Decolonizing the literary AI in the age of LLMs and digital neocolonialism

Large Language Models (LLMs) are usually considered neutral technological advancements. However, critical digital studies increasingly emphasize the need to challenge their potential to perpetuate colonial power structures in cyberspace. This paper argues that LLMs function as powerful apparatuses of digital neocolonialism. It aims to diagnose this phenomenon within the field of literary AI and to propose a decolonial framework for its future development. This study demonstrates how the protocols of extracting and processing data privilege Western epistemologies in a systematic manner. Then, it develops a conceptual framework for the praxis of decolonial AI based on the principles of reciprocity and epistemic justice. The analysis reveals that the extractivist data collection utilized by dominant LLMs treats cultural and linguistic data as territory for appropriation, privileging the Western literary canon and erasing marginalized languages and traditions. This has led to linguistic homogenization and epistemic injustice as well as the imposition of aesthetic standards of the global West. In response, the proposed decolonial framework has necessitated a paradigm shift from extraction to reciprocity, which involves community-led data governance. Furthermore, AI should be used as a collaborative, co-creative tool by literary writers and researchers. As a further decolonial step, Eurocentric evaluative criteria in this field must be reformed in concrete ways. The decolonial approach advanced in this paper seeks to fundamentally reposition literary AI. The ultimate goal of this repositioning is to foster a pluriversal aesthetic and epistemic framework.