Visual Orientalism in the AI Era: From West-East Binaries to English-Language Centrism

TL;DR

The paper introduces Visual Orientalism as an AI-driven dual standard that links Western political-modern symbols with Eastern cultural-traditional symbols. Using a multi-model, AI-assisted content-analysis of 396 images across 12 countries and 3 T2I systems, it develops Symbolization and Visual Orientalism Indices to quantify symbol use and framing. Findings reveal a shift from traditional West-East binaries to English-language centrism, with the United States showing political dominance while many non-Western countries face cultural exoticization; gender and festival contexts further intensify these biases. The study highlights governance and data-structure implications, arguing for curated training data and inclusive, interdisciplinary approaches to mitigate algorithmic Visual Orientalism.

Abstract

Text-to-image AI models systematically encode geopolitical bias through visual representation. Drawing on Said's Orientalism and framing theory, we introduce Visual Orientalism - the dual standard whereby AI depicts Western nations through political-modern symbols while portraying Eastern nations through cultural-traditional symbols. Analyzing 396 AI-generated images across 12 countries and 3 models, we reveal an evolution: Visual Orientalism has shifted from traditional West-versus-East binaries to English-language centrism, where only English-speaking core countries (USA and UK) receive political representation while all other nations - including European powers - face cultural exoticization. This algorithmic reconfiguration operates through automated framing mechanisms shaped by English-language training data dominance. Our findings demonstrate how AI systems function as agents of cultural representation that perpetuate and intensify historical power asymmetries. Addressing Visual Orientalism requires rethinking of algorithmic governance and the geopolitical structures embedded in AI training data.

Figures (8)

• Figure 1: Main Findings of Visual Orientalism in AI Image Generation. (A) Symbol distribution: Western countries cluster in the lower-left region with higher political representation relative to Eastern countries, while Eastern countries concentrate in the lower-right region showing strong cultural dominance with minimal political representation. (B) Flag effect: Western countries M=1.03 vs. Eastern M=0.16, $d=1.33$, $p=0.031$. (C) Festival effect: Western festivals are framed as more modern and political, while Eastern festivals are depicted as more traditional and cultural, with a large modernity gap between regions ($d=1.23$, $p<0.001$). *$p<0.05$, **$p<0.01$, ***$p<0.001$.
• Figure 2: Visual Orientalism Index (VOI) by Country: Evidence of English-Language Centrism. The VOI (ranging from -1 to +1) combines Political Sovereignty Index (PSI) and Cultural Exoticization Index (CEI) to measure the balance between political representation and cultural exoticization. Only the United States (VOI = +0.170) shows clear political representation dominance, while the United Kingdom (VOI = +0.024) shows near-balance. All other countries, including France (VOI = -0.437) and Germany (VOI = -0.488), show cultural exoticization dominance. This pattern suggests a shift from traditional West-East Orientalism to English-language centrism, representing the further polarization of Orientalism in the AI era. Countries are color-coded by region: United States (dark blue), Other Western (light blue: UK, France, Germany, Australia), and Eastern (orange).
• Figure S1: Gendered Visual Orientalism Analysis. Bar charts show mean values for five coding dimensions (political symbols, cultural symbols, flag appearance, sovereignty representation, and modernity level), comparing East vs. West for women and men separately. Eastern women show the highest cultural symbol count (M = 3.90) and lowest modernity level (M = 2.40), demonstrating compounded orientalization. Western women show higher political symbols (M = 0.47) and modernity (M = 3.12). Error bars represent standard errors. ***$p < 0.001$, **$p < 0.01$, *$p < 0.05$.
• Figure S2: Weighted Composite Indices Comparison. Western countries show higher PSI (M = 0.225 vs. M = 0.048, $d$ = 1.45) and lower CEI (M = 0.419 vs. M = 0.558, $d$ = -1.65), resulting in less negative VOI (M = -0.194 vs. M = -0.510, $d$ = 1.58). The VOI effect size is 76% larger than the original Symbolization Index ($d$ = 0.90). Error bars represent standard errors. ***$p < 0.001$, **$p < 0.01$, *$p < 0.05$.
• Figure S3: Model-Level Bias Comparison. Midjourney shows the highest bias (average $|d|$ = 1.54), followed by NanoBanana ($|d|$ = 1.27) and GPT-Image-1 ($|d|$ = 1.08). All models show consistent bias direction. Error bars represent 95% confidence intervals. ***$p < 0.001$, **$p < 0.01$, *$p < 0.05$.