The State of Industrial AI: A Look into the Future

We recently sat down with Peter Seeberg, industrial AI consultant, moderator, and podcaster, to gain his insights into where industrial AI is headed over the next twelve months. Seeberg has a significant following, and his guests are among the top industrial AI experts at some of the world's biggest companies.

Predictive maintenance and the rise of agentic industrial AI

In broad strokes, Seeberg explained that we’ll see an increase in the maturity of AI solutions and their adoption, as well as greater democratization of the tool. By that, he means specifically that AI as a tool will become more widely available, especially now with agentic AI. This is not just for industrial applications, but also for general use in business and by consumers. In simple terms, agentic AI is hard to set up, but easy to use.

More specifically, one of the main areas where AI is continuing to gain traction in the market (and on the factory floor) is in predictive maintenance. This was one of the first areas where industrial AI was deployed, and that solution is now mature enough to be an off-the-shelf solution. 

Predictive maintenance yields significant time and cost savings for industrial processes, especially those that rely on complex, heavy machinery, such as mining. Though it’s not limited to any one industry, predictive maintenance is showing value and measurable returns across industries. Knowing when a problem is developing before it shuts down production is invaluable. 

This is now a well-established industrial AI application, capable of collecting useful equipment sensor data and presenting it in a relevant, user-friendly dashboard. New AI applications are certainly being developed, but much will depend on the availability and relevance of data for training large language models (LLMs) underpinning many current and upcoming AI solutions, the compute capacity and availability in a tight technology market, and the skills needed to fine-tune LLM algorithms. 

Driving efficiency gains with industrial AI and lights-out factory automation

The aim of overall equipment efficiency (OEE) is to get to 100%, and some days a facility might reach 92% or 93%, but 100% is not easy to reach consistently. Here is where specific, tailored industrial AI applications, including predictive maintenance, can help move that needle closer to 100%.

“This year, there has been a lot of talk about the latest iteration of AI known as agentic, which in essence means we now talk to our AI systems and tell them what we want,” Seeberg said. As agentic AI applications increase — partially in response to the ongoing scarcity of skilled workers and partially to reach a “lights-out factory” state — we’ll start to see humans called in by AI agents to review issues. Humans will have agentic AI as a tool to help them optimize. 

On the factory floor, automation has been ongoing for some years with the introduction of robotic devices, but now those too can be run by AI, instead of a human behind a keyboard manipulating a robotic arm. “Adding AI to robotics makes it possible to get to ‘lights-out factory’ sometime in the next five to 10 years,” Seeberg said.  

The concept of a lights-out factory isn't new, but we’re not there yet. Some manufacturers, like BMW, have achieved “lights out” in a section of their facility. That section is fully run by robots and AI. Human supervision and programming adjustments can be performed remotely, rather than on the factory floor. That section of the facility doesn’t need the lights on because robots are programmed to do their work without overhead lighting; hence the term “lights-out factory.”

Data quality and accuracy

Seeberg took a small detour by going back in time before moving forward. He explained that, roughly 10 years ago, the translation industry was among the first to experiment with AI in the form of machine translation. Now, the fruits of those labors are enjoyed by consumers who can simply use their phones to access a translation app. 

From those early machine translation days, we’ve learned the value of data in populating a machine learning system. Today, data scientists understand that a large language model is only as good as the data it’s been trained on. And we’re seeing that there aren’t currently many models with a lot of industrial information. To bring AI into your production and processes, you’d need to train it on your protocols, systems, equipment, and regulations, tailored to your industry. 

However, there is one company that’s spearheading a project to ensure data accuracy for large industrial foundation models. The German-headquartered company Siemens is working to build AI foundation models that can be applied across most industries. The key steps in building these are gathering the right information and making sure that the data is correctly annotated so the LLM can learn what something is and how it works. At this time, not much more is known about Siemens' ambitious project. It’s clear, though, that it will fill an important gap in the industrial AI market.

Taming AI’s power demand and bridging the skills gap

Seeberg went on to say he’d spoken with some experts in the field of microchip power usage who are looking into ways to make AI less power-hungry, but there’s nothing concrete to report in that area yet. With the rapid growth of data centers and their demand for power, a solution that can reduce that demand would yield both cost savings and a step closer toward sustainability.

In closing, Seeberg touched on the ongoing challenge across industries of finding skilled workers. One possible solution is to automate, if possible, particularly the entry-level jobs that are proving hard to fill. Agentic AI could potentially fill that gap by providing entry-level workers with a tool to help them. 

To sum it up, industrial AI is continuing to grow and develop, with agentic AI being the most exciting innovation.