GeoBS: Information-Theoretic Quantification of Geographic Bias in AI Models

Zhangyu Wang, Nemin Wu, Qian Cao, Jiangnan Xia, Zeping Liu, Yiqun Xie, Akshay Nambi, Tanuja Ganu, Ni Lao † \dagger, Ninghao Liu † \dagger, Gengchen Mai † \dagger · 2025

Read GeoBS: Information-Theoretic Quantification of Geographic Bias in AI Models by Zhangyu Wang, Nemin Wu, Qian Cao, Jiangnan Xia, Zeping Liu, Yiqun Xie, Akshay Nambi, Tanuja Ganu, Ni Lao † \dagger, Ninghao Liu † \dagger, Gengchen Mai † \dagger (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-23482.txt.

Description

AI models can treat places unevenly, yet geographic bias has drawn far less study than social bias. Zhangyu Wang and 10 co-authors build GeoBS, a scoring framework grounded in information theory. Older measures fit single models or hide where the bias sits on the map. GeoBS is meant to work for any model and to show the spatial pattern itself.

Three new scores in the framework track how bias changes with map scale, distance, and direction. Tests span three tasks, eight data sets, and eight models. Both task-built GeoAI models and general foundation models showed forms of geographic bias. The authors aim to make spatial fairness part of how AI systems are designed and checked.

Catalog record

CREATORZhangyu Wang, Nemin Wu, Qian Cao, Jiangnan Xia, Zeping Liu, Yiqun Xie, Akshay Nambi, Tanuja Ganu, Ni Lao † \dagger, Ninghao Liu † \dagger, Gengchen Mai † \dagger
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE47 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-23482

Browse the full library · About the Tower of Babel