Cyprus as AI Saw It: Digital Colonialism and AttnGAN Text to Image Synthesis

Abstract

This paper discusses an ongoing project pivoting on an AttnGAN Text to Image synthesis as the means to critically comment on affairs of bias in historical and digital colonialism. It zooms in on Cyprus — a geographic region that is largely underrepresented in AI programming — as seen through the lenses of 19th century colonialism. A well-known AttnGAN pipeline is appropriated and trained over a DIY dataset to produce images that more genuinely reflect the (post-)colonial Cypriot reality — thus giving agency to a geographic region, the cultural and historical idiosyncrasies of which are generally not reflected in AI programming. Sample imagery is presented and the results of a qualitative survey are discussed in some detail. In this fashion, the authors intend to contribute empirical data to the ongoing discussion about digital colonialism, and to initiate a broader conceptualisation process about methodologies to decolonise current affairs in AI research. Technical, critical and ethical implications of the above are further discussed, with references to Digital Colonialism and AI-driven Text-to-Image synthesis.