AI Sees Your Location, But With A Bias Toward The Wealthy World

TL;DR

This paper evaluates visual-language models' ability to recognize geographic locations from images, revealing significant biases favoring wealthy, densely populated regions and raising privacy concerns.

Contribution

Introduces a benchmark dataset and evaluates VLMs, highlighting regional biases and privacy implications in geographic recognition tasks.

Findings

VLMs achieve up to 53.8% accuracy in city prediction.

Models perform better on developed, dense regions (+12.5%) and worse on less developed areas.

Regional biases include over-predicting certain locations like Sydney.

Abstract

Visual-Language Models (VLMs) have shown remarkable performance across various tasks, particularly in recognizing geographic information from images. However, VLMs still show regional biases in this task. To systematically evaluate these issues, we introduce a benchmark consisting of 1,200 images paired with detailed geographic metadata. Evaluating four VLMs, we find that while these models demonstrate the ability to recognize geographic information from images, achieving up to 53.8% accuracy in city prediction, they exhibit significant biases. Specifically, performance is substantially higher for economically developed and densely populated regions compared to less developed (-12.5%) and sparsely populated (-17.0%) areas. Moreover, regional biases of frequently over-predicting certain locations remain. For instance, they consistently predict Sydney for images taken in Australia, shown by the low entropy scores for these countries. The strong performance of VLMs also raises privacy concerns, particularly for users who share images online without the intent of being identified. Our code and dataset are publicly available at https://github.com/uscnlp-lime/FairLocator.