A city does not become better simply by adding AI. A useful project should solve a recognizable problem: making a service easier to access, identifying maintenance needs or helping staff handle information more effectively.
For residents, the important outcomes are ordinary ones. Can you get where you need to go? Can you understand a notice? Can you correct an error without spending a day chasing it? Those questions are a better starting point than a vendor’s promise of a “smart city.”
Start with the public problem
Ask the city to describe the problem before describing the technology. A late bus, a confusing application form and a leaking pipe need different responses. AI might help with part of the work, but so might clearer instructions, better maintenance or additional staff.
Some projects described as smart use sensors, online forms or ordinary rules-based software. Those tools can be useful without being AI. Conversely, a sophisticated AI model does not make a poorly designed service accessible.
A sensible proposal explains what the system will predict, generate or recommend, who uses the output and what action follows. That makes both the benefit and the risk easier to evaluate.
Look for concrete examples, with dates
In its October 2023 AI Action Plan announcement, New York City described work including a business-information chatbot pilot and a machine-learning tool to flag unusual utility bills. The announcement also outlined governance, staff training and procurement work.
These examples show the range of possible uses. They do not establish that a chatbot always gives correct regulatory advice or that every project delivers savings. An announcement states intentions and describes activity; a later evaluation must show results.
If an article presents a years-old pilot as a current success, look for an update from the responsible agency. Check whether the project continued, changed or ended, and what evidence is actually available.
Measure what residents experience
Consider a fictional city testing AI to improve bus-arrival estimates. A useful evaluation could examine how closely predictions match actual arrivals, whether performance differs between routes, and what happens during disruptions.
It should also ask whether the information is usable. Is it available at stops as well as in an app? Can someone using assistive technology access it? Is there a clear indication when live data are unavailable?
Now imagine the same city celebrates thousands of chatbot conversations. That count does not reveal whether residents found the right form or completed their applications. A service should be judged against its purpose, not simply how often people interacted with the software.
Data collection needs a clear purpose
Ask what information the project collects and why. Counting vehicles is different from identifying individuals. A proposal should explain whether personal information is involved, who can access it, how long it is retained and whether a contractor can reuse it.
More data are not automatically better. A city should be able to explain why the collection is proportionate to the problem it wants to solve. Claims about public safety deserve scrutiny of both demonstrated benefits and possible harms.
The UNESCO Recommendation on AI ethics emphasizes privacy, transparency, accountability and human oversight. For a local project, those principles should translate into plain-language information that residents can actually find.
Keep a route for people and corrections
If a generated answer concerns a permit, fee or eligibility rule, verify it against the official source before acting. A helpful explanation is not the same as a binding decision from the responsible agency.
A city should explain where people can report an incorrect answer and who will deal with the underlying problem. A complaint that disappears into another automated system is not a meaningful remedy.
Residents should also ask about alternatives for people without compatible devices, reliable connectivity or confidence using a particular service. These needs are not confined to any age group. Maintaining access should be part of the project from the beginning.
Six questions for a meeting or consultation
- What problem is being solved, and what non-AI alternatives were compared?
- What evidence shows an improvement over the existing service?
- Who might experience more errors or lose access?
- What data are collected, retained and shared?
- Who can correct a decision, and how can residents reach them?
- What happens if the supplier changes its terms or the system fails?
NIST’s AI Risk Management Framework offers a voluntary structure for organizations managing AI risks. Invoking a framework is not certification; ask what evaluation and monitoring have actually been done.
You do not need to be an engineer to contribute to this discussion. Your knowledge of how a service works in daily life is relevant evidence. For the broader picture, read how AI changes society, including who benefits and who carries the costs.
Featured photograph: Marek Rucinski via Unsplash. Public transport photograph for context, not evidence that the tram uses AI.
Updated September 24, 2026. This AI-assisted article explains the linked sources and includes clearly labeled illustrative examples where used. It is for general information. See our disclaimer.
