UK researchers develop AI system to track urban inequality

By on 24/04/2019 | Updated on 24/09/2020
The story in a streetscape: new AI system can identify deprived neighbourhoods (Image courtesy: Matt Ross).

Artificial Intelligence (AI) could be used to identify areas of poverty in cities across the world, after researchers at Imperial College London developed a system that scans street images.

Dr Esra Suel and colleagues from the university’s School of Public Health used deep learning image analysis to train a computer programme, identifying social, economic, environmental and health inequalities within four UK cities.

The programme was first trained on a total of 525,860 Google Streetview images from London, corresponding to 156,581 postcodes, and provided with government statistics on local incomes, health, crime, housing, and environmental conditions. It was then trialled on three further cities – Birmingham, Manchester and Leeds – where, once researchers had benchmarked the data by manually rating 1% of the available images, it was able to correctly predict areas’ economic and social wellbeing.

Eye in the sky

In their academic paper, ‘Measuring social, environmental and health inequalities using deep learning and street imagery’, Suel and colleagues write that “training in one city can be transferred to predictions in other cities in the same country, especially when networks are fined-tuned with as little as 1% of target city images.”

The authors hypothesise that the algorithm detects visual signs such as pollution and signs of disrepair in urban locations. “Some features of cities and urban life, such as quality of housing and the living environment, have direct visual signals in the form of building materials and disrepair, sources of air and noise pollution and green space,” they write.

“Others, like poverty, may be visible because they influence or are related to features like housing and neighbourhood environment, the type of vehicles that people use, or even the type of shops.”

Real-world applications

The researchers found the application of deep learning to street imagery provided more accurate predictions of some measures of equality, such as income and living environment, than others – including crime and self-reported health.

After their successful UK trial, the team now plans on using the technology in developing countries, where there is less up-to-date statistical data available to keep track of policies aimed at reducing inequality, The New Scientist reported.

About Natalie Leal

Natalie is a freelance journalist whose work has been published by The Sun Online, The Guardian, Novara Media, Positive News, and Welfare Weekly, among others. She also writes reports and case studies on global business trends for behavioural insights agency, Canvas8. Prior to working as a journalist Natalie worked for the public sector in social services for several years. She switched careers in 2013 after winning a fully funded NCTJ in a national writing competition. She holds a Masters degree in social anthropology from Sussex University where she specialised in processes of social change and international conflict and reconciliation processes.

Leave a Reply

Your email address will not be published. Required fields are marked *