AI-powered cities: How artificial intelligence can transform government services

Click to listen on your favorite streaming platforms:

Apple podcasts      Spotify

Today’s guests:

Costi Perricos, global gen AI business leader, Deloitte UK

Michael Flynn, global infrastructure, transport and regional government leader, Deloitte Ireland

Bill Eggers, executive director, Deloitte's Center for Government Insights

Jumbi Edulbehram, global business development leader for smart cities and spaces, NVIDIA

Nick Holmes, director of sustainable infrastructure and transportation, ServiceNow

Cities are made up of a web of systems and services that provide for residents’ needs. When these systems work in harmony, the city and its residents can thrive. When they don’t, it can be a different story.

Cities are also laboratories for new technologies. Electricity and the gasoline engine transformed how city residents live and work. Now AI is on the cusp of creating what could be an even greater transformation.

“A technology change like this comes once in a very, very long time,” said Jumbi Edulbehram, global business development leader for smart cities and spaces, NVIDIA. “You can't come up with a single city department, a city function, [or] a citizen service that can't be enhanced with AI.”

Edulbehram’s point is supported by the report “AI-powered cities of the future,” produced by Deloitte, ServiceNow, NVIDIA and ThoughtLab. The report surveyed 250 cities around the globe about their AI plans, investments, and implementations. It reveals how these cities are using machine learning, robotic process automation, and generative AI and how they are planning to incorporate agentic AI.

In this episode of Government’s Future Frontiers, we explore the promise of AI for city services, and the obstacles city managers may face in implementing the new technology. Effective use of AI will likely not only require multiple departments sharing data freely, but could also depend on robust partnerships with private enterprises and academia. But Bill Eggers, executive director of Deloitte's Center for Government Insights, contends that the effort will pay off.

“Anticipatory government … when it’s powered by AI and analytics can enable governments to not only predict undesirable activities, but also prevent them from occurring, from spotting fraud to combating the opioid epidemic. An ounce of prevention really is worth a pound of cure, especially in government.”

Tanya Ott: The concept of the “smart city” isn’t new. A network of sensors—all connected via the Internet of things—aims to improve efficiency, livability and responsiveness to the needs of city dwellers.

But artificial intelligence is changing the game. All of that data coming in from various sensors and connected devices can now be parsed, combined, and amplified to make something greater than the sum of its parts.

Costi Perricos: What AI is going to do is radically supercharge the journey to smart cities that started a decade or more ago with the digital revolution. AI is going to increase the speed, pace, and effectiveness of smart cities.

Ott: I’m Tanya Ott, and in this episode of Government’s Future Frontiers, we explore how AI is turning the promise of smart cities into reality.

I’m joined by several guides on this journey. You’ve already heard from one—that was Costi Perricos, global gen AI business leader at Deloitte UK. I’ll also be speaking with Michael Flynn, global infrastructure, transport, and regional government leader at Deloitte Ireland, and Bill Eggers, executive director of Deloitte's Center for Government Insights.

In addition, we’re getting insights from two firms who are deeply embedded in the world of AI-empowered cities: global AI platform company NVIDIA, where Jumbi Edulbehram oversees global business development for smart cities, and ServiceNow, which is embedding AI agent and other functionalities into its systems. Our guest Nick Holmes is director of sustainable infrastructure and transportation there.

We were inspired to tackle this topic by “AI-powered cities of the future,” a report produced by Deloitte, NVIDIA, ServiceNow, and ThoughtLab. By analyzing the AI plans, investments, and implementations in 250 cities in 78 countries, it helps build out a road map to those future cities.

That road map will vary city by city. But the goal is the same: Make the user experience for living in a city as seamless as possible. What might that look like?

Perricos: An AI-powered city for me is an urban environment where artificial intelligence is actually integrated into the fabric of the city itself.

The moment you step outside, you get into an automatic cab to [take you to] your workplace. Having your lunch already there and delivered because AI knows what you like to eat for lunch. Coming home and having an automated taxi wait for you to take you to the train station—timed perfectly because AI knows exactly how the train times are arranged—as an example.

Bill Eggers: One of my favorite examples is in Raleigh, North Carolina—which is just a couple hours from me here—where they have sensors that detect early signs of blockages and forecast flooding and overflows during storms. Machine learning can predict water breaks, directing the municipal water utility where [one needs] to perform preemptive maintenance.

Nick Holmes: Let’s use AI-driven predictive maintenance for city assets like roads, bridges, [and] utilities. Let's make it really easy for citizens to interact with the government with AI-powered virtual agents [that] handle inquiries, service requests, emergency reporting.

Think about emergencies. We’ve had a lot of chaos recently with these weather events and other unpredictable acts. But what if we could use AI to enable automation for tracking incidents, dispatching emergency services, and even optimizing response times?

Jumbi Edulbehram: Traffic congestion is one of the top—if not the top—items that comes up in any survey of the quality of people’s lives in a city.

[Think of] having to stop at a light when there’s absolutely no cross traffic, which in a lot of cities you have to do because the lights are just timed—they don’t actually have any intelligence on traffic patterns. [And in] a lot of cities today, especially at intersections, there’s a massive issue with accidents—literally thousands of deaths at intersections—just because there’s no intelligence in the way the lights are controlled, depending on how the intersections need to be used for many different situations—for bicycles, for cars, etc. AI can analyze a lot of this data and citizens immediately feel the impact on both efficiency [and] safety.

Michael Flynn: I think it’s having infrastructure working as well as it can do for the citizens. You go to the city, and it works for you, and you feel connected. That now means that the cities can do all the things that each person who designed each piece, each asset, originally envisaged—now they're allowed to interact. And I think that's going to be pretty cool and make them really great places to live.

Ott: That was Costi Perricos of Deloitte UK, Bill Eggers of the Deloitte Center for Government Insights, Nick Holmes of ServiceNow, Jumbi Edulbehram of NVIDIA, and Michael Flynn of Deloitte Ireland, giving us their picture of how residents may experience an AI-empowered city.

All of these examples have one thing in common: to the citizen, the driver, the person turning on the tap, things just work, and they work smoothly, without effort on their part. According to Nick Holmes, that’s by design.

Holmes: Think about your smartphone—what do you use as an end user? You use the front of the phone, right? Think of that as a single pane of glass. As a citizen, I don't care about what's behind the scenes, I want that single pane of glass.

What I want is simplicity. I don't want to around a website with tons and tons and tons of different links. I want to go in through one front door. I want to be able to do everything that I think I should be doing, and then, I don’t care what’s going on in the background, what departments I’m hitting, what IT systems I’m hitting, what different areas I’m touching on.

But what if it gets even more interesting? What if I was going in there to renew my driver's license, and then all of a sudden, it popped up that I had a few parking tickets I needed to pay. What if the AI was then enabling even better citizen services? That's all about connecting the dots behind the scenes. That's what AI does a really good job on. So that you as the citizen, the visitor, or the business, you don't see that complexity—just like I don't see the complexity on my phone, but I get to be able to have that wonderful experience with just one touch and one go.

Ott: But while the user might not care what’s going on behind the scenes, city managers do. They have to evaluate how to provide services when and where they’re needed. NVIDIA’s Jumbi Edulbehram says those services fall into two major buckets

Edulbehram: The first bucket is services based on the different kinds of data that the city has. Behind the scenes, cities need to make data across all the different departments available in what we call a large vector database, where it can be analyzed, it can be correlated, you can use it for predictions. So cities need to build out that data infrastructure on the back end.

Ott: Bucket number two involves systems that often require decision-making on the fly.

Edulbehram: A lot of the old utility infrastructure, for example, where people are accessing energy, water, waste management.

Ott: These services need constant tweaks to handle changing demands: think electricity spikes on hot days, water usage changes during droughts, or extra waste after big events.

Edulbehram: Cities need to make that infrastructure much more efficient. And obviously residents can feel the impact of that, but cities get the biggest impact from cost reduction while maintaining quality of service.

Ott: Those different buckets require different solutions. First, let’s take a look at the data. This first goal, says Costi Perricos, is to collect as much as possible.

Perricos: These new algorithms are even more data-hungry and much more sensitive to data than previous generations of AI algorithms. So the ability to place sensors around the city and to be able to collect data—high-quality data—from these sensors so that these algorithms can learn and deliver AI-enabled services becomes very important.

Ott: Once a city has that data … that, says Jumbi Edulbehram, is when AI can really get to work.

Edulbehram: AI is really good at ingesting multimodal data, [both] structured data, which is forms, text files, etc., and unstructured data, which are things like audio files or video from cameras. It’s really good at taking all that data and putting it in formats that can be uniformly accessed.

To take massive amounts of data, store it and make it accessible, just to be clear, is a nontrivial task in a city, but this is something that AI Is good at doing and once the data are in a format that can be analyzed, using it for predicting, for example, traffic, energy usage, health issues across the city, is just absolutely amazing value-add.

And we're at a point now where enough cities have done it and have examples of what value that can bring. Departments that were traditionally hesitant to share data are willingly doing so now.

Ott: What sets an AI-enabled city apart is its ability to take all of the data being collected at various points across the city, crunch that data and then make evidence-based predictions.

The sorts of predictions that have real value to citizens, says Nick Holmes.

Holmes: I think AI does a great job of helping us make better decisions. We've got so much data, the data that can then be processed very quickly.

We can think of optimization, we can think about recommendations, and then we can surface that to you as the decision-maker and say, “Hey, look, this is the decision that we would recommend. This is the consequence of those decisions.

Ott: Bill Eggers says city managers can use those predictive powers to produce powerful results.

Eggers: A really cool thing [cities are] doing is using digital twins and AI to guide development. So they do microclimate modeling to enable them to avoid development that could create heat islands, which is becoming such a big issue in cities today. This approach addresses the challenge of extreme heat and its impact on residents and future infrastructure development.

We’re seeing a lot of examples like this across public safety, across environmental issues, across living, health, and public transportation and mobility, and even just the physical space that cities have.

Ott: I asked Michael Flynn to give us a deeper view into what a digital twin is, and what it can do.

Flynn: A digital twin is a digital representation of a city or a district or even just a single building. If you put in all of the potential use cases into the one thing, you can use that to predict, to plan, [and] to run scenario tests on “what happens if ...”

Now you can, if you use [a] digital twin appropriately, keep that digital representation of an asset all the way through post construction. So during the time you can look at, “if I'm building it this way, how do I operate it? How do I make it more efficient? How do I make it that the operation is efficient, is cost-effective?” But [you can also see] how it interacts with the rest of its environment, with the other buildings that are sitting in the area. What does it do to the transport plan for the city?

You’re able to scenario test all of that. Now when you step forward and the asset is built, not only can you run scenarios on resilience testing, emergency, [you can model] if there [are] any changes, particularly changes in regulation over time.

You can start looking at the operation piece and centrally managing that, and the interaction of that operations with other parts. [Say] that asset is a road: Well now, you've got that all set up from the day you started thinking about it and strategizing about it, you've worked all the way through to now operating it and making it the most efficient use of that asset, using AI to make this an efficient asset for citizens and for the city.

Ott: That’s a taste of what data can do for an AI-empowered city. Now let’s look at Jumbi’s second bucket.

Edulbehram: Systems where decisions need to be made in real time.

Ott: Different types of AI have different strengths. Right now, the world is hearing a lot about generative AI—algorithms that can generate text, pictures, video, or other content. Jumbi says this type of AI isn’t quite flexible enough to handle city systems.

Edulbehram: Generative AI is trained on a particular model, and it gives you answers to particular questions. But if you go way beyond that, to try and get it to do stuff in real time, it may require it to be retrained, just like a person that's doing a particular task and you put them on a completely different task that they had no training to do.

Ott: Costi Perricos agrees.

Perricos: If you've used some of the existing tools that are out there, currently, it’s all around prompting and eliciting an answer, and then you're using your own intelligence to ask the next question until you've got everything that you need.

Ott: But a new iteration of AI is advancing the possibilities—agentic AI. Unlike generative AI, which needs specific training for each task, agentic AI can adapt and act in real-time without needing constant retraining.

Perricos: Agentic is, for me, the next level of intelligence or automation in AI.

Ott: That ability makes all the difference when AI algorithms meet all of the dynamic systems that make up a city, Jumbi says.

Edulbehram: I’d say [the] three large areas where we’ve seen agentic AI systems being super useful are, one, in traffic management, because traffic is fairly dynamic—things happen, accidents happen, road conditions change, etc. Being able to analyze all that information in real time—from sensors, from cameras, from data—and being able to operate the lights to change traffic patterns to make it more efficient and safe [is] one of the key areas for the use of agentic AI.

[Another] one is utilities, because again, the demand for utilities—water, electricity, waste management—is fairly dynamic, and optimizing the usage of utilities and optimizing the production and the supply in real time is a huge benefit. Again, agentic systems are really good at that.

The last one is just broadly public safety, which is enhancing the response to emergencies by analyzing situations in real time and making sure that the right resource is deployed for the right situation.

Think about someone calling an emergency number, 911 in the United States, for example, and speaking a language like Croatian. Someone on the other side may not even recognize what language that is. Being able to get a translator, being able to offer services like these to citizens, which had never been possible before, is super exciting.

Ott: The possibilities that agentic AI brings to the table are exciting. But implementing these solutions aren’t easy. A number of obstacles must be overcome.

Obstacle one: power.

Perricos: You will definitely need more compute infrastructure. These algorithms are very power-hungry, and that comes with additional energy needs. And so, particularly when you're looking at urban planning and cities, understanding the energy requirements that this sort of automation and artificial intelligence brings becomes very, very important.

In a number of cities around the world, the electricity grid is already struggling with people having converted their homes to heat pumps and car chargers, which are all very energy-hungry. You overlay AI onto that, and you do need to think about power delivery as a key factor, but you do need infrastructure.

Edulbehram: You have to make sure that you can meet the peak demand, just like an energy utility is designed to meet the peak energy demand, for example, in the middle of the summer where everyone's using their, you know, ACs and cooling systems.

So predicting that peak demand of what citizens might want to use is essential for making sure that you can meet the needs.

And of course, you need to make sure the back-end infrastructure is super scalable and scalable on the fly. A lot of cloud-based systems have the ability to not just scale quickly, but continue scaling to meet the demand. So a centralized infrastructure that's easily scalable is probably the best way right now to supply all the AI needs across the city.

Ott: Obstacle two: trust. Nick Holmes of ServiceNow explains that the challenge lies with us as everyday citizens.

Holmes: The bigger barrier that we've got right now is us. It’s us wanting to share that information. It’s us feeling comfortable that the information that I share with you is not going to get surfaced on the internet, on some wiki page somewhere else.

It becomes a cultural argument, and it becomes a personality argument on this.

Ott: The solution: technological guardrails—robust security systems and built-in systems that can spot problems before they occur. As we discussed earlier, that’s something AI is quite good at.

But another solution is moving that cultural needle, and Costi Perricos says cities already have a compelling argument.

Perricos: As consumers, we share our data with different technologies and apps every day. This is perhaps just another facet where we might do the same.

Ott: Obstacle three: silos. For AI to truly deliver on its potential, no city department can be an island, says Costi.

Perricos: This is not about a single government entity running AI in a city. All these things require ecosystems of technology players, of data providers and of government agencies.

For example, you’re commuting to work, you’ll be using roads, you might be using trains, you might be going through areas that need policing. You'll probably be touching a number of government services and you'll be touching a number of private services, and they need to work seamlessly, and all know about you as a consumer so that they can give you the best experience possible.

There [are] all sorts of agencies and all sorts of technologies that need to come together to make that a seamless user experience. So being able to curate the right ecosystem and have that ecosystem not only share data, but share technology seamlessly becomes really important.

Ott: But it’s not just city departments working with each other, says Jumbi Edulbehram.

Edulbehram: A lot of AI development is obviously happening in the private sector. Public-private partnerships are absolutely essential, [and] not just for the transfer of knowledge.

Ott: The public-private partnerships can provide funding, establish early successes, and expand the reach of existing services.

Edulbehram: The other kind of collaboration that seems to be super important is collaboration between cities and academic or research institutions. Cities don't generally have these folks in their IT departments.

Ott: So cities can work with these academic institutions to train up a new group of city workers who have the skills to develop, run, and collaborate with AI systems.

Edulbehram: It's important for cities to collaborate with other cities and learn from what they've done. And luckily cities are very willing to share successful projects. So it’s important for cities to come together in consortia to make sure that all these applications that they’re doing across the globe are quickly shared so each city doesn't have to start from scratch.

There are also global networks, like, for example, the Global Smart Cities Alliance. And then, there are a lot of conferences like the Smart City Expos, where cities will send their chief innovation officers or their CIOs to start setting up these collaborations. Having attended a lot of these conferences and being part of some of these alliances, I can tell you, “Yeah, there is absolutely a huge need, but [also a] very large degree of openness from cities that have applied or are starting to deploy AI to actually share their best practices.”

Ott: And the last, and possibly greatest obstacle: Where does a city even start? Nick Holmes has some advice on that front.

Holmes: The turn of phrase I really like is “Think big, start small, scale fast.” You can change that scale fast to fail fast as well.

The analogy I use on this one is when [you] were growing up, [you] always wanted to ride a bike, right? When you jumped on your bike for the first time, did you cycle off into the distance? Did you never fall off? Did you never stumble? Did you never get banged up knees and elbows? Probably not.

But if you don't get on your bike and try, then you're never, ever going to be able to learn how to ride it. You miss every shot that you don't take. And so that's what I would encourage with government organizations: [You have] to jump in there and don't try and boil the ocean.

Think big, think about where you want to be headed. Start small on something that's discreet, something that's measurable. Make sure you get a really good sponsor. Make sure [you] have really good partners. We talked about the guardrails in place. And then just try it. Then all those benefits I talked about at the beginning, all of those citizen experiences, business experience, visitor experiences, that's what you can do with your city.

Ott: More and more cities across the globe are taking those steps, as outlined in AI-powered cities of the future. This is leading to a future where AI is an integral part of the city, says Jumbi Edulbehram.

Edulbehram:  The production of AI knowledge needs to have the same profile as a utility. It needs to be easy to consume, just like plugging in an appliance into a socket or turning on a tap to get water.

It also has to be scalable, because just like water or electricity, as the need increases, as the consumption increases, you need to be able to scale the supply. It has to be reliable, of course, because if your power cuts off or your water doesn't come to your house, there's a huge disruption.

Ott: Bill Eggers adds one more requirement: It needs to be widely available.

Eggers: Certainly making sure that the access is spread, it’s the same thing that cities have to go through when they’re determining any sort of infrastructure development and planning.

Ott: With AI, governments can become more dynamic. Bill suggests that they can become …

Eggers: Anticipatory government … when it's powered by AI and analytics can enable governments to not only predict undesirable activities, but also prevent them from occurring, from spotting fraud to combating the opioid epidemic. An ounce of prevention really is worth a pound of cure, especially in government.

You can apply this predictive analytics now across a wide range of areas from human services to health care to climate-oriented events and that can help government to get ahead of things and be more anticipatory. That also can apply to human needs, using analytics and AI to better predict what some of those needs of the future might be.

A great one in this case is looking at the jobs of tomorrow and the skill sets that are going to be needed. A number of jurisdictions around the world now use AI both to help predict that, but also as a way of helping to connect individuals to those job-training opportunities.

Ott: AI opens these possibilities, and more. I’ll let Jumbi Edulbehram sum it up:

Edulbehram: A technology change like this comes once in a very, very long time. I’d say the last big sort of inflection point in technology was the advent of the Internet. Now, with AI, we’re at kind of another inflection point. And this one is going to be much bigger and much, much more impactful. Because it really impacts every single aspect of the way the city operates. You can’t come up with a single city department, a city function, a citizen service, which can't be enhanced with AI.

Ott: Thank you for listening to Government’s Future Frontiers with me, Tanya Ott, brought to you by Deloitte Insights. I want to thank all of our guests today: Costi Perricos, global gen AI business leader for Deloitte; Michael Flynn, global infrastructure, transport, and regional government leader at Deloitte; Bill Eggers, executive director of Deloitte's Center for Government Insights; Jumbi Edulbehram, who oversees global business development for smart cities and spaces at NVIDIA; and Nick Holmes, director of sustainable infrastructure and transportation for ServiceNow.

You can learn more about how cities are using AI to change the way they interact with residents, maintain infrastructure, and improve livability in “AI-powered cities of the future.”

We’ve considered the way AI is changing other aspects of life, like the medical industry:

Sara Seigel: Many clinicians say 30% to 50% of their day is spent on clinical admin [tasks]. If we could free up all of that time, that would be [like] a doubling of the workforce.

And that is, I think, one of the greatest promises of generative AI. Doctors and nurses spend time looking through notes. They spend time looking for diagnostic results. They spend time writing up notes, putting things in different systems. These are things which can all be streamlined, and can all be digitized, especially with tools like generative AI, which can search a patient’s longitudinal health record for the whole of their life to see if they’ve ever taken a drug and had an adverse reaction to that drug.

Ott: You can check out our previous episodes wherever you get your podcasts—like however you’re listening now.

 If you’ve already subscribed, you’ll get new episodes delivered automatically. If you’re not subscribed, maybe hit that button right now so you don’t miss out.

This podcast is produced by Deloitte. The views and opinions expressed by podcast speakers and guests are solely their own and do not reflect the opinions of Deloitte. This podcast provides general information only and is not intended to constitute advice or services of any kind. For additional information about Deloitte, go to Deloitte.com/about.