Reproducing Coloniality Through AI: A Case Study of LLM Integration in Pakistani Organisations

Abstract

As large language models become embedded in professional life across the Global South, questions about whose knowledge they carry and whose they erase have become impossible to ignore. Pakistani organisations are adopting these tools fast, within a context shaped by colonial language hierarchies, structural underrepresentation of Urdu in training data, and no community-led infrastructure for cultural knowledge. This thesis examines how LLM integration in Pakistani organisations reproduces or challenges coloniality, and proposes a service design intervention in response.

Employing a qualitative critical case study approach grounded in a critical interpretivist and decolonial paradigm, the study draws on semi-structured interviews with design and marketing professionals in Karachi, supplemented by interactive visual elicitation activities. Data was analysed using thematic analysis and critical discourse analysis, interpreted through the frameworks of Quijano, Mignolo, Couldry and Mejias, Fricker, and Mohamed and colleagues.

The findings reveal four interconnected patterns: LLMs framed simultaneously as neutral tools and culturally loaded infrastructure, a growing dependency on systems that can fail silently, the reproduction of English as professional infrastructure at the expense of Urdu and Pakistan's regional languages, and a pattern this thesis names recognition without transformation, where structural conditions are identified but that identification changes nothing. In response, the thesis proposes an open source, community governed cultural knowledge repository for AI, designed to reduce the invisible overhead Pakistani professionals absorb alone. Decolonising AI infrastructure requires not just better tools but the collective infrastructure that makes acting otherwise possible.