From Data Chaos to Urban Oasis: AI's City Makeover

AI can significantly de-risk the development of smart city applications as it transforms raw data into valuable intelligence.

This is one of the premises of a new whitepaper, A blueprint for using AI to create smarter cities, published by SmartCitiesWorld in association with Dell Technologies and Nvidia.

The whitepaper focuses on two use cases that use fields or subsets of AI: intelligent transport systems and services for residents.

AI can contribute to intelligent transport by increasing safety with intersection technology and advanced detection, boosting revenue through vehicle plate detection, and enhancing planning using a digital twin.

In services, AI can increase the efficiency of processing transactions, impact the citizen experience by reducing waiting times through chatbots, and drive more accuracy in interacting with the city’s database of citizen information.

“Having identified the outcome of the smart city use case, the next step is to fully understand the required workflows,” says the whitepaper, which is written by SmartCities World’s Sue Weekes.

“This will vary depending on whether the city needs a customized outcome or can leverage existing outcomes delivered by readily available applications.”

Data discovery

The whitepaper goes on to outline an optimal workflow, from data discovery through to model implementation.

It begins with data discovery and access and then moves to exploration and enrichment before moving to model development and implementation.

Dell Technologies and NVIDIA advise that cities take a broad view of the range of data sources available so they don’t limit the potential of the insights.

“Cities need to collect all the data possible because they won’t understand its full value until they start building their models,” says Wayne Arvidson, global director of market development and strategy at Dell Technologies.

“It’s often the case that a data stream a city might not have considered important takes on a new meaning when they assess the insights. Additionally, cities may not realize that they have data in silos, often associated with very specific workflows such as building infrastructure systems or vehicle data.”

Both use cases should also draw from unstructured data, such as images, social media posts, multimedia and web pages, and structured data.

Unstructured data like this will be increasingly crucial in sentiment analysis, which is key to citizen services.

“We’ve also seen customers using it to detect unconscious bias in systems and processes, which will be increasingly important in an AI-driven world,” says Arvidson.

Data processing

The next step is to decide where the data will be processed: does it need to be ingested — moved or replicated — to a cloud-based server, central data lake or data warehouse, or will it be processed at the edge, closer to its source?

This will depend on the use case and desired outcome. Suppose a city department tries to optimize traffic flow and dynamically re-route one of its fleets in real time because of an incident. In that case, it will need to process data at an edge server, which could be located at an intersection.

Conversely, if the city is building an AI model to power a chatbot in a city call center, it is more likely to occur in a centrally located server.

“Cities need to collect all the data possible because they won’t understand its full value until they start building their models.”

On data exploration and enrichment, Nvidia’s Charbel Aoun, the company’s smart city & spaces director—EMEA, highlights that it is also essential to introduce the concept of RAG (Retrieval-Augmented Generation) at this stage. This is especially relevant in the citizen services use case to increase accuracy.

“When you use a generative AI platform like ChatGPT, you are using a third-party model that is trained on a set of data up to a certain point, but RAG can serve as a connector to a reliable database or data source,” he explains.

So, in the case of citizen services, RAG could aid a chatbot by cross-referencing the generative AI responses with verified information held on internal databases or wiki in near real-time.

When moving to the model development process, cities must choose the right data science tools and machine learning algorithms to deliver the right workflow and achieve the desired outcomes.

Even if a city is taking the build approach, Dell Technologies and NVIDIA advise cities to begin by assessing if a suitable commercially available model exists to provide the desired insights.

“Typically, we either start with one of these existing models or utilize an integrated model from a third-party applications vendor as a base and then customize it further,” said Arvidson.

For intelligent transportation use cases, cities may look for models capable of object identification, direction detection, traffic signal optimization, or route optimization. These can then be integrated with natural language processing and generative AI tools to create a customized and unique algorithm.

Aoun explains that this is the stage at which the power of these latest generative tools can truly be utilized.

“They move us from simply data retrieval to data generation, which is really powerful,” he says. “As well as expediting really complex tasks, they enable us to interact with data differently.”

Monitor the pipeline

Finally, there is a stage of observability and performance, where the data pipeline is monitored, any drifts in the machine learning model are managed and the data product is integrated with operational systems. At this point, continuous A/B testing is also performed.

In Dell Technologies’ and NVIDIA’s experience, once a city starts obtaining actionable insight from the model, it begins to understand that additional insights can be derived from the data or combined with other tools to provide new outcomes.

Tools like natural language processing and generative AI make it relatively easy to apply more learning and data to an algorithm to uncover more insights and information.

“Just as the AI model itself never stands still, then nor should the city’s aspirations for what it can help them achieve,” the whitepaper says. 

Image credit: iStockphoto/jamesteohart