Hi Jonny

Lots of ads showing AI already walking the floor to identify things that are wrong!

Lots more ads offering to get your risk assessments and method statements to be generated electronically - which is fine if your factory makes widgets in exactly the same way that another widget factory that was used for the intelligence to inform the bot.

But we are using modern technology to remove or reduce many risks.

A couple of decades back I was asked to advise on the confined spaces risks associated with the internal inspection of a road bridge. 

In very simple terms there is the deck on which the traffic moves, the void beneath and the deck below, all supported mostly on two hollow pillars.

Now in normal use one or two people might regularly walk through the void and the natural ventilation from one end to the other would be such that any confined space risk could reasonably be deemed "negligible" - any wind whatsoever and the effect was a howling gale through the void!

But, this full structural inspection meant bringing lots of people in and erecting scaffolding down both pillars and the howling gale was certainly not going to permeate to the bottom of those pillars, so everyone recognised that these were definitely to be classified as "confined spaces" - for the scaffolders erecting and dismantling and for the structural engineers doing the examination.

So, I asked one of our experts in bridges - "why can't they simply put a camera down and do the inspection remotely?".

"Because the standard in "DMRB" requires every part of the structure to be inspected from within touching distance." I checked and he was correct.

....and so that is what happened.

The standard has subsequently been amended. These days the use of drones for remote inspections of bridges both internally and externally has become quite common.

So AI will be used to inform an ever expanding range of decisions that have health and safety implications.

What is important is that the AI is informed by people who understand what they are asking the AI to do and who can see the pitfalls that may exist. 

Depends on what you mean by AI. Bradly speaking there are two approaches to AI.

The first relies on  creating a massive database, which the algorithms refer to when they come up with an answer. So this depends on what has been included in this universal database and it could be information that enables an AI to distinguish between your widget factory and the rather different one that it had been told about when it was first created. The problem is that you don’t know if the answer you are getting to your question is as a generic one or the specific one you want, unless you are a H&S professional which means that you don’t really need the AI in the first place!

Alternatively, the AI could have a learning ability. It does its best based on what is has been told but its work is checked by someone who might be a review group. They tell the AI if what  it is doing ok and it learns what is expected of it and eventually, in theory, it can work independently. The problem here is that nobody understands how it comes to its conclusions and if it comes up with odd answers, it is hard to challenge it.

AI could be useful for things like writing up reports based on a survey but the conclusions should rely on the human touch…for now.

Originally Posted by: Jonny95

and under experianced AI Handlers

Is this the new AI with spelling errors to appear human?