Reproduce “AI Sees Your Location, But With A Bias Toward The Wealthy World” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jingyuan Huang, Jen-tse Huang, Ziyi Liu, Xiaoyuan Liu, Wenxuan Wang, Jieyu Zhao

A study revealed that Visual-Language Models (VLMs) consistently show geographical biases, favoring economically developed, populous, and Western-aligned regions when inferring location from images. The FAIRLOCATOR benchmark quantified these disparities, highlighting concerns regarding user privacy and the potential for perpetuating global information inequalities.

Key insights

  • Visual-Language Models (VLMs) exhibit remarkable geolocation capabilities, significantly outperforming human participants in identifying locations from images.
  • Three distinct types of geographical biases were uncovered: favoritism towards well-known cities, uneven accuracy across diverse economic and population-based regions, and spurious correlations associating urban scenes with developed nations.
  • VLMs showed increased accuracy on user-uploaded images without metadata, amplifying privacy risks by enabling precise location inference from personal photos.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.