AI in Safety
What is AI, where is it being used, and who is is responsible?
AI in Safety
AI is already being used in film, television, and event safety.
People are using it to research unfamiliar activities, summarise guidance, organise recce notes, draft documents, identify possible hazards and controls, interrogate plans and ask that very useful question – what have I missed?
And increasingly, AI is being built into specialist safety tools themselves.
Some of this is genuinely useful. Some of it considerably less so. And some of it can produce something that looks remarkably like professional safety work without necessarily being anything of the sort.
So this isn’t about why we shouldn’t use AI in safety. It’s about what we should trust it to do.
Because underneath the increasingly impressive technology, Large Language Models such as ChatGPT or Claude are – essentially – probability engines. They learn statistical relationships from enormous amounts of human-created material and use those relationships to generate a plausible response to whatever we ask them.
And the word plausible matters.
An AI can produce an answer that is fluent, detailed, confident – – and completely wrong. It doesn’t know or care whether an answer is true or not. It has no experience of the project you’re working on, no understanding of the people involved, and no responsibility for what happens next. AI doesn’t own the consequences. You do.
But that doesn’t make AI useless – far from it. A useful way to think about it, is as a remarkably capable junior assistant. Give it the right task and it can save enormous amounts of time. It can organise information, interrogate a document, summarise guidance, challenge an assumption or suggest something you may have overlooked.
What it can’t do is become the competent person, even though the output sounds competent. And that’s the distinction we’ll explore in this short course. Because as AI gets better at producing convincing safety information, the important question isn’t simply what can it produce, it’s what should we trust it to do?
And answering that starts with understanding a rather unusual problem with AI: quality and accuracy aren’t the same thing.
