DATA ORIENTALISM IN THE MACHINE: A POSTCOLONIAL AND VISUAL DISCOURSE ANALYSIS OF AI-GENERATED CLIMATE IMAGERY OF PAKISTAN

Abstract

Generative AI is increasingly used to visualise the climate crisis, yet few studies have examined the ideological work these systems perform when representing non-Western contexts. This paper conducts a postcolonial Visual Discourse Analysis of AI-generated climate visualisations produced by the Llama 4 model. Focusing on a corpus generated with the geographic tag “in the context of Pakistan”, the paper analyses a purposive sample of three contextualised images along with three comparator images. Drawing on Edward Said’s concept of Orientalism, Homi Bhabha’s account of colonial discourse, and postcolonial ecocriticism, this study finds that the GenAI resorts to a typological inventory of crisis iconography: dust, flood, and the suffering child, while erasing Pakistani institutional agency and geographic diversity. Read alongside comparator imagery, these findings reveal a colonial spatial logic coding suffering onto the Global South, solution onto the Global North, and governance onto a racially homogeneous professional elite. This reproduction, at scale and by default, of colonial representational conventions embedded in the training data of generative systems are manifestations of  Data Orientalism. The paper concludes with recommendations for counter-mapping the global visual economy of climate communication.