Whose city is it? Mapping perceived urban livability with citizen-guided AI

Area of Science:

• Urban planning and remote sensing
• Artificial intelligence and machine learning
• Geospatial analysis and social equity

Background:

• Urban livability metrics often neglect the experiences of residents in deprived urban areas (DUAs), potentially reinforcing inequalities.
• Conventional indicators may not capture the nuanced preferences of diverse urban populations.
• Integrating participatory approaches is crucial for equitable urban development.

Purpose of the Study:

• To develop and evaluate lightweight deep learning models ('AI-voters') for mapping urban livability.
• To compare AI-voter preferences trained on DUA residents versus city planners.
• To assess the potential of AI for scalable, inclusive urban planning.

Main Methods:

• Developed lightweight deep learning models ('AI-voters') using open-source satellite imagery.
• Trained models on livability preferences from both DUA residents and city planners in Ghana's Greater Accra Metropolitan Area.
• Employed a two-step urban form sampling strategy to reduce data requirements by 90% for scalable participatory mapping.

Main Results:

• AI-voters trained on DUA residents' preferences better reflected local insights on urban livability compared to those trained on city planners.
• City planners exhibited disagreements among themselves and consistently assigned higher livability scores, overlooking DUA residents' preferences.
• AI-voters accurately mirrored human preferences based on physical urban features like greenery and building density.

Conclusions:

• Integrating community perspectives into AI models is vital for mapping urban livability equitably.
• AI-voters trained on DUA residents' preferences can serve as scalable proxies for local insights.
• This approach can expose hidden spatial inequities and promote more inclusive urban development.