Authors: Zhibin Zheng*, McGill University, Renee Sieber, McGill University
Topics: Urban and Regional Planning, Urban Geography, Political Geography
Keywords: Smart Cities, Automated, Human-Centered, Machine Learning, Topic Modelling
Session Type: Virtual Paper
Start / End Time: 8:00 AM / 9:15 AM
Room: Virtual 33
Presentation File: No File Uploaded
Even as researchers recognize smart cities to be socio-technological assemblages, the drive towards automating urban activities continues, whether because of increasing amounts of big data, a wide range of interoperable sensors, or easier-to-use machine learning (ML) methods or because automation appears more universal, objective, and inductive. It is often overlooked that ML methods are not out-of-the-box but “semi-automatic” solutions requiring considerable human intervention. This non-automatic feature is acknowledged in human-centered ML (HCML) research, which helps in identifying the role played (e.g., humans-as-annotators). HCML contains its own lacunae (e.g., emphasizing steps in implementation where humans can act as more intuitive machines). Instead of arguing whether researchers should automate more or less, we are more interested in defining where the interventions occur and considering how ML can cope with social problems and ethical concerns early in its design.
We conduct a case study on utilizing an ML method called topic modelling to analyze applications to a pan-Canadian smart city challenge grant competition. The applications consist of 137 documents from 199 communities and contain approximately 1.5 million words. We document ten steps in topic modelling where instances of human intervention are needed during its implementation and interpretation. We then argue for a transformation from this automated mentality to a more human-centered deployment of ML that prevents ML being uncritically and opaquely adopted in smart cities and considers placing society in the loop, acknowledging wicked problems, and challenging urban cybernetics.