Human-Centric AI: Why IIoT Systems Should Help People, Not Replace Them

Illia Smoliienko is the founder of Sivense and an expert with over a decade of experience in industrial IoT and predictive maintenance.

When I first started working in the industrial IoT sector, I witnessed a situation that has stayed with me ever since. Hundreds of IIoT sensors had been installed for a client to detect equipment defects, but within a few months, they began failing en masse, even though there were no technical reasons for it. It turned out that the maintenance team was deliberately damaging them because they saw them as a threat to their jobs.

I still recall this incident as an illustration of how automation is sometimes perceived: more technology means fewer employees. It’s a logical conclusion if you measure the effectiveness of automation by the absence of humans in the process.

Many years have passed, and technology has changed, but today we see the same fears and expectations in the development of industrial analytics with AI. The question we need to answer is this: how to design human-centric AI, where technology helps people make better decisions rather than trying to replace them.

Why AI Should Not Replace Analysts

The condition of an industrial machine isn’t just about signals from IIoT sensors. It involves actual load conditions, repair history, the specific characteristics of a particular component, the quality of sensor installation and countless other factors. Data alone doesn’t provide the full picture, no matter how much of it there is.

Let me explain using bearings as an example. For a neural network to classify bearing defects based on vibration data from IIoT sensors, it needs data on various defect types, their stages of development, equipment operating conditions and specific bearing types. If we consider roughly 30 defect classes, 35 operational contexts and about a thousand common bearing variants, that results in over a million potentially suitable combinations for training.