Industrial AI won't get adopted until engineers can verify it
Key Highlights
- Process engineers are accountable for outcomes, so they'll bypass or double-check AI that can't show its work, especially in safety- or quality-critical decisions.
- AI recommendations need a direct line to the process variables, historian data and maintenance records behind them, so engineers can validate an insight without rebuilding the analysis from scratch.
- The more effective approach connects historians, control systems and maintenance platforms through strong APIs so OT teams can build purpose-built AI workflows themselves, rather than IT restricting what tools can be used.
Industrial AI is starting to catch quality deviations and equipment drift before they turn into full-blown process upsets, and Bain & Company projects industrial AI will generate $70 billion in new value by 2030. But for process control engineers, the technology's real ceiling isn't computing power or data volume, but trust, writes Mark Derbecker is the co-founder and chief product officer at Seeq, in this article on Automation World.
Unlike enterprise AI, industrial AI touches decisions with physical consequences, from worker safety to product quality to equipment integrity, so any recommendation must hold up to the same scrutiny engineers apply to first-principles analysis and statistical process control.
The article lays out why AI has struggled to earn that confidence: without governance grounded in engineering context, AI outputs can feel like an answer with no visible path to how it was reached. That's a nonstarter in regulated environments like pharma manufacturing, where every input to a batch release or deviation investigation must be defensible to auditors and regulators alike. Derbecker argues that governance shouldn't be treated as a compliance checkbox but as the mechanism that makes AI adoption possible—keeping recommendations traceable back to validated operational data.
For plants looking to move from reactive troubleshooting to proactive intervention, the article frames trust-building as a three-part discipline: embed engineering and process knowledge into how AI models are built, make every AI recommendation traceable to the underlying historian, sensor, and maintenance data, and shift IT/OT governance from restricting tools to enabling secure data access.
