Hardware
Cities start to think
Smart cities may get a new lease on life as public services are being rebuilt to act with machine intelligence, writes ARTHUR GOLDSTUCK.
Imagine a drone that flies over farmland, spots a weed, and deploys a laser to kill it. Or imagine a digital court assistant trained on legal protocols, issuing permit guidance based on thousands of regulations. Or a disaster management system that predicts typhoon damage before a drop of rain falls. It’s not science fiction. These are working systems, already deployed, and they signal a shift in how cities and governments are beginning to think.
“We believe the future city has a brain that can think for itself,” said Hong-Eng Koh, global chief public services industry scientist at Huawei, in his keynote address at Huawei South Africa Connect 2025 in Sandton last week.
“Everybody knows that we are in Industry Revolution 4.0, digitalisation, right? Wrong. I think we are already in 5.0,” he said. “We’re going to see a lot of collaboration between human and AI.”
This 5.0 era, according to Koh, is defined by intelligence embedded directly into public infrastructure, from automation and connectivity to reasoning.
“Traditional smart city, you still need to build very big intelligent operation centre, you still need human to do a lot of activities,” he said. “But we have already seen what a cognitive city is.”
Huawei first introduced the term “cognitive city” in 2020, offering a vision beyond IBM’s smart city model. The new approach integrates AI into urban systems capable of analysing real-time data, making decisions, and executing actions without needing command centre oversight.
The shift from digital to cognitive is visible across sectors, said Koh. In Vienna, drones monitor crop health and trigger autonomous vehicles to destroy weeds using boiling water or lasers.
In classrooms, AI tracks student engagement and teacher performance, while lessons are auto-recorded and tagged for review. In hospitals, digital avatars offer basic consultations and diagnoses.
“Of course, what is holding back is the law,” said Koh. “Whether the law allows the AI to prescribe you control drugs.”
To achieve this kind of intelligence, said Koh, systems needed “secondary training” of large language models, fine-tuning AI using internal, organisation-specific data. He pointed to DeepSeek, the Chinese open-source model with 760-billion parameters, as a turning point.
“The training cost compared to OpenAI is less than 10%, the inference cost is 3%,” he said.
But the more important factor, he said, was sovereignty.
“When you use a closed-source large model, you are literally sending your internal classified data to a third-party server for training,” said Koh. “With DeepSeek, 100% open source, it means the base foundation… can be in your data centre behind your firewall for you to do the secondary training.”
Koh laid out three core use cases.
First: information discovery, where users interact with avatars that respond with up-to-date, contextual answers.
Second: operational assistance, such as AI automating permit approval by applying every relevant building regulation. “In the old days, the public server will have to check one by one,” he said. “But now with all these rules uploaded as secondary training, DeepSeek can take over.”
The third use case, decision support, deals with large-scale, real-time data to tackle complex problems like fraud detection and disaster prediction. The city of Shenzhen, for example, now operates 2.2-million sensors to monitor environmental threats.
“Not just cameras, also something to detect all these unusual activities,” said Koh.
He warned that countries lagging on digitisation would find AI adoption a challenge.
“No data, no AI,” he said, recounting a visit to a police chief in Africa who requested help deploying AI despite still using paper-based reports.
South Africa, he suggested, has made unusual but valuable policy decisions that can accelerate transformation, notably the State IT Agency Act, or SITA Act, which mandates unified cloud infrastructure.
“This law was debated in parliament over a long time, but in most countries there’s no such law. This is why it’s very difficult to push them to come together, to overcome the silos.”
He outlined eight success factors for intelligent transformation, ranging from vision and governance to digital culture and security. Huawei’s own warehouse in South Africa served as a case study.
“By using only 5% of the rooftop for solar panel, we are already saving ourselves R350,000 per year,” said Koh. Once energy trading becomes legal, Huawei plans to expand rooftop solar coverage at its South African warehouse to 100%, aiming to sell surplus energy back to the grid and generate an estimated R5.7-million in annual revenue.
He cited similar transformations underway in the UAE, where the national research network Ankabut has moved beyond its academic roots. Now it operates high-performance computing, AI-as-a-service, and public cloud built on Huawei infrastructure.
He gave the audience a stark reminder that the human element cannot be sidelined.
“Singapore, the largest bank… announced this year they’re going to cut 4,000 people because of AI. And the remaining 30,000 people must learn AI.”
* Arthur Goldstuck is CEO of World Wide Worx, editor-in-Chief of Gadget.co.za and author of The Hitchhiker’s Guide to AI.
