Industrial AI Where Hallucination Is Not an Option

In a recent interview published in Tivi (Opens in new tab), Harri Saikkonen, SVP, Software and Embedded Solutions at Etteplan, explains how industrial AI differs from traditional software and why reliability, transparency, and domain expertise are essential when intelligence is brought into physical systems. 

Generative AI models are based on probability. They predict patterns in language. In many office AI applications, occasional inaccuracy may be acceptable. In industrial environments, it is not.

“Industrial AI cannot be a black box. In factories, there is zero tolerance for AI hallucination.” Saikkonen states. Industrial AI operates in physical systems. It must account for vibration, temperature, pressure, and mechanical behavior. Errors are not minor inconveniences. They can damage equipment or create safety risks.

According to Saikkonen, decisions made by industrial AI must be explainable. In critical environments, systems must justify their outputs and operate within the constraints of physical laws. Intelligence in this context is connected to machines and physics, not to language structures.

Industrial AI often runs on embedded systems at the edge of the network rather than in the cloud.These environments introduce practical constraints:

“When developing intelligence for industrial processes, we must consider hardware constraints and energy efficiency,” Saikkonen explains. Edge systems do not have the same computing resources as cloud environments. Optimization is therefore essential. AI must be designed to function reliably within these limitations.

Saikkonen highlights agentic AI as the next development step in industrial intelligence. Traditional predictive maintenance systems detect anomalies and generate alerts. Agentic systems go further. 

For example, an embedded AI agent may detect abnormal vibration in a pump and determine that a bearing failure is developing. Instead of only sending an alert, the system can initiate coordinated actions such as checking spare part availability or triggering maintenance processes. This represents a shift from passive monitoring to active orchestration. 

Artificial intelligence is central to Etteplan’s growth strategy. The company aims to have AI-driven service solutions account for 35 percent of revenue within three years. Developing industrial AI requires broad technical and domain- specific expertise.

Engineers must understand:

Industrial AI must meet strict safety requirements. Systems cannot introduce new risks into production environments.

At the same time, AI is changing software development itself. Saikkonen notes that AI tools help developers work faster, reduce routine tasks, and improve code consistency. This leads to significant productivity gains.

Saikkonen emphasizes that AI initiatives should not be isolated within individual units. Greater impact is achieved when intelligence spans the value chain rather than optimizing single functions. He also highlights that technology alone is not enough. Organizations must address ownership, accountability, and change management. AI systems require continuous development and maintenance to remain reliable.

Industrial AI operates in environments where reliability, transparency, and safety are essential. It must function within physical constraints and strict operational requirements. 

As Saikkonen emphasizes, technology alone does not create value. Successful implementation requires clear ownership, a strong understanding of industrial processes, and continuous development. In critical industrial settings, intelligence must be engineered with the same discipline as the systems it supports. 

If your organization is exploring industrial AI, edge solutions, or agent-based systems, careful planning and clearly defined responsibilities are essential.

Contact Etteplan to discuss how to implement industrial AI safely and effectively in your operations.