# AI is helping design cities. How should urban planners proceed? > Source: > Published: 2026-08-20 17:46:55+00:00 # AI is helping design cities. How should urban planners proceed? Using AI in urban planning can predict floods and simplify policy, but Northeastern researchers say it needs careful oversight. AI is showing up in nearly every aspect of daily life — from internet searches to visits to the doctor’s office. It could one day even play a role in the layout of the street in front of your apartment building. Urban planners are increasingly exploring the emerging technology for flood prevention and mitigation, generating renderings of future neighborhoods and communications with the public. That’s according to [a recent Nature](https://www.nature.com/articles/s44284-026-00492-2) review article published by Northeastern professors and Network Science Institute members Esteban Moro and Ryan Wang. In their report, Moro, a professor of physics, and Wang, a professor of civil and environmental engineering, analyzed more than 100 studies in the fields of urban science, computational social science and geospatial AI, which integrates artificial intelligence technology with mapping and satellite imagery. For the study, the researchers examined two types of increasingly popular forms of AI showing up in the urban science space — distribution-fitting generative models and foundation models. Distribution generative models are a type of AI that can create new images and renderings based on a large corpus of data. “One example in urban settings, you could send to a model a distribution of satellite images around a city and they will be able to produce another part of the city,” Moro said. The team found that urban planners are using this type of AI for a range of applications, including to predict flood scenarios. ## Editor’s Picks “You train the model with all the different pictures you have about flooding in different parts of the world, and then you could ask it ‘How will a flood, for example, in the city center of Boston look like?’ and they will be able to actually reproduce that.” Foundation models, on the other hand, are more instruction-based and learn primarily through human language – think large language model-based chatbots like ChatGPT and Claude or Google’s Notebook LM tool, which can translate complicated documents into plain English. Those tools are increasingly being used to make complicated and dense policy documents more understandable for the general public, he said. In one study the researchers analyzed, AI agents were even used to represent the perspective of residents in two neighborhoods in Beijing during urban planning meetings. As members of the meeting, the AI agents were able to speak on the issues that were important to the community, and even influenced future land development plans, according to the report. “The study found improvements in measures, including participant satisfaction and inclusion,” said Moro. “This is a particularly interesting example of AI facilitating community participation rather than simply automating a planning decision.” But while generative AI holds a lot of promise, it also comes with a hefty set of challenges and limitations as well, the researchers noted. Like all forms of generative AI, these models can “inherit and amplify biases” found in their training data, the researchers wrote. Additionally, they often can produce false or misleading information. “In high-stakes settings such as disaster response or large-scale infrastructure planning, these limitations become governance risks,” they wrote, referencing their tendency to make inaccurate statements and the challenge of interpreting how they came up with their responses. That’s why as communities continue to adopt the technologies it’s important to develop strong frameworks surrounding their use and development, Morro and Wang argue in the article. They also offered an ethical framework urban planners should consider built around five key themes — fairness, privacy, explainability, controllability and community participation. “Communities have adopted a lot of these AI technologies, but we haven’t seen a deep reflection on what is needed for these technologies to be accepted both at the level of our research [and] at the level of practitioners,” said Moro, whose research focuses on the intersection of big data and human dynamics. One of their biggest pieces of advice is for communities to develop robust validation and reliability system checks, explained Wang. This responsibility should be shared by both the AI providers and the urban planners implementing the technology, he said. “While we use it it’s important to evaluate it honestly,” he said. “We can’t just say there’s a new toy. Let’s use it.” Moro said ensuring that human review is part of these processes will likely be costly. It’s a lot cheaper to let a Large Language Model (LLM) work in the background and accept their output at face value than to hire a human worker to validate those responses, he said. However, the researchers said they were heartened to see communities are taking that responsibility seriously. “The human experience is really valuable,”he said. “We see a lot of initiatives to incorporate humans in the loop.”