AI’s Role in Urban Planning
Image generated using Adobe Firefly.
Right now, every tech company is rolling out various AI tools to help brainstorm, write, and illustrate for all manners of industries. ChatGPT is being integrated into Apple products, Microsoft is forcing Co-Pilot on everyone, and Adobe is stealing our images and documents to train its AI. With all of this innovation, surely there must be a role for AI in urban planning, but there is not.
Urban planning and city building in general is fundamentally a community-based activity. Each community has a unique set of goals and desires that cannot be accounted for in a city plan written by machine. Even with the most well crafted prompts, machines will never be able to fully incorporate the nuances of a community as well as a planner who is physically present in the community, even if just visiting.
One of the most feasible uses of AI in urban planning would be to describe the historic and current conditions in a community. A good example of this would be the requirement for cities to describe population and income trends in their housing elements. It’s conceivable to design a large language model (LLM) like ChatGPT to take in Census data and output appropriate text for a housing element. Developing a model that could do this would cost millions of dollars for the potential of hundreds of thousands of dollars in revenue. The financial return just isn’t there. Hiring quality planners to analyze the data and write compliant housing elements is far less expensive.
LLMs work by providing the next most likely word in a sentence. They are simply fancy predictive text models barely better than your phone suggesting the next work when you text. The problem with using the “next most likely word” is that everything LLMs write is average. Everything they write starts to look identical. They can’t respond to the unique character of a place. Writing planning documents with an LLM is like lining a city’s arterials with 5-over-1s and hoping for a unique place to emerge.
The other problem with LLMs is that they lie. There is no validation that what they say is true or correct, they just spit out words. As a result, the product of an LLM requires considerable fact checking and editing. This process can take far longer and be more costly than simply having a planner prepare a good first draft of a document initially. The time it takes to edit is often longer than preparing the initial draft of a document, and that time only grows exponentially as the quality of the initial draft decreases.
Planning is fundamentally a community-based activity that requires nuance, listening, and eventually compromise and agreement within a community of people. There is no role for AI in this process, computers cannot respond to the human factors as well as a well-trained planner. We are all better off hiring quality planners who can respond to all of the human elements required for city building, rather than trying to insert AI into the process.