The Un-Curated Past: Cultural Memory and Power in AI-Generated Geopolitical Narratives

Abstract

This study interrogates the epistemological status of geopolitical narratives generated by large language models such as ChatGPT, challenging the prevalent assumption that artificial intelligence constitutes a neutral or objective source of truth. Rather than producing experiential knowledge, these systems operate as probabilistic language models that recombine and reproduce linguistic structures derived from human-generated corpora including books, news media, websites, and social platforms. The paper examines the operation of cultural memory and discursive power within AI-generated texts addressing geopolitical themes, specifically terrorism, border conflict, and religious extremism. Through a qualitative analysis of selected ChatGPT narratives, the study employs a combined methodological framework of Post-Humanist Discourse Analysis (PHD) and Norman Fairclough's three-dimensional Critical Discourse Analysis (CDA). This framework facilitates examination at three interrelated levels: (1) the micro-linguistic features of the generated texts; (2) the interdiscursive relationships between AI outputs and established media and political discourses; and (3) the macro-social significance of ideology and cultural memory in shaping geopolitical representation. The findings indicate that AI systems do not generate original geopolitical meaning but instead statistically reproduce the most dominant and recurrent patterns present in their training data. In doing so, these models function as conduits of cultural memory and as discursive apparatuses embedded within broader systems of power. The paper concludes by arguing for the cultivation of critical literacy in AI engagement, particularly in contexts concerning geopolitics, identity, and collective memory.