Text-to-image generative AI systems exhibit significant limitations when engaging with under-represented domains, including non-Western art forms, often perpetuating biases and misrepresentations. We present a focused case study on the generative AI system DALL-E 3, examining its inability to properly represent calligraphic Arabic script, a culturally significant art form. Through a critical analysis of the generated outputs, we explore these limitations, emerging biases, and the broader implications in light of Edward Said's concept of Orientalism as well as historical examples of pseudo-Arabic. We discuss how misrepresentations persist in new technological contexts and what consequences they may have.
@article{arxiv.2502.20459,
title = {Broken Letters, Broken Narratives: A Case Study on Arabic Script in DALL-E 3},
author = {Arshia Sobhan and Philippe Pasquier and Gabriela Aceves Sepulveda},
journal= {arXiv preprint arXiv:2502.20459},
year = {2025}
}