual bridge between technical concepts from the deep learning literature and natural metaphors for international development. We start with a rather technical review of four of the characteristic traits of deep learning technologies: representation, reuse, robustness, regularization (4Rs of deep learning). Based on the empirical evidence of 24 case studies, we derive four characteristics of the use of AI4D that align with the four technological traits, namely development foci on local and distance intelligence, and mirrored and detailed reality representations. In isolation, each one of the identified issues presents a plethora of opportunities to contribute to international development, especially to the attainment of the Sustainable Development Goals. However, in combination, they create a clear tension between a looming threat of a hegemonic intelligence indoctrination pushed by global economies of scale and the potential promise to not only honor but also celebrate local diversity with the help of flexible AI designs.