AI-Generated Visual Data for Assessing Social Well-Being in Sustainable Cities

Abstract

Integrating social well-being metrics into sustainable city planning remains a significant challenge, primarily due to the subjective nature of well-being and the high cost/sparsity of traditional survey methods. This paper introduces a novel methodological framework that leverages AI-generated visual data (e.g., from Stable Diffusion models) to quantitatively and qualitatively evaluate the social well-being implications of urban environments. We synthesize controlled, realistic urban scenes based on specific socio-spatial parameters. These synthetic images serve as a testbed for extracting both objective urban features (e.g., green space ratio) and subjective human perceptions (e.g., safety, beauty) via computer vision and crowdsourcing. These multi-modal data streams are integrated into a composite Social Well-Being Index (SWI). A comprehensive case study, simulating the assessment of varied neighborhood typologies, demonstrates the framework's efficacy. Results show statistically significant disparities in SWI scores across different urban forms, with pedestrian-friendly, green, and mixed-use profiles scoring highest. The study concludes that AI-generated visual data provides a scalable, cost-effective, and controllable medium for pre-emptive social impact assessment, offering valuable implications for participatory planning and equitable urban development.