In this installment of EHSDA Shorts, Mwangi Ndonga, CIH, CSP, CHMM, Chair, AIHA Technology Initiatives Strategic Advisory Group, explains how EHS professionals can take advantage of machine learning to improve workplace safety.
This clip is taken from a webinar titled Improve Workplace Safety with AI, which is available for free on-demand here. This webinar is sponsored by KPA.
Transcript (edited for clarity):
Ndonga: There are many branches to AI: vision systems, natural language processing, and machine learning is one of those. It’s probably the biggest. And the idea around machine learning is this: You’re essentially coding or writing code or giving the machine data to allow it to learn.
And what does that learning look like? You give it data, you allow it to come up with patterns, and then you test the machine with certain pieces of data where I know the answer may be excluded. And then you see how the machine performs any more, like when you and I were taking tests in school, right? Maybe somebody gives you an old test, you learn the answers, but then when you take the actual test, there are no answers. So that’s what machine learning is.
So whether it’s LLMs [large language models], whether it’s vision systems to open your phone, these machines look at our faces and learn our patterns and so I can open my iPhone with my sunglasses on. That’s what machine learning is.
And so the reason I’m very interested in machine learning is it helps me calibrate or measure what kind of data I have and whether a model can learn from it. So again, if I’m trying to employ an LLM but I don’t have good literature, how do I expect this machine to perform?
Now here’s the thing, machines are very stupid, so if I just tell it to learn off bad data, it will do it. The question is whether I’m going to be really confident in what it may produce, knowing that I have bad data. So again, the concept of machine learning is you don’t just throw a bunch of data at the computer. You have to kind of tell it what you want.
You may say something like apply a classification algorithm because I want you, computer, to separate these data into three groups and then you give it the data or something like that. So there’s a variety of ways to do that.
I think if, if you’re asking yourself about AI, I would argue that can we take the data you’re collecting already or maybe enhance or pivot a little bit on it because that’s where we want to go. We want the machine to learn what you’re doing so that you can step away and maybe spend more time with the employees.
I utilize machine learning, myself either like in a desktop format with my own, let’s say IH [industrial hygiene] data, or something like that. But also I’ve worked on some projects enterprise-wide applying to machine learning.