Using AI to Manage the Risks of Infrequent Events
By Ed Rutkowski
Many people marvel at old photos of iron workers walking across beams several stories above the ground, with no harnesses to save them if they slip. Thankfully, workplace health and safety has come a long way since those days. But one thing that hasn’t changed is the human mind’s ability to understand risk: we’re no better at that now than the iron workers, or, for that matter, our long-dead ancestors.
“Our brains are bad at understanding infrequent risks, like the risk of dying,” is the way Keith Bowers put it in an educational session held yesterday at AIHA Connect 2025 in Kansas City. Bowers, who spent 20 years in OEHS at Honeywell before founding his own consultancy in 2015, was recently trained to develop code for artificial intelligence systems. Now, in partnership with Owens Corning, Bowers’ company is creating AI tools that can help predict high-consequence, low-frequency events.
At AIHA Connect, Bowers and his co-presenter, Ed Puhala of Owens Corning, discussed the AI systems they’re using to identify risks that would otherwise remain hidden until a tragedy occurred. One of the reasons these risks are so difficult to recognize is that they tend to arise from everyday situations. When workers walk next to moving forklifts every day for years without incident, for example, they become inured to the serious risks presented by these massive machines.
At their session, Bowers and Puhala showed videos of near misses involving powered industrial forklifts, which they have identified as a major risk at some Owens Corning facilities. The videos were captured by temporary cameras over a period of weeks, and the data was analyzed by a computer vision model.
One of the videos, taken at a facility that manufactures roofing products, showed a worker repeatedly walking beneath a raised load on a forklift. “A potential fatality if I ever saw one,” Bowers said. “This would have gone completely unnoticed until a tragedy happened.” Puhala emphasized that the facility has a sterling safety record and goes many months to years without a recordable injury.
Another video taken at a composite lumber factory indicated that workers let their guard down around forklifts when the machines were on the opposite side of a pallet stacked with products. The perceived safety was an illusion: if a forklift, which weighs thousands of pounds, had struck a pallet, it certainly would have injured or even killed anyone on the other side of it.
Still in their infancy, AI systems have weaknesses that OEHS professionals need to be aware of. They aren’t particularly adept at estimating scale or range, Bowers said; for this reason, they can’t be used to sound an alarm, for example.
But for the kinds of applications Bowers and Puhala discussed, AI represents a new tool for protecting workers. Until recently, Bowers said, OEHS staff might have attempted to identify risks by reading incident reports, trying to determine if any of the descriptions represented a potential fatality. Now, OEHS professionals can use AI systems to flag risky situations instead of relying solely on professional judgment.
“We cannot rely on our gut,” Bowers said. “We have to use data.”
Ed Rutkowski is editor in chief of The Synergist.
