Encoding/decoding in artificial intelligence: Global AI and local languages

Abstract

Amid artificial intelligence rapidly transforming our social life, this article uses Stuart Hall’s encoding/decoding model to critically assess the design and use of AI-based language technology for developing a more inclusive and fair global AI system. Generative AI predominantly trained on data in high-resource languages like English encodes and reinforces dominant ideologies as a preferred reading of the global society. In sub-Saharan Africa, the post-colonial linguistic politics and oral traditions contribute to indigenous African languages, despite being spoken by millions, as being low-resourced, which in turn exacerbates the marginalization of African linguistic and epistemic systems in the AI age. In addition to efforts to facilitate inclusive encoding using datasets featuring local contributions, another way forward is to emphasize user agency in critical decoding of meaning structures embedded in global AI and in the use of technology, allowing for the possibility of negotiated or oppositional positions that resist the dominant.