The silicon gaze: A typology of biases and inequality in LLMs through the lens of place

Title	The silicon gaze: A typology of biases and inequality in LLMs through the lens of place
Authors	Francisco W. Kerche, Matthew Zook, Mark Graham
Magazine	Platforms & Society
Date	01/20/2026
DOI	10.1177/29768624251408919
Introduction	This paper proposes the concept of the silicon gaze, detailing how large language models (LLMs) inherently reproduce and exacerbate spatial inequalities. Analyzing a 20.3-million-query audit of ChatGPT, the research highlights systematic biases in the portrayal of geographical entities such as countries, states, cities, and neighbourhoods. The findings suggest that biases in LLMs are inherent, stemming from historical data imbalances and design decisions. Through a nuanced, power-aware approach, the study introduces a five-part typology of bias—availability, pattern, averaging, trope, and proxy—illustrating how LLMs favour certain places while marginalising others.
Quote	Francisco W. Kerche, Matthew Zook and Mark Graham. The silicon gaze: A typology of biases and inequality in LLMs through the lens of place. Platforms & Society. 2026. Vol. 3. DOI: 10.1177/29768624251408919
Element	Silicon (Si)
Topics	Machine Learning in Materials Design
Industry	Research & Laboratory