I’m curious where people have landed on AI in Health & Safety so far – beyond the hype.
I’m seeing some genuinely useful cases starting (first-pass drafting, summarising incident notes, pulling themes from audits / inspections), but also plenty of AI-washing – products being described as “AI-powered” when it feels more like a rebrand of existing features with a bit of automation bolted on.
As someone who’s naturally pretty risk-averse, I’m cautious about any AI roll-out in our world – but as a fan of tech and efficiency, I’m also excited to see where it can genuinely remove admin and help people focus on the stuff that actually prevents harm in the future.
Small anecdote from me: I was mid-conversation with a market leading (paid-for) AI tool last month about something completely unrelated (ironically an internal presentation on use of AI / AI policy) … and it suddenly generated an unprompted AI version of The Last Supper but with Lego characters. No prompt for it. Funny, but also a decent reminder these tools can be surprisingly confident and occasionally random – which isn’t what you want anywhere near safety-critical decisions.
A few questions to kick things off:
Is anything AI-related actually working for you in a practical H&S context?
How are you managing the risks (accuracy, bias, confidentiality, over-reliance) when AI is used at work?
What are your hopes for AI in H&S over the next 12–24 months?
Keen to hear real-world examples – good, bad, and “meh”.
I tend to use it as a research assistant. I might ask it to ‘tell me what you know about topic x, y, z’. It then goes and does all of the searching for me and gives me a concise precis of what’s out there. That, in my opinion, is better than doing the same Internet search myself. I can avoid the sponsored posts, endless GDPR cookie screens, etc.
But it’s important to remember to ask for references, links, etc. so that I can fact-check what I’m being told before using that information to help me make decisions.
I’d like to see some of the data analysis beyond the usual accident stats, maybe around what are people doing when they have the accident. More than the usual categories of manual handling or work at height. Are they trained or not, how long since training, time of day. A lot more to narrow down the scenarios we need to guard against. We are still using the sword of damocles on training and competence.
Good point. AI could be used to ‘read’ some of the accident descriptions and pull out keyword and themes (e.g. distracted, noisy, tired, pressure, etc.)
It sounds like a comic book villain, but apparently it’s the unauthorised / unregulated use of AI tools in organisations .
Fortunately, and this is this is excellent timing - I’ve just created an E-learning course on ‘Using AI at Work, and we’ll be rolling out internal training here at Opus soon
.
