AI is taking technology advances in industrial environments to a new level, as Steve Rogerson found out in a chat with Linir Zamir from Telit Cinterion.

Factories ain’t what they used to be. They have moved a long way from the dirty and often dangerous places of the 1970s, when the teenage me earned his keep during school and university days, through the sci-fi-like robot car factories that appeared in adverts on TV screens in the 1980s, to the modern, connected entities of today.

In recent years, the advent of the IoT has had a massive impact to the extent that IIoT (industrial IoT) has become a trend in its own right. Sensors give plant managers visibility into what is happening on the production floor as never before. Machines, belts and storage areas are streaming data constantly, there are cameras showing real-time views of what is happening, but with that comes a problem – what to do with all those data.

Industrial IoT has hit a point where relying on humans to act on those data is slowing decisions and creating delays. Yet, as in other sectors that have seen IoT change the way they operate, the answer could already be here thanks to the massive developments in the past couple of years in the use and scale of AI.

That is the view of Linir Zamir, AI engineer and team leader at Telit Cinterion (www.telit.com), whom I caught up with last week.

“Many factories are so much more advanced and they have data going through them at light speed,” he told me. “They are so big, it is impossible that someone can follow that massive data flow. But now we have the tools to help the manufacturers; we can add a layer of understanding.”

Before agentic AI, most factory IoT installations operated using thresholds on temperature, pressure and so on. If those thresholds were breached, an alert would be provided and a person would have to investigate.

“That can be pretty tedious,” said Linir, “as the person tries to read and understand all the data and work out why there is a fault. Now we have agentic AI that can add context. It can suggest reasons for the fault and what can be done.”

However, manufacturers have been working for years with robots, PLCs and the like, and adding agentic AI is not straightforward.

“They also don’t want to send their data to the cloud for security reasons,” said Linir.

The catalyst here has been advances in edge computing. This lets them keep their data within the factory walls.

“Agentic AI works in a small device that can sit in the factory and run fully on the edge,” he said. “The data never have to leave the factory. This is relatively new. The hardware has really caught up with the software in the past year.”

There is, however, a certain amount of justifiable scepticism that software can figure things out more quickly and better than an experienced human operator.

“They have to learn to trust it,” said Linir. “There will always be a person that verifies the agentic AI and makes sure it is doing a good job. This is how you gain trust.”

He said this was especially important in the early days, as the AI will make mistakes but there will be fewer errors over time as it learns. With every second a robot is down costing money in lost production, speeding up the fault finding will be a blessing for many workplaces.

“The AI can read the error code and go over the entire documentation, find what the error meant and what steps need to be taken,” said Linir. “Then the operator can act on that information. But this is only the first step. The next step will be the agentic AI acts on the instructions, but still with permission from the operator.”

In the future, the AI may be able to do everything without consulting the operator, but for now there will always have to be a person in the loop as AI is still not 100%.

“Someone has to make sure the AI is not hallucinating,” said Linir.

That happens. Agentic AI does go off on its own tangents and comes up with stuff that can only be described as hallucinations. One day that will be fixed, but we are not there yet.

But where we are is AI that can combine information from video and sensors, spot things that might go wrong before they do, accept data in various formats, work out what has caused a fault before the operator even gets of their seat, and – this is important – suggest to the operator what the fault might be and, in some cases, ask for permission to fix it. That is not yet the full autonomous factory of the future but just think how much time this will save and is saving for factories that have embraced the technology.

“As time goes by, the hardware will become more powerful and available in terms of price,” said Linir. “Right now, it is expensive, but moving forward that will become less of an issue.”

As to what’s next, that is difficult to predict as the AI world is moving so fast.

“We never know what tomorrow might bring,” said Linir. “Every day there is a new company that has created a new AI or hardware. We are getting surprises regularly.”

However, the science-fiction dream of factories that run themselves completely may still not be here, but in the past year advances in AI mean the dream is now a goal.