Global disparities and structural imbalances in AI integration for urban climate resilience research

Abstract

Artificial intelligence (AI) holds transformative potential to decode urban climate resilience (UCR) complexities, yet its unguided expansion risks amplifying systemic biases. Here, we leverage large language models to conduct a systematic analysis of around 12,500 UCR publications from 2005 to 2024. Our findings reveal a fifteenfold increase in AI-related publications since 2018, yet this growth remains geographically concentrated: the United States, China, and the United Kingdom account for ~40% of AI-integrated studies, highlighting a deepening digital divide in both research capacity and geographic focus that limits the transferability of AI models to low-income regions. Crucially, we identify a statistically significant structural imbalance: while AI is heavily deployed for physical hazard prediction (40%), it remains under-represented in governance resilience domains (<4%). This asymmetry suggests a greater emphasis on quantifiable, engineering-oriented problems than on social-institutional adaptation needs relevant to vulnerable populations. We argue that bridging this gap requires a pivot from purely data-driven optimisation of physical systems toward human-centric AI paradigms that integrate unstructured socio-economic knowledge, ensuring technologies address the holistic needs of complex urban environments.