Answers are grounded in this paper and cite their sources.
Cross-cultural study compares AI-generated images of older adults, highlighting cultural biases and representation issues.
This cross-cultural study examines 400 AI-generated images from Doubao (v1.47) and ChatGPT-4o, comparing how each model visually represents Chinese and American older adults. The findings reveal distinct platform-specific patterns. Doubao produces polarized depictions, in which Chinese older adults are frequently shown in culturally specific settings (e.g. tea houses, community centers), while their American counterparts more often appear in commercial or transport spaces and are associated with negative expressions (22 out of 100; 5 out of 100 for Chinese). ChatGPT-4o generates uniformly optimistic portrayals across both groups (e.g. over 60% positive expressions) while systematically erasing social vulnerability. Both models show limited racial diversity in American depictions (88% White in ChatGPT-4o; 100% White in Doubao). These patterns suggest that AI image generators may reproduce particular cultural associations under the guise of algorithmic neutrality. The study calls for bias auditing that moves beyond detecting overtly negative content to addressing covert structural biases, including systematic omission and rigid cultural templating.
No takes yet. Share an insight, caveat, or question.
Xiaoyun Zhong (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: