Innovating the Netherlands: how AI is shaping Dutch smart cities

By Sabine Oskamp-Dubbeldeman

A quiet revolution is unfolding across Dutch cities. They're becoming smart and transforming into living labs, where sophisticated technologies and AI are being deployed to address pressing urban challenges — from crowded streets to climate change. 

The Netherlands is a frontrunner in smart city innovation, with initiatives that prioritize citizens and urban livability, as well as a strong commitment to ethics and privacy.

For a deeper insight into the current state of AI in Dutch smart cities, we interviewed four leading experts who work at the intersection of AI and urban innovation:

• Egor Bondarev, Associate Professor of AI and Computer Vision at Eindhoven University of Technology

• Iva Gornishka, AI Specialist at the City of Amsterdam

• Max van Meerten, Manager of Digital Innovation & Smart City at The Hague

• Tom van Arman, Founder & Director of Tapp at Smart City Prototypes

Smart cities: beyond the buzzword

The term "smart city" has become a catch-all for various urban technologies, but what does it mean in practice, especially in the Dutch context?

Tom van Arman, who has been working in this field for over a decade, puts it into perspective: "Smart city is a huge, broad topic — you're talking about smart energy, future mobility, IoT, open data, AI, circularity, it goes on and on. What I find most interesting is using data to show how areas can be improved and become more livable."

This focus on livability and driving real results is a key theme in Dutch smart city initiatives and among the experts we talked to.

Max van Meerten, Manager of Digital Innovation & Smart City at The Hague, explains their practical approach: "Our goal is to explore what opportunities new technologies in the public space can bring to make our city more sustainable, safer, and enhance the quality of life for citizens. We have a broad view of technology: sensors, fiber, computing, data streams, drones. And AI plays a major role because it's the big emerging technology that ties it all together."

Importantly, the technology serves the city, not the other way around. Max emphasizes this: "Our approach is focused on the city, the city's citizens, and urban challenges. And now AI can help us to power the city and make it better for people."

The evolution of AI in smart cities

AI in smart cities has evolved significantly over recent years. Iva Gornishka, who has worked as an AI Specialist at the City of Amsterdam for almost five years, has witnessed the changes firsthand.

"When I started, computer vision was hip and happening. We did a lot of work to explore how computer vision could help with asset management and monitoring of the public space. Out of that came projects for blurring panorama images and detection of garbage containers," she explains.

But the technology landscape has shifted rapidly: "With high resolution images, you can see things in the public space well, but you cannot easily localize them. So the next big thing was point cloud data — a new data source that would allow us to more precisely localize objects, such as street lights or trash bins. This helps when you have high resolution images and can see things well in the public space, but you cannot easily localize objects. Then suddenly ChatGPT hit the world, and up until that moment, NLP was something we had to fight and push for. But all of a sudden, the tables turned, and everybody was coming to us saying 'I want to do something with language models.'"

Egor Bondarev, Associate Professor of AI and Computer Vision at Eindhoven University of Technology, shares a similar perspective on the evolution of AI use in smart cities: "For smart cities right now, computer vision and machine learning are the most used AI technologies. And generative AI is the new kid on the block."

The Netherlands has developed a reputation for sophisticated smart city initiatives. Tom van Arman notes how advanced Amsterdam is compared to other world cities: "I'm super spoiled here because it's such a mature city to work for. I just got back from Boston, their city innovation department is called New Urban Mechanics with a staff size of seven to explore, experiment, and evaluate new approaches to government and civic life. Here, the same innovation department is around 150 people, doing everything from mobility, energy, circularity, and of course, AI."

Real-world smart city applications

Crowd management

One of the most mature smart city applications in the Netherlands is crowd management. Max van Meerten describes the need for their Crowd Safety Manager system in The Hague: "We have 14 million visitors a year in Scheveningen, with the majority coming on warm summer days to enjoy the beach. So managing those crowds and keeping Scheveningen accessible is a huge challenge."

The system provides real-time information about crowds on the beach, the boulevard, congestion on roads, and parking garage capacity. But its predictive capabilities are particularly valuable: "The Crowd Safety Manager can also predict crowds. A couple of days before a weekend, the coordination team will come together and use it to see what kind of crowds we're expecting in the upcoming days. It's based on historical data as well as 30 other factors. It can predict not only how busy it will be, but also whether people will come in the morning or afternoon. With this information the city can prepare and take measures beforehand."

This AI-powered prediction is transforming how cities manage public spaces. As Max explains, "We use machine learning to analyze historical data alongside weather forecasts and dozens of other factors. The AI models learn patterns from past events and continuously improve their predictions with each new data point."

Tom van Arman describes a similar system at NDSM in Amsterdam: "The cameras there are using AI algorithms trained on models to anonymously detect people — not individuals, just numbers of people. What they want to do is measure how many people are visiting, for example, public art installations. This helps the artists know how many people visited, and helps organizers know when the peak moments are."

"We co-developed this AI with the municipality to use object detection and computer vision logic to count people. The system is trained to recognize humans regardless of posture, clothing, or whether they're carrying items. It's pattern recognition at scale."

Public safety and security

AI systems are also being deployed to enhance public safety.

In Eindhoven, Egor Bondarov’s team worked on one such safety system: "We created a system of multi-cameras in the city centre of Eindhoven. This means we can basically re-identify people from camera to camera. We were able to build a map of moving trajectories of every person in the city, and detect people who are scouting around or moving in strange trajectories."

Egor explains how this works: "The AI models we're using can detect anomalies in movement patterns. They learn what "normal" pedestrian behavior looks like in different parts of the city at different times, and then flag any unusual patterns. It's not about identifying specific individuals, but about recognizing potentially concerning patterns of movement."

Tom shares an unexpected discovery from one of his camera monitoring projects: "We called it the 'Swim Counter' at Marineterrein. And during the COVID lockdowns, the camera was counting all sorts of crazy activity in the winter. We were like, who the hell is swimming now? So I walked to the boardwalk 'detection zone', and sure enough, you see these people dropping their skivvies and diving into ice cold water! Turns out they were all doing the Wim Hof breathing method to boost their immune systems during the pandemic!"

These sorts of insights help Marineterrein improve its services and facilities, for example by adding more changing rooms.

Citizen services 

Amsterdam has been pioneering the use of Retrieval Augmented Generation (RAG) systems to improve citizen services. Iva describes their system for handling citizen reports: "Citizens make reports (or ‘signals’) about the public space — things like 'trash bin needs to be emptied,' or 'the light is broken in front of my house,' or 'the pavement needs to be repaired.'"

The city found that in about 20% of cases, no new action was needed because either a resolution was already in progress or it went against city policy (such as not cutting branches for biodiversity reasons). "By using AI, instead of citizens making a signal and waiting for weeks to get a reply," Iva explains, "we can proactively provide the information that we have and say, 'Hey, did you know this is our policy?,' or 'Hey, the garbage truck is on its way and it's going to be there in the afternoon, so you don't need to make a report.'"

Iva elaborates on the AI used: "We used large language models with retrieval augmented generation to create an agentic system that can analyze citizen reports, understand their intent, and then retrieve relevant information from diverse internal systems and knowledge bases. The AI assistant can handle natural language input, even with typos or colloquial language, and respond in a conversational manner that feels helpful rather than robotic."

This approach not only improves citizen satisfaction through quicker responses, but also reduces the workload on city staff who would otherwise need to manually process each report.

Environmental monitoring and sustainability

Cities are also applying smart technologies to address environmental challenges. In The Hague, Max van Meerten describes their Smart Energy System: "We created a local energy network as part of our Living Lab, using AI to predict both the demand and supply of energy from local sustainable sources like solar panels. This way, sustainable energy can be used locally instead of going back to the big energy network of the city, which also helps with network congestion."

"We use AI to analyze weather patterns, historical usage data, and the availability of sustainable energy to make hourly predictions about energy production and consumption. The models continuously learn and adjust based on actual outcomes, getting smarter over time."

Tom highlights the "Green Mile" project in Amsterdam, which uses sensors to measure air quality and sound levels: "It's a wonderful project because citizens host these sensors and have an invested interest. One of the citizens is hosting a sound sensor on his houseboat along the busy Stadshoudekade street. He has access to both real-time and historic data and can use this data to monitor the decibels on his doorstep. So he can check: is this an anomaly or a pattern? If it's a pattern, then he can go to Stadsdeel and say, 'It's too noisy at night — you have to fix this problem.'"

Tom continues, "The AI in this system is doing pattern recognition and anomaly detection. It learns the baseline noise patterns and can identify statistically significant deviations. It can even classify different types of sounds — distinguishing between traffic noise, music, conversation, or construction."

Harbor management and water quality

Several Dutch cities are applying AI to their waterways as well. Max van Meerten explains The Hague's harbor registration system: "We're working with Brain Creators, an AI company based in Amsterdam, for our harbor registration system. They've developed image recognition technology to detect ships in our harbor, which helps with efficient management of maritime traffic and harbor resources."

"This AI tool not only detects the presence of vessels," Max elaborates, "but also classifies them by type, estimates their size, and tracks their movement. This gives harbor authorities real-time insights that were previously impossible to obtain without constant human monitoring."

Meanwhile, in Amsterdam, Tom is involved with water quality monitoring: "This spring, Marineterrein became an official public swimming area. To get approval for this, we needed to measure water quality for the past few years. We have sensors to measure water this quality. The data collected is automatically pushed to dashboards and to public kiosks, making this information transparent and accessible to everyone." This real-time monitoring helps city officials quickly identify and address water quality issues, while also raising public awareness about the health of urban waterways.

"Living labs:" testing AI in real-world environments

One of the distinctive approaches to Dutch smart cities is the use of dedicated "living labs" — urban spaces where AI and other technologies can be tested in real-world environments before wider deployment across a city.

These living labs serve a crucial purpose, as Tom points out: "You're able to test these AI-trained devices, in real and rough environments — whether it's rain, insects, or vandalism. You get to field-test these things, and if they start working and performing as expected, then you can replicate those projects in other parts of the city."

In The Hague, Max runs a living lab in Scheveningen. "It stretches from the harbor to the boulevard, around three kilometers. And it's a test bed in the public space where we can test new AI applications on a small scale," he explains. "If you look carefully, you'll see smart lighting posts with sensors and cameras that are all fiber-connected."

Max describes how they approach AI testing in their living lab: "First, developers have to develop and train AI themselves. Then, when we think it's ready, we test it in our living lab with various scenarios to see how it performs. If it works well in the living lab, the next phase is to scale it up to the rest of the city."

The Marineterrein Amsterdam Living Lab, where Tom works, functions similarly. "It's a wonderful playground to explore these concepts because it's not only a free test zone, but you can ask complete strangers for feedback, or get a better understanding of the public value of these technologies. As you're doing experimentation, research, and design, you can actually get feedback from people."

Egor emphasises the importance of these testing environments: "The problem with AI in controlled lab settings is that the algorithms might perform perfectly with clean, structured data, but then fail completely in real-world scenarios with noise, unexpected inputs, and edge cases. Living labs provide the messy, unpredictable environments needed to develop robust AI that can actually function in the real world."

This real-world testing is especially important for AI systems. Because they need to be able to deal with the messiness and unpredictability of urban environments. As Egor notes: "We analyze what's going on and we are able to point officers of cities to problematic areas, for example where the garbage was not removed or was placed in an unauthorized location."

Multi-sensory cities: beyond visual data

While cameras often come to mind when thinking about smart cities, the approach increasingly relies on multiple sensing technologies, what, Egor Bondarev explains, is called a "multi-modal" approach.

"The placement of visual cameras is not always allowed in our cities because of the GDPR law and AI Act by EU," Egor explains. "City services are often afraid of placing additional cameras. Therefore, we need different modalities which do not share the private data of citizens."

These alternative sensors include thermal cameras, acoustic sensors, imaging radars and LiDAR. Each has distinct advantages, as Egor explains: "If you look at the street during the day, video cameras can indeed see things. But once it's night without proper lighting, they see nothing. The same in bad weather conditions, like rain, or fog. Radar can see through fog, see through rain, see through darkness. Thermal cameras can see dark scenes as well as normal scenes."

"With multi-modal sensor setups, you have different senses, like the five senses a human has. And if you combine them in one model, one AI algorithm which combines these four or five streams into one, you get powerful results in a very robust analysis of what's going on."

Tom explains that this gives “the power to mix and mash data sets, as long as we have access to the data.”

At Eindhoven University, Egor's team is pioneering research with these diverse sensors. "The AI models for multi-modal analysis try to learn which sensor data to look for at a specific moment. In darkness, we really take into account the radars and thermal. In case of gunfire or strange loud situations, they put more attention towards acoustic sensors. In a normal situation, they take into account mostly visuals."

Tom explains how audio AI is becoming particularly valuable: "A wonderful citizen science group called Sense Makers is proactively training AI to recognize classifications of urban sounds, like scooter alarms, firetrucks, and loud music that can become a neighborhood nuisance. We also train AI to recognize classifications of sounds — scooters, music, motorcycles, car doors. But what's really interesting is that you can also use sound as a qualifier for urban health. If you detect more species of birds, you're talking about a healthier ecosystem or biodiversity. Or if you hear kids laughing, and the frequency of laughing increases, then you can talk about measuring joy. So you can use AI to detect certain sounds to measure livability or happiness."

The Dutch approach: ethics, privacy and citizen engagement

What distinguishes the Dutch approach to smart cities is the emphasis on ethics, privacy, and close citizen involvement. These elements are sometimes not included at all elsewhere in the world.

Privacy by design

Dutch smart city initiatives take privacy seriously, often designing systems with privacy protection built in. Tom describes their approach: "All these things are privacy by design and citizen-centric, meaning we're not going to be using any video or images. We're detecting and counting objects accurately, but only the numbers will leave the device."

This reflects a fundamental difference in how AI is deployed around the world, as Tom points out: "Depending on which side of the hemisphere you're looking at, the East will use similar technologies like cameras and microphones, except they use them to control people or penalize them. In the US, it's for massive profits and corporate commercialization. But here in Europe, especially in Amsterdam, we're in a GDPR world, with general data protection rules that protect us."

Egor adds, "When we design AI systems for public spaces, we already think about privacy in the earliest design phase, not as an afterthought. For example, with our camera systems, we blur faces from the beginning, and the data processing is done on the edge, so raw footage never leaves the device."

Ethical frameworks

Amsterdam has been particularly proactive in developing ethical frameworks for AI. Iva explains: "We are a multidisciplinary team, and what we always do is take a user-centred approach. We try to truly understand what citizens need, or if end users are internal to the organization, what our colleagues need, what their pains are, and how we can help them."

The city follows the TADA principles for ethical data use: "TADA is based on six values co-created with citizens. In our projects, we create ethical leaflets where we sit together as a group and discuss the specific values, risks, and opportunities of each project."

Iva emphasizes that this is not a superficial exercise: "It's always 50% ethical considerations. For example, can we translate or simplify responses so that more citizens can use the system, making us more inclusive? Or how can we design an interface that doesn't let experts overtrust the AI?"

Citizen engagement

Engaging citizens in the development and deployment of smart city technologies is another characteristic of the Dutch approach. 

In Amsterdam, Tom regularly conducts "sensor safaris" with various stakeholders: "I do it with policy makers, architects, students, citizens, kids — showing them and getting their perspectives. The feedback I get depends on the generation. If you're an older generation, this is all 'surveillance state' and 'big brother.' But younger kids are like, 'I feel safer with these kinds of things.'"

Max also highlights the importance of citizen awareness in The Hague's living lab: "We put up signs to make people more aware that we are collecting data. We're showing what kind of data and the purposes of the data we are collecting."

This transparency extends to more creative approaches, as Tom describes: "We're going to be working on co-creating a new AR app to reveal all of these sensor types what they do. And we’ll hopefully create a crowdsourced version of the sensor registry so people are aware that there's sensing going on, and decide, what kind of public space do we want? There are no 'cookies', or consent built into real-world environments, so how do we opt out?"

Max stresses the importance of citizen involvement: "Our Smart Garden project is a great example of citizen participation. About 100 citizens have sensors in their garden that measure moisture level and temperature of the soil. This helps them make their gardens more green and sustainable, while conserving water. I have a sensor myself, and I didn't use much water last summer because the data showed the soil was wet enough."

Challenges and limitations

Despite progress, implementing AI in smart cities faces significant challenges.

Technical challenges

Egor identifies a fundamental challenge in handling the sheer volume and variety of data: "Even if we have the freedom and possibility to install all these multi-modal sensors around the city, the problem is still how to merge all these 500+ data streams from different sensors and from city authorities, municipalities, city services, garbage collection, logistics? No one knows yet."

This is especially true for city-wide implementations: "We can play a little bit with small areas like the central city square. We can incorporate 20 visual sensors and five acoustic sensors together. But when it comes to large-scale, no one knows how to do that yet."

"The technological solution for really multi-modal AI at large scale is not there yet," Egor continues. "Academia is working only on small, specific problems, like how to merge video data with thermal data or acoustic data with video. But not at the large scale for a whole city with 25 modalities and data streams."

Organizational challenges

Max points to organizational challenges within municipalities: "One of the big challenges is that people want to prevent risks and fear things will go wrong or aren't allowed. Sometimes we have to push and try to demonstrate that it is allowed, safe, and tangible to do it."

Cross-organizational ownership issues also arise: "An innovation sometimes cuts right through the organization, leading to discussions about ownership. For instance, with crowd management, multiple parties are involved: the police, the municipality, event organizers. There is no sole owner."

Iva echoes this challenge: "When it comes to AI implementation, a lot of the obstacles come from the organization rather than the technology itself. Because in the municipality, we often have quite a lot of private data, including very personal, sensitive citizen data. This leads to a mindset of being better safe than sorry, where five privacy and security officers need to look at your project to give you a green light."

Privacy and regulation

European GDPR privacy regulations present both opportunities and challenges. As Egor notes: "The last five to seven years, the Netherlands dropped back a lot because of GDPR and more recently the EU AI Act, because researchers basically cannot get their hands on training data. A lot of training data is now in the process of deletion because of GDPR."

This creates a competitive disadvantage on the world stage: "China and the US don't have GDPR. They have tons of data. They can put up 100 cameras and train whatever they want and build very good AI. Therefore, in this domain, I think it will be very difficult for us to compete with the US and China."

Tom adds a pragmatic perspective: "Yes, we have to comply with all the rules and regulations, but I think the idea behind the AI Act is aligned with the approach we already have.”

The future of AI in Dutch smart cities

Looking ahead, our experts identified several promising directions for AI in Dutch smart cities.

Predictive urban intelligence

Egor anticipates more sophisticated predictive systems: "Current video AI models are trying to predict what will happen in a few seconds or minutes, based on the situation they see now. It's not yet there, but several research directions are moving towards that. Security officers could get an alert for predicted violence on a specific street, so they can already send officers there to monitor the situation and prevent it before anyone gets hurt."

"The real power," Egor continues, "is combining real-time prediction with longer-term pattern analysis to understand the root causes of urban issues before they develop. This requires both immediate sensing and long-term data collection and analysis using AI."

Sustainable AI and edge computing

Tom is concerned about the energy demands of AI in the future: "The more we depend on AI solutions, the more processing power we need, which means more energy consumption. Which means competing either for homes or for data centers. All these companies come to Holland to get huge tax breaks and build massive data centers that employ hardly anyone and use all this energy we could be using for households or businesses."

Iva: "At the City of Amsterdam, we are very mindful about the environmental impact of implementing AI. My team is trying to make the environmental impact more practical or tangible. When benchmarking large language models, we want to explicitly measure the environmental impact of each, to give more tools to people when selecting a model."

Tom highlights the need for more sustainable AI approaches in the future: "I like edge devices, where you train a tiny microprocessor to just do one thing, inside the machine, and nothing ever leaves that machine except for the numbers that you need. That, for me, is a much more sustainable use of AI in the future — using microprocessing and not these huge data sets."

Evidence-based urban design

Tom also sees AI becoming essential for validating urban planning decisions: "I think we need to use AI as a validation engine for sustainable transformation. We need to use AI to help us make good decisions about how things are performing. It's not so much historical data and big data lakes, but more about what systems could make public spaces or buildings perform better — energy management, mobility flows, resource management."

He continues, "Using AI, we can model scenarios before investing too much time or money doing it the old-fashioned way. As a designer, if you can sometimes intervene or add some component — maybe a crosswalk or a potted plant — you can see how behaviors change. The AI becomes really important because you want to validate your assumptions with data, not go through emotional, subjective discussions about what you think is nice or works."

AI for social inclusion

Max hopes AI can address social inequities: "I think the social divide in The Hague — we have rich parts and more upcoming parts — is a challenge AI could help with. There's also a digital divide within our city, and maybe AI can help make that gap smaller. I really hope it can do that, because it's important for our city."

Iva envisions AI helping vulnerable citizens navigate complex systems: "Supporting people in finding social benefits or social services that fit their situation. Quite often the information is out there, but people who need social services don't have the language skills to digest tons of very specific financial and tax information and complex documents that explain what they're entitled to. AI can make these systems more accessible."

Advice for AI professionals

For AI professionals considering a move into smart cities work, our experts offered practical advice and encouragement.

Tom is enthusiastic about the opportunities: "Hell yeah! There are two aspects to it that are particularly promising. One is using facts- and evidence-based design decisions — what are the instruments that collect data to show conclusively what's happening, removing all the arguments. The other is data visualization — visualizing things helps so many different types of people understand complex information in an accessible way."

Max recommends focusing on real problems rather than technology for its own sake: "I think the most important thing is to focus on the challenge, not on the technology. Try to learn what kind of problem you want to solve and how you can really help a city."

He acknowledges that working with municipalities brings unique challenges: "Cooperating with a city can be challenging. Things can go slow, and sometimes they can move really fast. You have to invest in the relationship and the cooperation. Not only does a startup have to do that, but also we as a municipality have to invest in relationships with businesses."

Egor emphasises the need to understand the regulatory landscape: "The fact that we have strict privacy regulations in Europe means that AI professionals need to develop skills in privacy-preserving machine learning techniques. Those who can deliver high-performance AI while respecting privacy constraints will be in high demand."

For those looking to enter the field, Iva offers encouraging advice: "Data scientists are very needed in the public sector! The cool thing about working here is that you have a direct impact on your own life and the things you see around yourself. If you see an issue with society that you want to fix or make better, it's a really exciting place to be. You can have a direct impact. So just go do it and put your skills to good use!"

Conclusion: the distinctly Dutch approach

These conversations highlight a unique Dutch approach to smart cities. It blends technology with practical solutions, ethical thinking, and active citizen involvement.

Tom sums it up well: "How we use AI is different from how Americans use it, or how the Chinese use it. Here it's not for profit, it's not for control, it's more about measuring and improving things so that we can create more liveable cities."

In the Dutch context, smart cities are not about futuristic visions or surveillance capabilities, but about solving concrete urban challenges in ways that respect privacy, promote sustainability, and enhance the lived experience of residents.

As AI continues to evolve and urban challenges grow more complex, the Dutch balance of innovation, ethics, and practical problem-solving offers a compelling model for cities around the world.