A review published in *Nature Cities* outlines the urban applications of generative models and emphasises the importance of context-based validation and the continued role of human judgement.
Under the title "Generative AI in urban science and practice", Yunke Zhang and her co-authors have published in Nature Cities a review of the uses that urban science and practice now make of generative models. Their initial observation: these tools know how to produce images, model certain aspects of human behavior and reason on heterogeneous data, three capacities that are beginning to change the way we observe a city, how we model it and how we prepare decisions that concern it.
The applications identified range from flood forecasting and mitigation to the production of views of future neighbourhoods, including communication with the public and the organisation of citizen participation. The authors also point out the setback: the arrival of synthetic content in the geographical field raises the question of the reliability of spatial data and machine-generated urban plans.
Their conclusion shifts the debate away from the technical field. What will determine the usefulness of these systems is not their novelty, they write, but the possibility of evaluating them without complacency, validating them in the precise context in which they will be used, and deploying them in a way that leaves judgment and human responsibility at the center — which presupposes an algorithmic governance anticipated by the cities themselves.
Read the full article on the original source
Go to the source →