Cultural Bias and Technical Glitches in Artificial Intelligence (AI) Video Production for Higher Education: From Prompt to Screen

Abstract

The rapid evolution of Generative Artificial Intelligence (AI) has extended beyond text and static imagery into the realm of realistic text-to-video synthesis. This article critically examines the practical and technical challenges of utilizing the latest generative video models (specifically the Google Veo model) for higher education content creation. While the technology offers significant potential for accelerating creative workflows in higher education, its application is constrained by distinct limitations. Adopting a Practice-based Research methodology, this study analyzes a corpus of 15 experimental video projects (n=10 convocation promotions; n=5 academic program promotions) produced using an AI-assisted workflow. Findings highlight three primary categories of limitations; (i) Cultural and Regional Bias, where the model hallucinates foreign cultural elements (e.g., Indonesian demographics) over local Malaysian contexts; (ii) Linguistic and Phonetic Inaccuracies, specifically the inability of Text-to-Speech engines to process local dialects and phonemes; and (iii) Physical and Logical Hallucinations, such as defying laws of physics or adding unauthorized visual artifacts.