In the spirited cultural debate over the possibilities and risks of artificial intelligence, the imagined pros and cons have tended toward the sensational. There’s been little mainstream attention paid to the technology’s potential impact on the everyday tasks that keep our cities humming—things like construction permit reviews, development application processes, and planning code compliance enforcement. But the needs in those areas are quite real, and experiments to apply newer AI breakthroughs to these kinds of operations are already well underway. Municipalities large and small, from Florida to New England, and Canada to Australia, have announced AI-related pilots and other exploratory efforts.

While the approaches vary, the challenges are practically universal. Determining whether proposed construction or development projects meet all land and building codes is a detail-intensive, often slow process: It can be confusing for applicants and require extensive back-end work for municipalities and other authorities. The hope is that AI can help make that process—or “the tedious parts of city planning,” as the publication Government Technology bluntly put it—speedier and more efficient, as well as more accurate and comprehensible. Ideally, it would even allow planning departments to streamline and reallocate resources.

But as city officials working with the new technology make clear, there’s a long way to go to get to that point. And given that some of AI’s most publicized moments to date involve embarrassing failures (such as Google’s AI search tool advising users on the benefits of eating rocks and adding glue to pizza), they are proceeding with caution.

There’s often a “hype cycle” between a new technology’s early promise and its eventual reality, cautions Andreas Boehm, the intelligent cities manager for Kelowna, British Columbia, a city of about 145,000. His team is specifically charged with seeking new opportunities to leverage tech innovations for the city and its residents. Despite a lot of chatter, we still haven’t seen many “concrete, tangible examples” of AI as a “transformative” force in planning systems, Boehm says. But we may start to see real results soon.

Canada is experiencing a housing shortage, Boehm notes, and moving faster on new construction could help. The permitting pipeline is clogged with inquiries from current property owners about zoning and code issues for more routine projects. For a few years, Kelowna has been using a chatbot to answer common questions, Boehm says. That has helped, but the more recent “generative” version of AI can handle a much broader range of inquiries, phrased in natural language, with precise and specific responses. So Kelowna began working with Microsoft to build a new and much more sophisticated version of the tool incorporating Microsoft’s Copilot AI functionality, which they now use to aid permit applicants.

Boehm says the Intelligent Cities team and its consultants worked with a range of residents (including those with no permitting knowledge) as well as experienced builders to develop the tool; it can give high-level responses or point to specific code provisions. It has notably streamlined, and sped up, the application process. “It frees up our staff time” because fewer questions need to be addressed by staff early in the process, Boehm says. “So now they can focus on processing applications that are coming in. And often these applications are much better quality because people are using these AI tools as they’re putting these applications together, and getting all the information they need.”

On the other side of Canada, the city of Burlington, Ontario, near Toronto, has been developing generative AI tools in collaboration with Australian property and tech firm Archistar. Chad MacDonald, Burlington’s chief information officer (and previously executive director of digital service), says Burlington, population 200,000, also faces a housing crunch. With little space available for single-family housing construction, the city’s focus is on improving the process of handling larger projects, including industrial and commercial proposals, with an eye toward creating a single platform that would work for all kinds of projects. The system the city is developing aims to integrate not only local zoning and bylaws, but also the Ontario Building Code, which affects all structures in the province.

Testing this system involves checking whether it correctly assesses previously submitted plans whose outcome is known. This process also trains the AI. “Every time we correct an inaccuracy in the algorithm, it actually makes it smarter,” MacDonald explains. “So the next time it gets more and more accurate.” And if the proposed solution to one permit problem could create two more problems in the application, the system is designed to point that out immediately, avoiding a lengthy resubmission process. An “extremely successful” round of testing was completed in May, MacDonald says, and he expects the city’s use of the technology to expand.

MacDonald envisions the technology advancing to the point of creating code-compliant designs. But won’t that put engineers and architects out of business? He counters that it’s vital to keep humans in the loop. “This is about speeding up these really mundane processes,” he says, “and then allowing these very highly educated and specialized experts to focus on the things they really need to focus on.”

In Honolulu, expanding the use of AI tools is part of a more sweeping tech-plan upgrade to address a significant permitting backlog—in 2021, the city’s mayor declared the process “broken” and committed to an overhaul. In 2022, a permit prescreen process involved “an intolerable six-month wait” to reach a reviewer, says Dawn Takeuchi Apuna, director of Honolulu’s Department of Planning and Permitting. The city added an AI bot that was able to review some of the prescreen checklist items in a newly streamlined process, and it helped cut that wait to two or three days. That success helped lead to a more expansive generative AI pilot with Chicago-based startup CivCheck, a relationship Takeuchi Apuna expects to continue.

“We have learned that there are enormous possibilities of AI in our business processes,” she says, “and that the most important piece is the people that are using it.” She emphasizes that this is just part of an overhaul that also includes better staff training and improved communication with applicants. “It’s a value that you must bring and continue to enforce as part of AI in order to get the best results.”

While these early results are promising, AI still presents plenty of challenges and wildcards. Some of the startup’s promising, powerful generative AI tools are untested. And as MacDonald points out, the technology isn’t cheap. There’s also a need to set standards around what data the process collects and how it can be used. (Kelowna, for example, is working with the nonprofit Montreal AI Ethics Institute on policy and guidance issues.) And, of course, there are broader public concerns about giving too much control to an automated tool, however seemingly intelligent and teachable that tool may be. “It’s not going to replace people,” Boehm says. “We’re never going to just issue you a building permit from an AI bot.”

In fact, he continues, that concern could be considered an opportunity, if cities use AI thoughtfully and transparently. Although government is often opaque and thus treated with skepticism by many, AI “is a great opportunity to demystify government,” Boehm says. “It [can increase the] understanding that this is really about people in the end and supporting them.” In other words, in the best-case scenario, AI might improve a knotty but vital bureaucratic process by giving it a more human touch.

Rob Walker is a journalist covering design, technology, and other subjects. He is the author of the Lincoln Institute book City Tech: 20 Apps, Ideas, and Innovators Changing the Urban Landscape. His newsletter is at robwalker.substack.com.

Lead image credit: PhonlamaiPhoto via iStock/Getty Images Plus.