With the popularity of artificial intelligence platforms, companies are asking themselves whether products such as ChatGPT can be employed to generate a health and safety manual. Yes they can, says BCCSA Regional Safety Advisor David Dunham — but unfortunately it won’t be very good.

He’s experimented with a range of generative AI platforms to see exactly what sort of documents they create and where they fall short.

“In most cases, the manualsit creates are generic, full of repetitive phrases and heavy on the adjectives,” Dunham says. “There’s also a real danger that it might commit your company to doing things it isn’t prepared to do.”

On the other hand, he says, platforms such as ChatGPT can be useful for a number of related tasks, such as helping you to begin articulating your ideas, priming the pump when you’re stuck, or performing research on your behalf by fetching legislation, case studies or WorkSafeBC decisions.

“I’ve asked it to help me craft an argument that is similar to one that I’ve supplied,” he says. “It will give you a decent start, but that output will still require a heavy amount of editing.”

Dunham has had some success in offering photos of workers to an AI platform and asking it to identify ergonomics hazards.

“You could instruct it to compare the photo to WorkSafeBC’s ergonomic risk factors and then ask it how to control those risks,” he says. “It can do that fairly reliably, although I’ve also seen it make things up.”

Dunham cautions employers using consultants to make sure they’re not being overly reliant on tools such as ChatGPT to generate material. For example, ask a consultant producing a confined spaces program about their experience in developing these programs and check the material they produce against the regulations to ensure it’s correct.

His verdict on the state of AI: at some point in the future it will be able to produce a relevant and usable health and safety manual.

“Right now, it’s better at showing me pictures of dolphins.”