Steve Rogerson talks with Digi’s Bob Blumenscheid about AI and security.

Not that long ago – I am talking two or three years – IoT looked very straightforward. You had various sensors that gathered information, these were linked through a network to somewhere the data could be processed, and information based on those data was presented in a useful way.

OK, we didn’t think it was straightforward. We debated on what types of sensors and how they were powered, we discussed the pros and cons of the networking technology – the Sigfox versus LoRa debate seems distant history now – and companies presented alternative interfaces and data formats.

But now, just a short time later, all that is being turned on its head with the giant leaps in AI technology. This intelligence is being deployed at the edge where those sensors gather the data so decisions can be made in real time, whether to control a piece of industrial equipment, an autonomous vehicle or, as is becoming incredibly common, a robot. This is all part of the physical AI revolution that has burst into our lives.

“It is a crazy time,” Bob Blumenscheid, senior product marketing manager at Digi (www.digi.com), told me this week. “Two or three years ago, our customers were telling us they needed an NPU in their products to future-proof them.”

We are talking IoT routers and gateways here. They weren’t quite sure why they needed an NPU but thought they better have something in there for the future. These were the wise ones.

“That has now completely changed,” said Bob, “because we have AI at the edge. They want to run AI on their routers and gateways and they don’t have the horse power or memory. So we are seeing a quick transition, we are seeing processors taking things up a notch to higher Arm cores, we are seeing customers building AI centric devices. It is no longer future-proofing; they need AI at the edge today.”

There are clearly many practical advantages of this, whether predictive maintenance on machines or traffic monitoring in smart cities, or anywhere decisions can’t wait for data to be streamed to somewhere else for processing, especially given the limited bandwidth of some traditional IoT networking technologies.

“You can move to having the analysis taking place right at the device,” said Bob. “This means you don’t have to stream data continuously. You don’t need large data rates, but you can analyse data as they go by.”

There are challenges. There is, as mentioned, a need for more processing and memory. There is a need to train these AIs, something that is not a one off but has to happen all the time.

“You train them in the cloud and deploy them at the edge,” said Bob. “So you need secure over-the-air updates. Once this was just for adding features.”

Talking of security, that is the other biggy, not for the future but here and now. The next major date in the EU’s Cyber Resilience Act (CRA) is just a month away. From 11 September, mandatory reporting arrives for actively exploited vulnerabilities and severe security incidents.

“We have added security engineers to manage the CRA,” said Bob. “September 11 is the day when customers have to be ready to report active vulnerabilities. We are helping them with that.”

He said this was not just for those in the EU, but a challenge for everyone, whether in the USA, UK, anywhere.

“Each embedded implementation is different,” said Bob. “At Digi, we have expanded our security team. We have added security people and they have to support the product people for the CRA. Each vertical has different requirements.”

So AI at the edge and new security regulations, all happening at once.

“It is evolving very quickly,” said Bob. “How companies react will define their success.”

The other effect, as always, is cost. IoT devices at the edge used to be cheap sensors. There are loads of them out there. That is fine if you have a sensor in your garden that determines when to turn on the sprinkler, but for critical infrastructure that was never cheap, and now it is going to be more expensive. On the other side, AI at the edge does reduce the amount of data that needs transmitting, so there can be savings there.

Bob told me the story of a customer with a traffic light application that used an old microprocessor. They thought they were OK with this as the lights didn’t need to go faster. But now they have to be secure and load data, and suddenly they needed a drastic upgrade.

“We are also seeing new markets springing up that are not tied to a traditional approach,” said Bob. “The most amazing of these is robotics. That is exploding. These robots are embodied AI. Whether they are making coffee or whatever, they need AI.”

Today, we are already seeing a lot of products being redesigned for security and AI. These are not bolt-ons. Other companies will follow in the coming months and years, and there will be new applications as AI develops and people realise its potential.

“It is very exciting,” said Bob.