Contemporary AI development rests on assumptions about intelligence, rationality, and human values that are not culturally neutral, according to a new preprint on arXiv. The paper argues that these assumptions come from a narrow set of cultural perspectives, and that this narrowness limits the ways AI could be designed to benefit people globally.
The authors contend that what is considered rational or valuable in AI systems is shaped by specific cultural contexts. Treating these views as universal may cause researchers to overlook alternative approaches that could be more suitable for other societies and needs.
As the title suggests, the paper turns to anthropological concepts to decode the hidden assumptions embedded in AI. By making these assumptions visible, the authors hope to broaden the range of possible AI designs and move toward more globally beneficial systems.