The global circulation of literature is undergoing a fundamental transformation as artificial intelligence systems increasingly mediate the translation, selection, and reception of texts across linguistic borders. This article develops the concept of AI-mediated canonicity to describe a new regime of literary canon formation in which large language models, neural machine translation systems, and platform algorithms, rather than human gatekeepers alone, determine which texts achieve global visibility and canonical status. Drawing on a mixed-methods design combining computational corpus analysis with comparative case studies of AmazonCrossing and AI-translated Chinese science fiction, the study examines how algorithmic mediation reshapes the conditions of literary value. Findings demonstrate that human translators consistently outperform large language models across all evaluated language pairs, with professional evaluators selecting human translations at rates of 79–87% compared to 8–14% for LLMs. Stylistic analysis reveals that AI translation systematically reduces variation, producing a homogenized register that flattens aesthetic particularity. Examination of LLM-generated literary canons exposes overwhelming Western, White, and male biases that amplify rather than merely reflect existing inequalities. Platform algorithms increasingly determine visibility through engagement metrics rather than editorial judgment. The article argues that the definition of world literature as "writing that gains in translation" requires substantial revision to account for algorithmic production and mediation. Four constitutive mechanisms of AI-mediated canonicity are identified: algorithmic selection, translation automation, corpus bias, and generative adaptation. The study concludes that the future of world literature depends on critical intervention in the algorithmic infrastructures that increasingly determine which texts travel and which remain in place, calling for a politics of technological transparency and accountability.
Copyrights © 2026