Building a safer path to autonomous industrial AI

Industrial AI is moving into a new phase in which foundation models, physical AI and agentic AI are making it possible to automate more complex tasks across industrial environments, according to Arti Garg, chief technologist at AVEVA, speaking on MIT Technology Review's Business Lab podcast. Garg said AVEVA has worked on AI for the industrial sector for more than 20 years, but the type of AI has changed substantially in recent years. She cited one study suggesting almost a 78% increase in industrial-sector AI adoption over the past two years, describing it as an inflection point and almost a step function.

What distinguishes industrial AI from purely digital AI, Garg said, is that it interacts with real physical systems such as power delivery and natural-resource mining, often in hazardous environments where an unexpected decision can affect safety and reliability. She noted that newer models are by design hard to understand and are not fundamentally explainable in the way earlier AI or statistical models were, and that their behavior can change over time as they learn and are tuned.

On data, Garg described correlating telemetry from equipment like pumps or mixers with service logs and original design documentation. Newer technologies, including graph databases and AI-based matching of disparate data sets, can let an operator with a tablet fetch and correlate that information in near real time to diagnose problems. Robots moving through hazardous environments could gather data using onboard computation or a connection to an operator.

AVEVA's responsible AI framework rests on a triple mandate: that AI be secure, efficient including environmentally efficient, and that it preserve human safety and human oversight above all. Garg said the company uses a multi-layered governance approach covering both its internal use of AI and how AI is deployed in its products, with human judgment, responsibility and ethics centered, and AI augmenting rather than replacing people in critical decision loops.

On sustainability, Garg said AI is critical to managing a grid with more intermittent renewable generation such as distributed rooftop solar, and that her team has worked with Idaho National Laboratory as part of its AI grid resilience project. She also said there is no agreed-upon method to measure AI's environmental impact, and that she chairs the IEEE P7100 Standards Working Group, launched a little over two years ago, covering electricity and energy consumption, resource usage, water consumption and carbon.

Why it matters: As industrial AI adoption accelerates, developers and product teams building for plants, grids and mining sites face pressure to combine automation with guardrails and human oversight. The IEEE P7100 effort, once complete, is intended to give organizations a single methodology for measuring and reporting AI's environmental footprint.