Examples of AI in the Public Sector: Public Safety

1. Using CCTV for understanding and optimising crowd behavior

The UK has one of the highest concentrations of CCTV cameras worldwide. While these publicly managed systems primarily aim to deter crime, the extensive data they gather can also be used to enhance public safety in other ways, such as managing and optimising crowd movements in busy areas. 

Improving public safety with computer vision and real-time contextual nudges

Singapore’s Smart Nation initiative leverages data from CCTV networks to identify how people use its services from bike paths to outdoor exercise equipment and to deliver real-time, context-sensitive nudges to residents commuting through the city. Digital displays in public spaces provide live crowd-density information, subtly guiding people away from overcrowded areas. 

Unlike CCTV-based solutions that track individual behaviors, which raise significant privacy concerns, this approach focuses solely on crowd behavior and flow, similar to systems used in airports. The initiative has also helped emergency services respond more quickly by using data to find the best routes through the city. The displays inform large crowds when emergency vehicles are approaching and request specific routes to clear a path.

2. AI in Public Sector UK: Driving Safer Roads Through Behavioral Nudges

With the rise of connected vehicles and smart mobility, modern transportation has become a vast network of real-time, geo-localised data. Vehicles, from passenger cars to e-scooters, act as "sensors on wheels," constantly generating valuable information on road conditions, traffic, and driving behavior. 

Coupled with the enormous amount of historical data on accidents, weather, and lighting conditions, this presents a goldmine for machine learning algorithms to identify risky hotspots and dangerous traffic zones. 

This enables authorities, municipalities, and urban mobility professionals to take proactive measures by changing the hazardous road conditions that contribute to accidents or by providing preventative warnings to drivers and pedestrians about potential dangers to help them take actions to avoid them. 

Machine learning to forecast accident hotspots

Zurich and other cities use Vianova’s machine learning solution to improve road safety by identifying high-risk zones. 

Their "Multimodal Road Safety" system—multimodal meaning the integration of vision, hearing, and speech— collects real-time data from connected vehicles, such as speed, location, and braking, to analyse traffic flow and road conditions. By combining this with historical crash data, Vianova’s algorithms detect hazardous areas and assign risk scores to roads and intersections based on factors like over-speeding and sudden braking. 

This allows authorities to implement targeted safety measures, such as redesigning intersections, installing speed cameras, or adjusting infrastructure to reduce accident risks.

AI trends in product development 2025 

• How can AI agents supercharge process automation across industries? 
• How can your business harness the potential of multimodal AI? 
• How can you safeguard your AI systems against emerging threats? 
• How will evolving AI regulations shape your strategy and path forward?

3. Predictive Policing with AI: Examples and Challenges

Predictive policing was a feature of the 2002 movie Minority Report, where a police division arrests criminals prior to them committing the crime. Today, many cities are experimenting with AI-driven crime prevention solutions that aim to proactively guide people away from potential misconduct. 

Forecasting criminal activity with machine learning to inform police presence

It’s well documented that police, or even police-like presence (fake CCTV cameras or cardboard cutouts of police officers) can reduce criminal activity. 

Predictive policing AI systems have also attracted attention for their potential to forecast and help prevent crimes. Such tools employ data analytics, machine learning, and pattern recognition to analyse historical crime data and other relevant factors, identifying crime hotspots and forecasting when and where crimes are most likely to occur. This proactive approach aims to enhance public safety by enabling law enforcement agencies to allocate resources more effectively and intervene before incidents happen. 

A common criticism of predictive policing is its susceptibility to data bias. Relying on historical crime data risks reinforcing existing biases, as areas with a history of high policing are more likely to be flagged, potentially leading to cycles of over-policing. This, combined with a low level of transparency in how the algorithms work, has sparked debates about their fairness and ethics. To address these concerns, regular audits of the data and algorithms are essential to spot and reduce any biases.

4. AI-assisted behavioural coaching to enhance safer driving habits

In the UK, five people die every day on the road and 58% of deaths had speed as a road safety factor. Alongside major campaigns to try to intervene, how can AI help provide some introspection into drivers’ habits? 

AI Driver Coaching: Real-Time Feedback with Smartphone Sensors

Cambridge Mobile Telematics’s DriveWell Fusion platform tracks driver behavior by collecting data from phone sensors, including GPS, accelerometers, and gyroscopes, and uses machine learning algorithms to analyse driving patterns like harsh acceleration, sudden braking, sharp turns, and speeding. 

The AI platform also looks at the phone's movement patterns to determine whether the phone holder is the driver or a passenger in the car, helping to prevent phone usage while driving. Drivers receive real-time feedback in the form of push notifications and weekly reports, coaching them to improve their habits. 

Another promising area for improving road safety is AI DashCam systems, which can monitor potential risks on the road, providing real-time alerts for hazardous conditions such as distracted driving or fatigue, and track drivers with inward-facing cameras, helping drivers remain focused and attentive. 

Going beyond reactive alerts, this AI technology could identify recurring issues like speeding or tailgating and offer personalised coaching. Facial or voice analysis would detect emotional states like stress or anger, allowing the system to intervene before risky behavior escalates and provide calming prompts or advise rest breaks.

Adopting AI for public safety

Here’s our take on AI adoption in public safety initiatives:

• Real-time feedback goes hand-in-hand with personalised coaching: While people often agree they could improve, they may struggle in the moment or lack a clear plan. Machine learning models can analyse behavior and provide immediate prompts to change behavior, and personalised coaching to improve existing behaviors can be both timely and even enjoyable. Whether it’s related to driver performance, large crowd control or law enforcement strategy, benchmarking someone’s performance and setting goals for improvement is generally effective.

• Quality data can create real-time nudges: Singapore has demonstrated that you can use high-quality data to sway the behavior of groups in real time. Furthermore, we often think that large datasets can only be used in a long, async analysis to provide insights, but creating conditional logic on top of machine learning models can be extremely powerful. For instance, implementing Singapore’s nudge system to improve emergency response times worldwide would have a significant global effect on lives saved.

If you’d like to know more about how AI-driven behavior interventions can bring long-term results in the public sector, check out our blog, and feel free to consult us about AI-enabled digital products if you need professional guidance.