The evolving role of AI in fire safety

Welcome to the FPA's podcast. Assembly Point provides a collective space in which industry leaders can explore the most pressing issues in fire safety and share expert information and advice. Join us for the 2024 Summer Series as we hear from some of the most respected figures in the industry and continue to move the debate on fire safety forwards by identifying ways to work together to improve standards. Please be aware that the views expressed by guests are their own and not necessarily those of the FPA. We hope you enjoy this episode. Hello and welcome back to the FPA's Assembly Point podcast. I'm George Edwards and I'm delighted to be joined by Adam Sanders, the technical director at RISSPACE and Peter Thomas, who's the strategic growth executive at Intuity and also the host of the My Safety Podcast. So Adam and Peter, welcome and thank you very much for taking the time to join us today to talk about the use of artificial intelligence coming referred to as AI. I think the first question is what is AI? It's a term that seems to be thrown around these days with companies rebranding on technology as AI. What does AI mean to you if I come to you first, Adam? Yeah, I think the thing is it can mean so much on it. It's for a lot of people it's using chat GPT to go and get an answer for something that they can't be bothered to type up or do the research for themselves or because they're struggling to write something. Even if they've got the knowledge, it's kind of easier to use AI to build those or chat GPT to build those prompts for you. But I think for me it's much wider than that. It's managing lots of data and it's being able to query lots of data. And I think there's a huge difference between using chat GPT which is just using whatever it finds on Google versus kind of enterprise AI where you're creating something for your own business or your own requirements which is going to be quite specific. Yeah, is that are you see a beta? Yeah, I love this question George, thanks very much. I mean I'm a safety professional so our CTO in surety is going to be cringing when he listens to this answer. But obviously we met up at the NEC show and when you went up to the, what was it a couple of months ago at the Health and Safety show and almost every single stand had some reference to AI on it up at the Health and Safety side. And like Adam said, it's everywhere at the moment but it's, I want to Google this, it says perform a task that would take human intelligence. But I like to think that we're moving way beyond that otherwise that means it can do what we can already do. But yeah, you're looking at LLM's, large language models, machine learning, artificial intelligence, a generative artificial intelligence. So it's, there's so many aspects to AI and I think that's the exciting for me is like the exciting thing is like how can we bring that into fire safety to improve to go beyond what we can already do as humans. No, yeah, I mean, in the world of fire safety, how, how can it be used at the moment? You know, I've heard of people using it until Health and Review reports and things, I think that's something that you both talked on at that show and yeah, be interested in more about your sort of thoughts and experiences to how it is being used and at the moment. But I think we've got, we've certainly got quite a few clients that are using it for, for the review and to suggest to them, have you considered these things. But the problem is the data isn't there to do that particularly effectively. It's a little bit of a backup, it's a bit of an aid memoir. And I think for a lot of people or for the people that are using it, it's quite fun. You know, we, I mean, I don't, but the fire safety world does some quite interesting stuff. Most of that interesting stuff is going out, visiting things, thinking about things, talking to peers about it. The really boring bit is sat back at the office, trying to write it up and trying to convey those thoughts in a simple language as possible to the end user who often isn't technical. And I think at the moment, one of the things that AI is quite good at is digesting some text and saying, you know, and asking it, how can I make this clearer? It's not necessarily, is it technically correct? But how can I make it clearer? And I think that's really effective for a lot of people. Asking AI to write something for you, I think, is generally terrible. And I can spot it a mile off, and I get really frustrated. It's got no character to it. It's got very little flow. But in terms of content, it's really cool. But I don't think there are many people in the fire world that are doing much more than that. I mean, at the show, we had Tom on from Tennyson Sweet, who's created his own GPT, which I think at the moment for most people is the best you can do in a small business. Yes, there's some much bigger, cooler stuff you can do. But in terms of, like, I do know what, I'm going to go and put ADB in a whole load of British standards. And I'm going to put that into my own custom GPT. And then I can ask it specific questions, rather than me having to wait through it. That's, you know, it's a time saver. It's not, it's not really replacing humans. It's not replacing knowledge, because you kind of know where that information is. It's just the time that it takes you to get to it and to be able to extract some specific wording. Yeah. So I think there's so much the fire sector can do, but I'm not sure we're doing that much with it, other than scratching the surface. Yeah, I think that concept is sort of uploading British standards and guidance and legislation up to a language model, which some of those sort of behind a paywall at the moment, so you won't get it from from chat GPT and, yeah, and doing that uploading it to a model and then using that to, you know, like like a like a colleague to chat to and it can pinpoint those exact bits and be a faster method and tell you where to look. That interest me quite a lot, as something that seems quite viable. And I don't know what your thoughts are. Feature, if this is any more, you think it can be done at the moment? Yeah. So I think for me, the concern at the moment would be around the governance and the management of the use of AI and fire safety. So because we don't have the official channels to use AI in terms of to my knowledge and someone please message me on LinkedIn if I've got this wrong, but there aren't that many fire safety companies out there that have these AI models. So people are probably using chat GPT and that's really concerning for me. So if I was responsible for fire safety within an organization, people could be using chat GPT to produce a document. So I give you an example. It's very easy to go to chat GPT. What is the relevant British standard or what guidance should I be looking at for X scenario and it will provide that guidance? But as you say, it's currently behind a paywall. So chat GPT won't have actually seen the British standard. It will have seen a third hand interpretation of that that it's found online. And that's a real concern for me because then it's like how do we as an industry ensure that this is correct and that this information is correct. So I think this is somewhere where we and this is not meant to be a negative, right? I'm not negative towards AI. I'm really excited about AI, but it's for us as a profession and a sector to catch up really, really quickly because I think we have got a lot of catching up to do. By buttered in there though, I think one of one of my worries as well is even if British standards, we talked about this the other day, if BSI turned around and said okay, here's a plug-in, you can now query all our standards with AI. What we don't want to end up is the fire sector going any further down a route where we're trying to tick boxes and saying would you know what to be to be safe? You've got to tick every single British standard compliance box. You know, I think the cool work that gets done is when somebody says look, actually here's a standard and it says this is our ideal position, but I believe the risk is really here and you're mitigating these risks so you don't need to do everything that's specified in the S, whatever it is. So I think that's a slightly dangerous route. What we want is to up-skill survey as assessors and engineers so that they're doing less donkey work and putting their application skills to use better. What's a moving from like the checklist style into an actual risk management function and evaluation of risk? Checklists are great, but it's the context that sits behind the checklist that's important. Yeah, I think big concern is when you question it and it is wrong. It's got the same confidence as when it's right and it doesn't give you any indication that it's made it up and yeah, it's seen it make up references and things before just completely out thin air that all look very believable. You can ask it by the way though, you can test it. Like, so if you are going to use it for this and by the way, I'm not endorsing it because I don't like the idea of using open LLMs for this type of work just because of the risks, but you can say, look, what is the relevant guidance and can you please provide me with the reference where I can find it and then you can go into your own QD elegance. So there are some work arounds, but it's not perfect for the industry that we operate in. and the level of risk in the industry that we operate in. - Yeah, you spoke a bit about sort of data earlier and how the data isn't there. If the data was there and we had everything gathered and I think fire industries, different to other industries and amount of data that's available, but yeah, it's much better how other industries are using it. And if we had that data available in the fire industry, if there are any ways you think, that we could be using in the future that could be quite powerful if I'd come to you first Adam. And I think, I guess there's kind of two sets of data that we're looking at here in my mind. There's going out and saying, I'm doing some kind of a report on a building where that's a fire, a success, and a strategy or whatever. What do I need to consider? Here's some information about a building. It's this tool, it's made of this, it's whatever. And it will say, well, actually, with that kind of occupancy, here's some things you need to consider or here's some more information I definitely need. And then an AI model could lead somebody down a path and go, based on all of the other fire strategies or all of the other risk assessments I've seen for buildings that are similar to this, these are the things that seem to be key topics or key risks or controls. But then there's the other side of it in terms of buildings that have been modeled and also existing fires, existing emergencies that have happened, that have occurred, that need to be analysed. Now, if we get that information from the fire service and if we can start to get that instruction data, then I think you can do some really cool stuff because actually we can start to back it up and say, well, actually point, not, not, not two percent of fires are related to this. But it would require the fire services to start to pick that information up. And ideally also, you know, near misses as well, fires that don't, the fire services aren't called to to also get picked up. So it really requires a massive amount of transparency from lots of people and in a world where we're all finger pointing at each other all the time and blaming each other, that becomes quite difficult and it also requires lots of open standards. You know, when we look at what's happened with the Building Safety Act and the Fire Safety England regulations, we can't even get, you know, each of the fire rescue services to come up with one system for everybody to enter the information into. Let alone have consistent information that wants to be gathered or captured. So I think, I think, well, to be fair, I think there's probably some things that the membership and the governing bodies could really do here. Yeah, I think there's a big desire to certainly have from a lot more information, probably available both from existing fires that happened and the fire factory port as well as from fire testing and that data bit and getting there. I think is a big challenge as lots of hurdles that need to be resolved with that. But yeah, I'd sort of ask same question to you, BigDif we did get across that, you know, do you think there's some big uses for AI in the fire sector? Yeah, absolutely. So it's a great question. I honestly don't have an easy answer because it is a challenging. Like data has always been a challenge within health and safety and fire safety. You know, it's a challenge. We're not traditionally very good at it in terms of collecting it, sanitizing it, controlling it and sharing it. And I think this is where AI can actually help us to solve the problem of data. But one of the things that I do look at with data say we forget like the numerical data, you know, in terms of like spreadsheets of numbers. Say we go to data in terms of information. Could AI help us with golden thread? Yeah, absolutely. Could it help us with KBI registrations? Yeah, absolutely. Could it help the regulator who has been, they must have been inundated with hundreds of thousands of pieces of data, hundreds of thousands of pieces of information. So actually starts to draw some correlations and look at risk rather than what we, that the elements that we perceive to be high risk may not actually be high risk because no one's ever done a deep dive into that much information. I would almost say that the submissions to the regulator recently are probably one of the biggest collations of building and safety and fire safety data and information that we've ever done as a profession as a country. So that's something that could you potentially look at utilizing AI in the future to do some analysis of that and actually give us some really great information as an industry as a profession. Yeah, I think we could. I think there's some real opportunities there. Yeah, I think it's a really good point. Is there say people are, people intuitive probability that they like to put things into never going to happen, always going to happen or 50, 50. And when you have lots of data like that, it's difficult for someone to process intuitively. And that is something that the AI and advanced algorithms are very good at. It's quite difficult for people to be objective, isn't it? As a risk assessor, there's quite a lot of weight on your shoulders. I mean, whether you're a risk assessor or whether you're doing something else, in that you're always going to want to earn on the side of caution because you don't want somebody to turn around to you five, 10 years later ago, but you said this. And without that context, without saying, well, this is why I've made this decision, it's quite difficult to justify that stuff. I mean, I think I am relatively objective. I'm quite data-ortated. And I see a lot of different risk assessments and the inconsistency between two people looking at the same building and it can be quite astonishing. And it's not because one person isn't as good as the other, it's because they viewed it from a different perspective. And when we look at, so I don't want to talk about risk-based, but we're a management platform. We've got loads of data. And one of the things that we will see is we might have four or five different consultants that have gone to a building, to a property. Might have been the building might have owned for different people over a period. It might have had slightly different use, but it will have had so many different touchpoints by various experts looking at things from different perspectives. And if you could take all of those surveys and put them into very structured data, the inconsistencies you'd find are astonishing. And sometimes it's really simple stuff like how many floors are there. Sometimes it's because those people can't count. Fine, that's just a simple mistake. Sometimes it's because it's floors above ground. Sometimes they've included the basements. Sometimes they just haven't seen something, but as soon as that data can't be trusted, or soon as one bit of data is wrong, none of the data can be trusted. So if it were one of the first things we can do with AI with these data models is to start going or wait a second, you've just chucked a new survey into this property or this building. Can we just double check these things? And this utopia that everybody has about will all the data needs to be perfectly structured, will have one form field, which says how many stories there are, doesn't work. Because that's so much better measuring it. I know at least three different ways of measuring the height of the building, and it depends on what you're doing on which method you need to use. Yeah, but it does sound like there could be some good uses for it. How, my next question is sort of how do we get that? How do we help people gather data in a way that's useful for AI? Is it different to gathering data for just a statistician who wants to see, is there a different way that AI likes data? On these things, how can we help get that data and start gathering over the next five, 10 years so that we've got that pool available to start a view feeder? Yeah, I don't think it'd be five years, to be honest. I think it's going to come very, very quickly. I think we're going to be surprised that the change of pace in terms of the implementation of AI in safety and then in to fire safety. I think one of the challenges that we've got in fire safety, if I may go there, that we don't necessarily have, like, so say in health and safety. You know, I work with an organization that specializes in AI for health and safety risk assessment. So you can upload a risk assessment and it will do an analysis of that risk assessment based against all of the available information in the system. So everything that comes from the HSE, everything that comes from lessons learned from other organizations, et cetera. One of the things that I fear about fire safety is that we don't necessarily have, like, one single one song in terms of the HSE guidance. That is the guidelines that we tend to follow. And I'll give you for an example, "Biodals." Okay? We've got a lot of conflicting fire guidance around fire doors. So if I ask AI a question about fire doors, and I say, "Well, what is the, "I don't know, the threshold gap of blar in this circumstances." It's going to go, "Well, I've come to PS8214. "You've got this." And double nine, double nine, you should check it on this frequency. And if it's health care HTML5, I'm bragging in our George, I'm trying to impress you with a little bit of my knowledge, right? It's not written down in front of me. BWS, BM Trader, Fire Safety England regulations. And this is one of the things that I fear about AI and fire safety is that we've got so much conflict that it's actually going to struggle to give you the answer. Because we can train AI in all of this information. But then it's in our industry, it's going to require someone to put that in the context of the building. a ddiodd gallan yr ia科ill. Hysigydd ei g lastlyredd renau ran echendawd13 ypneudau started dal ein ni'n bi wedi yr gallu idregu g MAC fo ei sefydnadach am sydd wedi头 Greflyn az trein o fiili unable acussions ar ysefan i de narrative rysadedde beth aeld eth considerer ac i'n ei cynig. Mae gyda Knowingletiaol a'i o pecn錦mbi i roedd yn rhafa cymu a'i'r cymdeu i'r cymdeu ymdol ymdol yma. A'i cymdeu yma'r cymdeu yma. A'i cymdeu yma'r cymdeu yma yn ddein o'r cymdeu yma. A'i cymdeu yma. A'i cymdeu yma. A'i cymdeu yma yn ddein o'r cymdeu yma. A'i cymdeu yma. A'i cymdeu yma. A'i cymdeu yma. A'i cymdeu yma. A'i cymdeu yma. A'i cymdeu yma. A'i cymdeu yma cymdeu yma. A'i cymdeu yma. A'i cymdeu yma. A'i cymdeu yma. A'i cymdeu yma. A'i cymdeu yma. and yeah you were talking to me about how, how AI, you know, it might not gonna spring upon us, it might just sort of slowly merge in and I think sometimes I sort of use it as a dirty word almost whether you were sort of just rebranded an old algorithm that's very simple and aren't using the term anymore and I wonder if we will stop using the term and it will just be integrated sort of like that or yeah. - Do you know I've got an analogy for this and if anyone listens to my podcast the most hatey tech podcast, there'll be sick and tired of this by the neck by the way because I say it all the time. But I remember when Tom Tom came out and like you'd say to someone, "I'm using Tom Tom" and they'd be like, "Why don't you just get a map out and write down the directions?" You know, and it's like, "Well, I got Tom Tom and it would always get people lost and it would be on the paper because someone are driving to a field or something." And then we very quickly moved on to the point where now we're on Google Maps and Waves and Waves doesn't just tell you which way to go but it actually predicts what the traffic is going to be like on the route at any point in the future that you decide in on a route that you don't know. So now it's this level of intelligence that's way beyond what we could have ever hoped for when Tom Tom first came out. And you know the weird thing is, I don't even know where Tom Tom went. I used to have one in my car. I don't remember binning it. I don't remember selling it. It's like they just disappeared and now we have an app. So we very quickly went from this resistance to sat nav to now we just have an app on our phone and people use it almost religiously. And I sort of feel that's going to be the same way about AI. So there will be resistance. There will be resistance change. There will be this element of catastrophisation, you know, in part with the press. But in terms of what we're using AI for, you know, in terms of like really a minuscule application in terms of what we're doing, it's going to be really powerful for our industry in our sector. But wasn't what map, what made digital mapping so effective was the data that came back from all the people using ways and Google maps so that they could predict it. And that's where Tom Tom kind of got stuck because it was a secondary, you know, it was a physical device that didn't have an internet connection, right? And it's really, it all comes back to get the data, look after the data. And it's something that I say all the time to everybody, whether they're consultants or whether they're managing properties. That information is so valuable. And I, it really frustrates me when you watch people just effectively throw it away, or just not use it properly. Yeah, you know, Tom Tom was still going in like 2021 by the way. I googled them the other day. I was like, how? Who was still using them in 2021? How, how? So I think how the experts were propping them up. I think they, they, they use their technology to help them, you know, car manufacturing things with the built-in satin house. But yeah, I think the big change there, let's say, was when the knowledge switched over and suddenly, you know, an experienced map reader was no longer faster than using these. And as you said, I'm sort of by starting to use this technology. And we were able to gather data and make it better and get to where we are now, which I think is, I think a big topic for all of this AI is, maybe starting to use it a little bit and, and gathering that data and how things get to where we are going in the future. Well, by the way, that it will get better because I've no doubt that driverless cars in 15 years' time, they will already be the technology that they will, if you do a regular route into work, it can send cars off in different directions in order to pre-proactively ease congestion prior to ever occurring based on how much traffic a road can take. And I've no doubt that that's already in that system ready to go so that when we do have driverless cars, it can direct people in different ways. So it's not just where we are now, it's like, even now with the tech that we have, it can go so much further, you know, with something that's pretty well established. Yeah, and it's a really touched on the golden thread when you were talking about how we can use it at the moment in the Fire and Strider when I come back to it. And for our listeners, golden thread, we're talking about the thread of information and the buildings being designed and preserving that throughout. So that's always available to tap into. And I just want to really see if we could provide some more specifics on for our listeners as to how that could be used and how they could dip into their information and use AI to help them there. I think it kind of comes back to the other conversations we've had earlier, which is the difficulty is, in terms of a building, is the portability of that information throughout its life cycle. So when you go and design a building, you know, if it's a big advanced building and they're using BIM, you still got three or four BIM models. You've still got a structural one. You've probably got the up interiors one. You've got an M&E one. They never quite get put together. So this idea of this perfect digital twin never quite works out. And even if it gets handed over as this perfect single model, the people that go and then go and use the building can't actually do that much with that model. They don't really care about it. They care about what do I need to do to maintain it, when this breaks, how do I fix it? You know, when something needs to be serviced, what do we need to do to it exactly? And it's that, you know, the cause and effect, the strategy, design requirements of every single thing. You know, when we're us in fire, we're looking at things going, well, you know, how does that fan work? That AOV, what does it need to do? Why is the fire alarm configured in that way? You know, with those lines of compartmentation, why is that 60 minutes? So there's somebody doesn't come along in five years time and go, look, there's an FT60 door. Well, it didn't need to be an FT60 door. It's just it was just easier to put another one in. And then that starts getting maintained as a as a fire door. But we've gone from designer, principal contractor, first owner, you know, it could be an investment company. Then you've got a freeholder. If it was a block of flats, that can then change. And then you've got leaseholders. You've got various block managers that might be managing or housing association might be managing it. And exactly the same would happen in a in a retail environment. In fact, perhaps that or a commercial environment. And potentially that building is going to change hands more often. So the difficult thing actually that I think is how do you package all of that information up so that each stakeholder, if you want to call it that can access it, whether that's an FM contractor, whether that's an alarm company, whether it's a gardener, it doesn't really matter. It's everybody needs to be able to access these models and then go well actually I've just changed that thing. It's now not which it A it's would you be or it's I've you know what I've oversized it this this vent because it was cheaper or easier or something. It doesn't need to be that performance, but it just is. And then that giving people access to that history and that's the difficult bit you need an information store. And you need 20 bits of software to go and talk to it. So to me, it's not you might use AI to query some of that data, but the difficulty is going back to that data and updating it. Yeah, I think a really interesting example because that's you know that's sort of a problem with all without AI and it's just one that gets exasperated if we try to use this data. Even more and for some time is against us today, but I just want to see if you had any any final thoughts before we wrap up if I come to you first Peter. Yeah, I think when we spoke London build last year, you know, I approached you and I said we've just got I by the way, I love the fact that the fire sector is really leading on this and saying, you know, we've got BS86 before, you know, digital management fire safety information. I was like, this is amazing. How do we now help the professionals in our industry to improve that digital literacy to be able to work out how we ensure that we meet the requirements of this digital management in fire safety. And that's something that I'm really passionate and really excited about and that's why I set the podcast and this is why I love the fact that George, you've got myself and Adam on here today to discuss this because I think as a profession, as an industry, we all now need to be curious about tech and what it can do and how it can help and start asking the right questions and the professional bodies in the industry need to start providing training, start providing education so that this doesn't feel quite so mysterious when it comes in like we know how to use the power of technology, how to harness that power in order to improve fire safety. Yeah, something that really interests me because you know, same technology in the past that were rubbish and people have ignored and then, you know, at some point they've suddenly become good and this does sort of feel feel like one of those and something, you know, keen to support and for the FBA to try and help with and get the fire sector to where it needs to be with that. Are there any any ever thoughts from from yourself, Adam? I think I probably talk too much, haven't I? You definitely nodded then by the way. Now I think I think it's great where everything's going and I think the more conversations like this that I had, the better really. I think the main thing I think we need is just more transparency, more people sharing ideas and information, fewer British standards, more open standards. Yeah, the more we'll courage that the more the more it will come. Well, thank you both very much for joining me today. It's been very interesting for me and I hope it's been interesting for our listeners as well. So thanks that they informative and thanks your time today. Thanks George. Thanks. Thank you for listening to the FPA's Assembly Point podcast. For previous episodes or more guidance and resources on reducing the risk of fire, please visit thefpa.co.uk. Don't forget to hit the subscribe button for future episodes and if there is a topic you would like to hear discussed, please get in touch.