The AI Running Your Power Grid Cannot Afford to Guess Wrong

Key points

• Industrial AI adoption rose roughly 78% over the past two years, according to a study cited by AVEVA chief technologist Arti Garg.
• AVEVA, which builds software for oil refineries, power grids and water treatment plants, has worked on industrial AI for more than 20 years.
• Garg sits on an IEEE working group developing a standard way to measure AI's environmental footprint across electricity, water, carbon and broader energy resources.
• The new wave of industrial AI combines three technologies: general-purpose foundation models, physical AI (software that powers robots and autonomous machines), and agentic AI (software that handles multi-step tasks without constant human direction).
• AVEVA's responsible-AI framework puts human oversight above automation speed in any decision touching safety or critical infrastructure.

A factory pump starts behaving oddly at 2 a.m. The operator needs to cross-reference sensor readings, the last maintenance log and the original engineering specs. Traditionally those live in separate systems that don't talk to each other, and the hunt takes time nobody has in a crisis.

That bottleneck is exactly what Arti Garg, chief technologist at AVEVA, says the latest AI tools can finally crack. Speaking on MIT Technology Review's Business Lab podcast, Garg described how graph databases, a data-storage approach that maps connections between different information sources rather than keeping them in silos, combined with pattern-matching AI, can hand an operator a coherent real-time diagnosis instead of making them chase it down.

It sounds like a clean efficiency win. The catch is where these systems live.

Why does getting this wrong matter so much?

Industrial AI runs inside oil refineries, power grids and mining operations. A bad recommendation from an AI assistant in a factory is not like a bad autocorrect. It can mean a valve stays open when it should close, or a drone misreads a sensor on a high-voltage line.

Garg draws a firm line between what machines should decide alone and where a human supervisor must stay in the loop. The AI suggests; the experienced worker decides. AVEVA's responsible-AI framework makes that explicit: guardrails define where automated systems can act and where people remain responsible.

Foundation models, the large general-purpose AI systems behind tools like ChatGPT, are new to industrial settings and harder to predict than the narrow, rule-based software plants have relied on for decades. Garg's argument is that the answer to unpredictability isn't slower adoption. It's building the oversight layer first.

What does the broader picture look like right now?

This isn't a lone company urging caution while rivals sprint ahead. When we reported on 30 September that researchers are calling Asimov's robot laws inadequate for the current moment, the same tension was visible: capability is outrunning governance, and the industry knows it. Meanwhile Arm pulled more than 80 companies into a shared robotics ecosystem and Universal Robots launched a seventh-generation cobot, a collaborative robot designed to work alongside people, to lower the barrier to AI on factory floors. The sector is accelerating. What Garg is pushing for is a framework that moves with it rather than behind it.

The sustainability angle adds a layer most coverage skips. Managing renewable energy grids already demands real-time, multi-variable optimisation that AI handles well. But AI itself burns significant power. Garg's IEEE working group is trying to build a standard scorecard so companies can measure what their deployments actually cost the planet, not just what they save.

"Autonomous systems, whether they're robots or drones, are really going to change the way that we work in plants, in power systems, on mining sites," Garg told the podcast. "In a way, that will make these types of operations more efficient, much safer for the human beings involved and more productive."

My read: the 78% adoption jump is real, and the technology is ready to do genuinely useful things in places it couldn't reach before. What lags is governance. The companies that sort that out first will have the cleaner safety record when something inevitably goes sideways. Garg is right that experienced workers need new ways to apply their expertise rather than find themselves bypassed by it. Watch the plants that figure out that handoff.