aiqbench analysis
This research reveals that frontier multimodal models exhibit a systematic cultural bias, with their "neutral" baselines aligning more closely with European and North American perspectives than other global regions. While prompting can shift AI outputs toward specific regional human descriptions, it fails to capture the full diversity of human perception, instead flattening demographic patterns and introducing sentiment-based self-favouring bias. These findings highlight a critical risk for deploying LLMs in cross-cultural urban evaluation tasks where objectivity is assumed but not guaranteed.
Culturally uneven urban perception in large language models
arXiv:2604.20048v3 Announce Type: replace Abstract: Large language models (LLMs) are increasingly used to describe and evaluate cities, yet the cultural structure of their urban judgments remains understudied. Here we introduce a measurement framework for testing whether LLM-based urban perception is culturally neutral, using a globally stratified street-view image dataset. Open-ended descriptions and structured scores generated by three frontier multimodal models all show that the neutral baseline lies closer to regional framings associated with Europe and North America than to other cultural framings. Comparisons between AI and human urban perception further show that prompting can move AI responses closer to specific regional human descriptions, but fails to recover the variety and diversity of human responses, flattening observed demographic patterns and introducing sentiment-based self-favouring bias. These results indicate a systematic risk in treating AI as a neutral tool for urban tasks, especially when model outputs are used to compare, evaluate or represent cities across cultural contexts.
Abstract:Large language models (LLMs) are increasingly used to describe and evaluate cities, yet the cultural structure of their urban judgments remains understudied. Here we introduce a measurement framework for testing whether LLM-based urban perception is culturally neutral, using a globally stratified street-view image dataset. Open-ended descriptions and structured scores generated by three frontier multimodal models all show that the neutral baseline lies closer to regional framings associated with Europe and North America than to other cultural framings. Comparisons between AI and human urban perception further show that prompting can move AI responses closer to specific regional human descriptions, but fails to recover the variety and diversity of human responses, flattening observed demographic patterns and introducing sentiment-based self-favouring bias. These results indicate a systematic risk in treating AI as a neutral tool for urban tasks, especially when model outputs are used to compare, evaluate or represent cities across cultural contexts.
Computation and Language (cs.CL); Computers and Society (cs.CY)
arXiv:2604.20048 [cs.CL]
(or arXiv:2604.20048v3 [cs.CL] for this version)
Full text under the publisher's open licence · 165 words
Related coverage
NVIDIA and Japan Bring Full-Stack AI and Robotics to Every Industry
NVIDIA · Jul 15, 2026
Amazon Research Awards recipients announced
Amazon Science · May 27, 2026
UrbanWell: Benchmarking Multimodal Large Language Models for Spatio-Temporal Urban Wellbeing Analytics
arXiv · Jun 16, 2026
34 Amazon Research Awards Build on Trainium recipients announced
Amazon Science · Aug 5, 2026