ChatGPT’s interpretation of contested memoryscapes favours the voice of current governments, capitalism and the far-right

Abstract

This study examines political biases in ChatGPT-4o’s interpretations of memoryscapes linked to dictatorial regimes in Spanish-speaking countries. Using a least-to-most prompting model and a Discourse-Historical Approach (DHA), we analysed responses in both Spanish and English about three key memorial sites: the Valley of the Fallen (Spain), the ESMA Museum and Site of Memory (Argentina), and the Monument to Salvador Allende (Chile). Findings show a preference for dominant, capitalist, and government-aligned perspectives, often reinforced by far-right rhetoric, particularly in Spanish language. The research highlights generative AI’s influence in shaping heritage narratives and the challenges of ensuring unbiased and multi-vocal outputs. This study advances the fields of cultural heritage, digital humanities, and education, while also shedding light on the wider implications of ChatGPT’s outputs for memory politics and democratic governance.