In this article, Peter Bateman from Thomson Reuters reviews reviews Cam Stevens' Auckland NZISM workshop on safety technology on 31 July 2026.
"AI mediocrity is like ordinary mediocrity but comes with better spelling and grammar so is hard to spot."
Such zingers came thick and fast during Cam Stevens' Auckland NZISM workshop on safety technology, with the emphasis on AI.
Another one? "Fluency reads as competency to humans." Former British PM Boris Johnson pops unbidden into your reporter's head for some reason.
Stevens began with a warning: AI's abilities and reach into everyday life means we are now perhaps only six months away from the point where H&S practitioners will either have got on board with AI or will have been left behind.
It will become a matter of authority: if anyone can use AI to generate a plausible risk assessment in a few seconds, what makes yours worth more? What value do you add?
He acknowledged some H&S people are frustrated because their organisations are not enabling them to work with AI. Nevertheless, he encouraged us to have a mindset alert to the technology trends shaping work – they are converging – and not to be seduced by shiny tech toys, instead focusing on the problem you need to solve.
"The marketing in safety tech is alluring and you will be enamoured," he warned.
His consultancy, PKG Safety Innovation, has a mantra: Problem first. Tech second. Always ask yourself: what is the problem I am trying to solve?
"There is no point in seeking solutions that are tech-driven if you don't have a good problem statement." Ironically, he noted, this is a good use-case for AI: helping you to frame your problem statement.
A quick survey of the 50 or so people in the room asked them to rate their feelings about AI, from 1 (Deeply concerned) to 10 (Highly optimistic). The result was 7.1, which coincidentally was exactly the score generated by earlier workshops in Wellington and Christchurch.
On the use of AI with CCTV to detect people vs mobile plant incidents, Stevens reckoned this was a great use of one of the HOP principles: how leaders respond matters. "Once you see it you can't unsee it." He noted vendor language can be provocative ("Spot the violation!") and cautioned that without trust the technology would not be successfully deployed.
On the use of LLMs like Claude and ChatGPT (or Copilot, which uses them), he has mostly abandoned the keyboard and just speaks into his computer. "AI loves voice. LLMs love to be spoken to."
Companies currently asking workers to type incident reports into the phones, in the field, are missing out crucial information. If they speak into an AI model instead they will use ten times the number of words, generating far more rich intelligence data for AI to summarise.
He gave a live demonstration of something he described as the bane of his existence: having to fill out a prequalification questionnaire. "One of the most frustrating activities you have to do as a health & safety professional."
He spoke into his laptop, telling it what he wanted in a lengthy, stop-start, slightly rambling set of instructions. The AI interpreted transcript appearing on screen was spot on, and in a few seconds the AI agent had used his consultancy's pre-existing documentation (work at height procedures etc) to fill in the prequal questionnaire.
Cue audible gasps from the room.
Stevens talked about AI agents – which go beyond just summarizing and take action – and said they aren't scary, "just a set of instructions".
However, another warning. "What happens to the skill of risk assessment when you can just press a button and generate a JSA?"
Hence the lesson about mediocrity and the challenge for humans of detecting it when it is delivered in mellifluous prose from the mouths of AI tools (or politicians).
This article is published with the permission of Thomson Reuters New Zealand. It was written by journalist Peter Bateman and appeared in their Safeguard Update enewsletter on Monday 10 August 2026.
