How GenAI can supercharge smart cities

Generative AI is shaping up to be the technology that finally makes smart cities truly viable.

Since we realised we could put small, cheap sensors into pretty much any device, the possibility of the smart city has seemed tantalisingly within reach. If we know everything about a city, we can run it better, surely? The challenge, though, has been turning that vast tide of data into a wave of insight we can surf.

One of the more exciting possibilities that generative AI opens up is transforming tech ideas of the past that have failed to truly take hold. Take the internet of things — the idea that we can improve devices by making them connected, thereby allowing us to both collect data and control them.

For enthusiasts, that might mean controlling the lighting and heating in your home via a mix of voice and algorithm. But for cities, this is digital transformation at an urban infrastructure level. Crucially, generative AI has the potential to fill the lacuna between the sensor data and the control, by automating adaptive processes based on demand and availability.

By instrumenting critical systems with connectivity and adding sensors to existing infrastructure, cities can better understand their resource utilization, citizen behavior and service gaps. The advent of AI, and more recently, generative AI, has opened up a world of possibilities for expanding access to services and enhancing efficiencies.

In essence, generative AI, as a predictive technology, manages to turn vast seas of data into insights. And that allows us to model the likely future from current datasets. And that’s what unleashes the internet of things in an urban setting.

The challenges and opportunities

There are challenges, of course: the compute demands of AI are well-known, and expensive in both financial and climate terms. The latter can be offset by, for example, fewer emissions based on a better-performing public transport infrastructure. The former, though, is more of a barrier for cities emerging from a torrid financial period.

But you also need the communications infrastructure in place to reliably collect, transmit, process, and transmit out again the collected data. But to what end?

Well, there are numerous potential applications of this:

  • Energy supply management. As we transition from consistent energy supply forms to more variable green energy, managing predicted peaks of supply and demand will be critical.
  • City maintenance. A combination of faster reporting of faults, possibly through AI detection, and predictive maintenance that allows repairs before the system fails, could keep cities running much more efficiently.
  • Parking management. Real-time prediction of demand could lead to drivers being directed to the place they’ll most efficiently find parking.
  • Traffic management. The bane of most cities, AI could help predict traffic flows, react quickly to emerging issues, and help public transport become a better experience.

Smart city mobility

The idea of smart city mobility is where we’re likely to see bit impacts quickly —and possibly even see a long-gestating idea come to fruition. Another technology that’s been a few years away for a decade or more is autonomous vehicles — self-driving cars and trucks. While some companies — like Tesla — that are prepared to play on the bleeding edge have shipped products, most are holding off. Why? Getting a computing system to constantly adapt to a changing landscape in realtime is really hard. But what if those systems could be predicting what’s likely to happen, and this only having to react to deviations from that pattern?

That is what a new wave of AI-first start-ups is exploring:

Waabi’s model works in a similar way to image or video generators like OpenAI’s DALL-E and Sora. It takes point clouds of lidar data, which visualize a 3D map of the car’s surroundings, and breaks them into chunks, similar to how image generators break photos into pixels. Based on its training data, Copilot4D then predicts how all points of lidar data will move. Doing this continuously allows it to generate predictions 5-10 seconds into the future.

Generative AI is essentially a predictive technology. The consumer versions we’re familiar with predict the next word or pixel based on their understanding of pre-existing patterns. That can apply to urban infrastructure as much as words and images. And, of course, those models could become more powerful if they’re not just fed data from the car’s sensors — but from the urban environment itself.

Avoid smart city conspiracy theories

That’s going to be a more tricky problem to solve. But the challenge here is less technical and more political. The idea of free-flowing, self-driving traffic complemented with efficient public transport is an attractive one. And then you point out that it might involve tracking the location of every single vehicle and sharing it with others, and the conversation changes…

As the IEEE puts it:

Smart cities’ reliance on various surveillance technologies and data collection platforms increases the risk of this personal data being misused, either intentionally (as in cases of identity theft or targeted advertising) or unintentionally (through careless handling or insufficient security measures). Hence, addressing data misuse should be a pivotal concern within the realm of smart city planning.

The challenge of building public acceptance should not be one we underestimate. Even benign ideas like the 15-minute city have become the centre of conspiracy theories and political protest. Once explicit data collection and reuse are brought into the equation, people will be very uneasy. As much time and effort will need to be put into ideas like anonymisation of data, control of whom it’s available, and — crucially — what technology partners are allowed to do with it. The AI companies’ rapacious greed for more training data has not gone unnoticed.

A new smart city infrastructure

However, the potential is there, and the support infrastructure to do it is starting to fall into place. On the strategic side, AI for Good has an Intelligent Cities Project, which helps cities work out which initiatives might help them hit the UN’s Sustainable Development Goals, for example. On the technical side, NVIDIA, the chip maker whose business has boomed through the use of its products in AI, has developed NVIDIA Metropolis, a framework for large-scale AI based on input from millions of sensors — exactly what we’d expect in a smart city.

That tide of connected data is shaping up to be the kind of wave we can surf into a more sustainable and more efficient future — if we can persuade people that their data will be safe.

Image by Peerapon Chantharainthron / Unsplash.