Beyond Efficiency: Reframing AI Hallucination and Refusal as Aesthetic-Moral Gestures in Cinematic Practice

Abstract

Generative artificial intelligence (AI) in cinematic practice is predominantly analysed through the lens of technical efficiency. This study proposes a critical reframing of algorithmic anomalies in film production, and its aesthetic and moral dimensions. Specifically, the research reframes AI hallucination and refusal, often dismissed as technical flaws, as aesthetic-moral gestures. The study conducts a textual analysis of AI-generated student short films, and applies cognitive and aesthetic-philosophical frameworks from Boden, Levinas, and Sontag. This study proposes a conceptual framework and provides illustrative case studies to explore how emerging creators negotiate cinematic authorship through these algorithmic behaviours. Hallucination is examined as a “creative deviation” that may prompt transformational curation within specific production contexts, where unpredictable visual anomalies are reinterpreted by creators as signs of memory, fragility, or affective ambiguity rather than as direct evidence of audience reception. Conversely, algorithmic refusal is examined as an “ethical pause” and a modern cinematic ellipsis. By restricting explicit depiction, this refusal prompted the selected creators to reconsider subjective moral boundaries and employ symbolic omission and hybrid workflows. Within these pedagogical contexts, hallucination and refusal form a dialectic of deviation and restraint. Generative AI is conceptualised not just as a tool for mechanical replication, but as an interactive catalyst expanding interpretive possibilities. Machine error and algorithmic limitation are thus recontextualised as opportunities for localised human inquiry and moral creativity.