What happened
- MIT researchers published a book, 'How AI Sees the City: Urban Visual Intelligence,' through Routledge, on using visual AI to study cities, MIT AI News reported. Authors are Fábio Duarte, Martina Mazzarello, Carlo Ratti, and Fan Zhang of Peking University's Institute of Remote Sensing and GIS. (MIT AI News)||- The book draws on work from the MIT Senseable City Lab, founded in 2004, which uses data to understand urban dynamics. (MIT AI News)||- Prior Senseable City Lab research cited by MIT AI News used machine learning to identify vehicle types across 331 New York City traffic cameras and estimate each vehicle's emissions. (MIT AI News) [1]
What visual AI is used for
- Potential applications named by MIT AI News include pinpointing causes of traffic congestion, comparing safety at intersections, and identifying which parts of plazas or parks draw the most people. (MIT AI News)||- The book also examines urban greenery; satellite imagery shows tree cover at a broad scale, while the article describes street-level imagery as near-ubiquitous. (MIT AI News)||- Duarte said with computer vision 'each image is a dataset,' and that techniques like Kevin Lynch's paper-and-pen urban observation can now be scaled up across many dimensions of cities. (MIT AI News) [1]
The concerns
- More images mean more data and more insight, but also more concerns about privacy and fairness, MIT AI News reported. (MIT AI News)||- Duarte said of treating images as data: 'We have to be careful about it.' (MIT AI News) [1]
Context
- The authors place AI within a longer history of visual urban study, from Roman marble maps to photography and the work of Kevin Lynch and William H. Whyte. (MIT AI News)||- Ratti said visual AI lets researchers 'build on that tradition' at a scale and level of detail previously impossible, and the authors argue AI can still be seen as a tool serving human purposes in urban design. (MIT AI News) [1]
Sources
- The promise and peril of using visual AI to study cities
MIT AI News · Primary source ·
