Abstract
This study investigates the explainability of generative AI artifacts and proposes a media-archaeological framework for criticism. Explainable artificial intelligence (XAI) has primarily focused on making algorithmic decisions understandable to human users; however, the explainability required for AI-generated art cannot be reduced to mathematical clarity. Generative artifacts are produced from relations among datasets, models, input conditions, platform interfaces, archives, and curatorial choices. This study conceptualizes "media-archaeological XAI" not as a technical extension of XAI but as a critical appropriation of its demand to explain the origins and limitations of AI outputs. Based on Wolfgang Ernst’s operativity, Friedrich A. Kittler’s media conditions, and Wendy Hui Kyong Chun’s account of software and interfaces, it creates three layers of critical explainability and a five-stage analytical framework that encompasses generation conditions, operativity, archival genealogy, interface analysis, and critical explanation. The study defines this framework as a theoretical proposal whose applicability and inter-critic consistency should be investigated in future case-based research. |