Course Support

Testing, testing. Can you hear me? Sorry, Keith, I was a reply to a question. Can you hear me? Yeah, I can hear you. Can you hear me? A couple of minutes, some students are arriving and then we can start. And then people will look for the signing campaign. Okay. We have what, one hour? I'm sorry. How long? Two minutes. One minute. No. How long do you want me to talk for? Last year was good. So same on last year. But it's up to you. It's you're in charge. I think we have a Zoom call for four. We can do it for 15 or something. No problem. Exactly. I think I will start closing the door. Are you recording the lecture or I am recording the lecture? I'm recording it here, so I can send you a copy if you want. Yeah, okay, perfect. Yes. Okay, I think we can kind of start. Guys, everyone, welcome to this lecture. As I mentioned, the first part of the lecture, we had Dr. Keith Steele. I'll leave him to introduce himself in a second. And yeah, Keith, the floor is yours. Thank you so much for being here again. It's always a pleasure to ask you an honor and a pleasure to ask you guys. So, yeah, the floor is yours. Okay, always a pleasure to talk to you all again. Let's just commence. Sounds good, guys. Can you hear? We're all good.

OK, the talk today is about crisis communication. Sorry, Kate, can you open up the window? Because we can see the window, but also the key. OK, let me just stop that. Share. We always have IP problems. I think now it's okay if you put the presenter. Yeah, one second. Not quite sure what went wrong there. Is that better? Yes, perfect. Oh, okay. Today's talked about crisis communication. I will give you some background details and first of all, a brief introduction of myself. A fellow of the Institute of Mathematics in its application, a fellow of the Institute of Civil Protection and Emergency Management, a specialist fellow of the International Institute of Risk and Safety Management, a fellow of the Institute of Place Management, a fellow of the Higher Education Academy, and a member of the Academy of Experts. That last one's probably most important. Because for over 36 years now, I've been consulting on crowd events around the world. Small, medium and large. I think the largest is now probably the Indian festivals and mass gatherings. I lectured to the police there for now 7 years. 100 police officers at a time, maybe three or four times a year. We worked on Hajj projects. Currently engaged with 53,000 crowd just off of one part of a site in Saudi Arabia Hajj, working with Hajj projects of 3 million people. And the smallest is the 500 people that come to my local fireworks display. The difference is they actually know where I live. So for 15 years I taught at the Emergency Planning College, left there to set up a master's programme at Manchester Metropolitan University. Have run projects globally and currently involved in four expert witness cases, two in the USA, two in the UK, looking at how accidents and incidents occur. And again, crisis communication follows on and within that. And also a whole range of experience for citywide event planning. As you know, I teach at University of Leyden, also at Brady University, the Polytechnic in Milan. NCS4, we run a virtual course online, run courses in Australia. But the four at the bottom there, India, Dubai, United Kingdom and Canada, we run directly with police on crowd safety and specifically now elements of mass communication, crisis communication. So quite a lot of experience in teaching and training around the world on that. And also because as an expert, I get called in to analyze what went wrong on sites. It gives us a deeper insight as to what we need to do to incorporate that into teaching and training. So if we can teach people how to prevent accidents and incidents, we're really doing our job.

So a lot of the work we do is focusing on this thing called the safety plan. Two key elements to that, a crowd management plan and the risk assessment. And as I've just said, we run a whole range of courses online and I've written a couple of books on the subject. We do site-specific courses which are more about how stewards are meant to assess risk and report risk. They might see something on the ground, at ground level rather, and they need to have a good method of communicating that to the control room in a language that both understand. For instance, if someone stewarded one site example, the Hong Kong Jockey Club, he sees something that might be a spillage or a trip hazard. In certain instances, he needs to be able to report that clearly to the control room, but they also need to be empowered to fix it. If it's a spillage, can he lock it off? Can he block it off? Can he mop it up? He or she, for instance. So any incidents really need to be thought about of how do we deal with risks, because it's usually the small things that tend to escalate.

So what's crowd science? How, where, why and when crowds form and move. So a little bit of mathematics in terms of fill rates and flow rates, quite a bit of psychology in terms of understanding crowd behaviour under both normal and specifically emergency situations. We look at how and where crowds reach critical density. Anything more than one person per square meter, and the crowd starts to behave as a conglomerate, as a dynamical system. Anything less than that, and people are influenced more by their environment. So information, particularly in a crisis, will have different forms of receptiveness, depending on the density, the number of people that you're trying to communicate with. We did a whole range of experiments that he's involved in emergency planning college, where we took a class full of people that didn't know each other. They come from all walks of life, from all different parts of the country, sometimes from different parts of the world. And we'd put them in a room and we'd start to introduce smoke through one of the emergency exit doors. And people would, if they don't know each other, start to ask what's happening, what's going on? Is this part of what we're supposed to be? It's just a learning exercise. There's no lecturers in the room, but we're filming them in order to use that as part of our training later. And as the smoke starts to fill up, eventually people start to talk to each other and they might come to a decision, should we evacuate and leave? There's no fire alarm going. We run the similar experiment, but with a class that's actually been with each other for two or three days. And they very quickly come to a consensus because they've developed personal relationships with each other. They know who they are, they've introduced themselves, they know who the various voices of authority are. And it's very quick for a group that knows each other to come to a consensus about risk and to move away from a site. But a group that is unfamiliar with each other, it takes much longer. There's a whole range of scientific experiments being done with that. And that's really very important when we think about how do we communicate in a crisis. We look at safety limits, but design effects and queuing systems. We look at crowd density, at what point does the crowd reach crush density? And we look at flow rate. So that really is embodied within crowd science, mathematics and psychology, human behavior. But it's all really about understanding what causes crowd accidents incidents.

So here's a couple of examples. This is a situation, of course, brought with danger. Crowds are corralled these days in football stadia, and if there is any pressure on from the crowd behind. Getting them all to shout together is one method to communicate with the crowd in an emergency, but you need to understand that had you designed that system more appropriately, this was a bottleneck, a funnel, a large number of people trying to move through to a narrow entry exit point. And that's what caused the crushing. And now, of course, they have to tell people to step back. So they didn't have a communication system in place. Other types of crisis communication deals with something like we saw in Turin 2017 where a rumour, in fact it was a couple of thugs were using pepper spray to rob a woman of her handbag and then somebody shouted gas and then somebody shouted terrorist and this is what happened. The PA announcement system there was communicating about the football. This was a fan zone in Turin. And what had happened, there was a mugging. Somebody used pepper spray in order to subdue the woman. She was okay, but now we end up with 1,527 injured. Because there was no method to cut across that and to communicate with the crowd, which you need to do instantly. Astroworld is another particular incident. The crowd are screaming because they're in distress. There's no way in which the organizers failed in order to understand what those risks were and had no method of communicating with the crowd once that pressure point started to build up. And then finally Seoul, South Korea. And this is an incident, it was a Halloween event. Thousands of people turned up. It is overwhelming the space that they had available. And we have on the right hand side of the screen there a progressive crowd collapse. So again, no method of communicating, no method of telling people to step back, no PA system in place, nothing to communicate the crisis other than trying to pull the bodies out. And you have about 6 minutes in that environment before people suffer from restrictive or concentrated asphyxia.

So why is this important? Well, if we understand the causes of disasters, we can improve safety. We understand our liabilities. And what we've done, our team here has been studying crowd disasters now since 1989 when I first got involved 36 years ago or so. But we've got records now to go back a couple of hundred years. And what we identified was a classification of what were the primary underlying causes of incidents. And quite a few of them, well, most of them, in fact, are related to design problems. A couple are related to just bad management. And then there are more that are related to just bad information in the environment. So for instance, it suddenly starts to rain. That's an environmental change. People suddenly run for cover, reasonably foreseeable, predictable and preventable. Unless you've got the space, you're going to end up with some form of crushing. So predominantly, the thing that we call the DNA of accidents are very similar underlying causes. And it was Lord Justice Taylor at the end of the Hillsborough report back in 1989, where 97 people died and several hundred were injured, saying it's not enough to aim at the minimum measure necessary for safety. That has been at best the approach in the past and all too often not even that standard's been achieved. What's required is a vision and imagination to achieve a new ethos. Now there were over 30 years of litigation involved in that one. Eventually pointing to the safety manager. And again, the problem was overcrowding at the outside of the environment, no method to communicate with the crowd, just tell them to step back. Bad decision to open the gates to allow people through was probably the only way to alleviate that pressure, but then the failure to have no communication system for the crisis. And then another failure where now the crowds are inside the stadium and the first thing we want to do is get down to the pitch side because it was a match that had already started. So failing to understand crowd behaviour in that environment.

But it's not just large crowds that have problems. This is a site, you imagine the size of a disabled parking space. We have some 600 or so kids queuing in that area, and ends up with three of them dying of compressive asphyxia. The problem there is that nobody wanted to give up their space in the queue. So failing to understand psychology, but there's no barrier system, there's no crush point in that. It's just a slow buildup of pressure. An accident is an incident which no one could have reasonably foreseen. No one's at fault. But negligence is failing to do something that could have been reasonably foreseen by others. And that's not necessarily a layperson. That's an expert or a trained crowd manager. So incidents occur and once they occur, we need to think about now about how do we communicate. And you need to understand human behaviour in that aspect. And as Hippocrates said, the chief virtue of language is clarity and nothing detracts more than the use of unfamiliar words. If you're in an environment and you're expecting a set of conditions around you and something happens that's out of the ordinary, people don't know how to respond to it.

Here's an example. This was a bomber in London. The police are trying to evacuate and he comes into a cafe. Rather than a rapid evacuation, you could see the lady in the foreground there, gathering her belongings, tidying up her area, making sure that she's got her baggage with her. And that's what happens in an emergency situation. People don't know how to respond, so they revert to what would be normal behaviour.

Now there are ways of analysing this and one of the models that we develop is to look at crowd communication. This is for Saudi Arabia, where we're looking at language issues and how best to communicate with African, European, South Asians, Arabs, Iranians, for instance. So you list the different demographics down the site. You list across the top the different methods you have of communicating with them, and you color code it. What's effective, what's very effective, what's not effective at all. Now, that allows you to then shuffle the rows and columns to see which is the best method to communicate with that crowd in an emergency. But it's not just crowds we need to talk to. We need to think about interprofessional communication as well. So facilitating links between key functions, promoters, production, security, stewarding, emergency services, voluntary organizations, and also facilitating links with critical locations. Now, unfortunately, there's a lot of jargon involved in this. A PA announcement that a pit team might call things rendezvous points. They might call them coded descriptions. It's all very difficult to communicate with people that might not know what those colloquial terms are. So very important to understand that clarity in communication is the most important thing. And there's nothing worse than the use of jargon or unfamiliar words.

So how do we understand how information might flow between agencies in a crisis? Because you can't risk a failure of communication during an incident. And you evaluate this using a matrix. Across the top, you have all of the different agencies involved in the site, event operating control, EOC, police, first aid, security, ushers, and so on. And then down the side, the same list. But how does EOC communicate with the EOC? How does EOC communicate with the police? And conversely, how do the police communicate with the EOC? And you see that it's not entirely symmetrical. Sometimes communication is best one direction. So ushers, for instance, trying to get in contact with City Health, they have to go through another agency in order to communicate. And this highlighted that the crowd, the bottom row, is crowd communicating. And you see the red one there with the police. There was no direct method for the crowd to communicate with the police officers on site. Melbourne Cricket Ground highlighted this to us. They said we've got an essential terrorist incident, what's the best method of reducing that communication in a crisis loop? And all of the signage said, if you see something, say something, and then a central telephone number to call. That central telephone number is in the middle of Melbourne. They are then calling Melbourne and saying, I'm standing here by the big blue door and I see a suspect package. And now there's a whole range of communication issues because they have to identify where's that big blue door, what's that type of package. They then take that information and dispatch it back to the site. So we've now got Chinese whispers that can be involved in that. Has that message been accurately recorded and sent through?

Now, if they were able to talk to the police on the site, then everybody knows what the local door colours are and where the areas are, and it's a much easier way of communicating. So when we highlighted that to Melbourne, we told them, you're using something like What3Words, which is a free piece of software. What3Words is a three-metre grid reference for everywhere in the world. So within a three metre grid, we can map any place in the world with this particular system. And now I can tell you exactly where that big blue door is, because I've now got a grid reference. But even that, in terms of an emergency, if I misspell a word, I could find myself in a completely different location. So again, communication in an emergency being vital.

So another tool that we were looking at, again, using this matrix approach, one of my students said we can look at site operating conditions and we took Notting Hill Carnival as the example. Time across the top, down the side different zones and in the boxes, not only the color to indicate the risk, but also an icon to demonstrate what's happening. So if we have a look at something like North Zone, we can see there's food and beverage, there's circulation, amber meaning that it's something we need to monitor. There's a performance, high density crowds, drinking and usually fighting. Now, in the control room, they've got a map, an idea of what's likely to come up. And hence, if we've got a situation, they can communicate that much more clearly, that we've got a north zone, it's 9 o'clock, we've got typical fighting with alcohol involved, and everybody was anticipating that ahead of the incident.

Another approach that we'd looked at was incident location risk and response. In this particular example, we've got different threats across the top, different locations down the side, because a suspect package in front of stage has got a far greater crowd impact than it would be at the main entrance during the middle of the performance where there's very few people around it at that time. But if there's a suspect package down in the front of stage area or around the merchandising or close to the toilet, there's going to be more people there. So a different level of risk. So this again gave us the ability to look at how severe the risk might be in order to then determine what type of communication tool we need to put into play.

Now, most of the work that we do is in planning. Anticipate, prepare, risk analysis, look at all the proactive things that you can do. When you've got time to think and you've got time to plan, it's relatively low cost. When an incident occurs, you've got a very short time frame. Because a CCTV camera, if you're not monitoring at that time, you then need to read, understand, respond to that, and then communicate with the crowd. And of course, once we've got the crowd out of the way, we're into emergency management, where it suddenly starts to become extremely time constrained, rapid responses, deployment of all sorts of materials.

Now, what I want to focus on now is how we communicate with the crowd in an emergency, because there's a lot of mixed messaging these days. When we were taught at school, in case of a fire, don't run or push, you line up and move out of the building in a very calm, collected manner. That was a message for fire. It still is the message for fire. Nearly every kid that goes to school gets trained on fire responses. So that's conditioned into our behaviour. If it's a fire, we walk in an orderly manner to the nearest exit. But if it's a weapons attack, now the phrase is run, hide, tell. And in the States, it's run, hide or fight in the last resort. But this is a conflict to our emergency training. So what happens if we've got an emergency that's not a fire? We now need to install a different type of behavior. We need to communicate with the crowd to say, no, don't walk, this time you need to run. So how do we get that information across?

Now here's an incident just in 2005, a couple of months after the July the 7th bombing in London. We have Notting Hill Carnival and somebody let off a firework. Bang goes, the crowd moves out of that area, the cameras all focus on it, police en masse move into that area, probably the wrong thing to do because if you've got a primary or secondary device, you've now got all of your emergency responders in that section. And it was a firecracker, it was a false alarm, but look what happened. That crowd then moved into an area and became extremely high density, high risk environment.

So you need to think about in terms of an emergency, there's a threat and there are two options. Do we move the people? And if so, how? Or do we keep them in place? Do we need to shelter in place? And if it's inside a building, we need to think about what the threat was. Could it be a contaminant, for instance? In a fire scenario, the fire alarms go off in a building and everybody leaves by the emergency exits. And most of the emergency exits in buildings, particularly the fire exits, have got positive pressure. They pump the air conditioning into that environment to keep smoke out. But if it's a contaminant, if somebody's opened an envelope and all this white powder came out of it, that's now been sucked straight into the air conditioning system. And now we've got the highest concentration of that contaminant in the emergency exit. So that's one of the decisions that you need to be making. And then where is your place of safety? How do you define that? And all of that sits within an information bubble. How do we communicate that to the crowd?

So what are the challenges with public communication? It's all about should information be given? If so, how much should it be given? By what means will it be given? And how is it best expressed? And to highlight that, here's a situation in Nathan Square a couple of years ago. This is in the Toronto Raptors victory area. And in the centre of the screen, we've got direct PA communications with the crowd. Down to the bottom end of the screen, they've got no communication with the crowd. It's outside the range of the PA system. And there's an active shooter, but very quickly he was brought under control. So there we see the situation where the crowd are evacuating. And yet nothing's happening in that top sector. No crowd movement because they're being talked to. But the PA system doesn't reach as far as the crowd at the bottom. And the television celebrity on stage had the presence of mind to be cool, calm and collected. He had an earpiece, and he was told the shooter has been brought under control, there was no threat to the crowd. So he's able to now keep them calm, talking to them constantly. But not everybody has that skill set or that presence of mind. And that's really the most important thing about crisis communication is training individuals to be able to respond in crisis, to communicate clearly.

At least that was the problem until this piece of equipment came along. It's a tool called Audiban. And it uses AI for mass communications. And the beauty about it is that I've got a whole range of pre-recorded statements in any language already to go. And it's in use in thousands of schools and colleges as a mass communication system. But I've been involved with them now for over a year. And I'm thinking in an emergency, one of the key problems is how do we communicate with the crowd? Their system already does all of this. Now, we demonstrated this to the police trainers in the UK on October the 8th last year. And they were saying, two of the big problems we have in a mass gathering on the outside, something like Notting Hill Carnival, there are these big disco panels and there are dozens of those. And normally we have to have a police officer on each of these sound stages in order to instruct the DJ to switch off the sound system in order that we can start up a secondary communication device, a megaphone or a set of police speakers somewhere else. But you've only got a few seconds between you switching off a sound stage at a major event like that before the crowd starts to get quite angry. And what we demonstrated to them is the device from Audiban is basically a computer about the size of a shoebox. And it plugs into these systems. You don't need to go along with your own PA. You can actually plug it into any existing PA system. And if it's got volume control, then we can control the PA system. So it means that any police officer with the app is able to communicate and control the DJ's panel and communicate using AI. Now, what's important to understand about that is that if I just repeat the same message over and over again, it's kind of like a fire alarm going. People don't know how to respond to it. But AI can deliver that message in a cool, calm, collected voice in the appropriate language and vary the tone so that it definitely sounds more like a human being rather than a recorded message. And as we said to the police trainers, instead of you having to go along with PA systems and train your operators, the commander can issue a set of commands down to the computer. You walk along with it at the start of the event, you plug it in and at the end of the event you unplug it, you take it back, put it in your vehicle and then you're off to the next site. But the commander can now issue a set of zoned commands to different messages at different parts of the site. Everyone that's got a NUC can have a different message in it. The device is called a NUC. It's a computer.

Now we've been involved in modeling and analyzing crowd crisis communication systems since right about 2003 when we were asked to look at Canary Wharf. That's a major financial district in the UK. And they had an issue where they would have a security threat, an alert. The police would then issue instructions to the Canary Wharf Management Limited, who would then issue instructions down to the buildings, who would then make a decision as to whether or not they would evacuate, at which point they would then phone the police and say, what do we do next? And it ended up in this loop of everybody asking everybody else what we should be doing, rather than a top-down evacuation strategy. Now, 2003, which is a couple of years after the 9-11 bombings, the Central Financial District in the UK had this issue. We're getting two credible threats per week. How do they do a top-down communication with their tenants? And what we proposed for them is this particular tool called CORUS. You click on the location of the threat, you click on your building, and it shows you which route to take. And we then couple that with a training program that actually has a video clip of them leaving the building. We then train the fire operators, the fire marshals, instead of pushing people out of a building to lead them out of a building.

Now, why is that so important? Well, if I were to say to you now, and I often do this at lectures when I'm in the rooms, I'll put a slide up that says, please leave this room immediately. And everybody looks at each other and they look at the sign and they think, is this part of the lecture? What's next? And I'll stop and I'll take a sip of water and I'll look at them and I'll point to the sign. And usually about two or three minutes later, somebody will stand up and leave. Now, it's not about the information. It's about how it was delivered. Everybody can read it and say, please leave this room immediately. Everybody understands that. But the context is out of place. There's no additional threat. There's no fire alarm going. There's no smoke. There's a lecturer standing there not doing anything. And that's the communication problem. And that's where Audiban really does have its advantages, because the tone of delivery, the additional information such as a low-level fire alarm, bell, buzzer, sound, anything that would allow them to get two or three bits of confirmation information rather than one contradictory piece of information, is very important to understand human psychology in emergency. So getting them to start to move, as Jonathan Simon proved, is about two-thirds of the total evacuation time. Getting people to start to move, getting them to respond. And that's all about the type of messaging that we deliver, how do we deliver it, and how do we tell people which way to go?

For instance, there might be multiple sets of threats, and I need to be able to analyze that, process it, map it, and then give people appropriate directions. And artificial intelligence now can facilitate that, whereas we didn't have that facility before. Now, I demonstrated this to my class for the master's program and issued them a challenge. If you can come up with something better, please, let's look at it. And somebody did, which is brilliant. You know, you teach something and it comes back tenfold. And they said, we can do it all in Excel. So using it for decision support. In other words, what they do is they take the site, and I think this is the Dublin Marathon. And they created a set of cordons which were scaled to the site. So 100, 200, 300, and 400 metre radius. And now you can just drag and drop. So there's your site. You've got a threat. You know what the location of that threat is. You move your cordon into the right location. And subsequently, it also appears on a network map. And now I know which routes to push people out from the site. And again, the important thing here is that if I'm telling you leave this area immediately, the first thing that pops into your mind is why, what's happening, what's going on. But if I had to say there is a threat, follow me, it will facilitate a much faster initial response. So training the fire marshals to lead people away from the threat, communicate that using artificial intelligence, mapping the site, analyzing the site, really reduces that crisis communication situation.

Now there are only five strategic responses. If we've got some form of threat, there's only five things we can do with the crowd. The first is total evacuation, which is usually something related to fire, in which case all the exits are available and it's just leave the area as quickly as possible. All exits are viable. I don't need to do anything more than have an alert saying leave this area or have your marshals coming in and saying leave by this exit, leave by that exit. The next one is directed evacuation. Now say we've got a suspect package by the big blue door. We did a demonstration of this actually in the USA. We were about 400 people in the room. What I always do when I'm lecturing in class is that I go along and I check with the facility, watch your emergency evacuation. And I think in the last 35 years or so, I've only ever had one response that satisfies what they would do if there was a suspect package. Everyone else was just the fire alarm. Switch the fire alarm on and they'll all leave in an orderly manner. And I said, but what if it's a suspect package? What if it's an armed intruder? What if it's somebody behaving with malicious intent? Then it was just fire alarm. I said, well, it doesn't cut it. So whenever I go into a building, first thing I do is ask what your process procedures. The second thing I do is then check to make sure that those emergency exits are clear. And in this particular instance, because of the security, they'd actually locked all the fire exits. And in fact, the room we were in had only one exit point. The other one was full of furniture. The emergency exit was not only locked, it had a chain over it. So I filmed this. And I showed them all of the exits, the options that they had available. In the event of emergency, there's only one way out. And if that's where the package was, if that's where the threat was, everyone was trapped. So thinking about directed evacuation and how do I communicate that, how do I put that into some form of analysing the site, determining which is the most appropriate message, and then messaging different zones with different messages.

Another evacuation strategy might be a phased response. I need to evacuate this floor by floor or area by area. For instance, if there's a fire in a building, at a high-rise building, usually it's the floor of the threat, two floors above and two floors below will evacuate and everyone else shelters in place. So that's a phased evacuation when I'm not evacuating the entire site. Stadia will often have a method of phased evacuation depending on where the threat is. So area by area, section by section. And in Salisbury, when they had the nerve agent, everybody had to stay in place. They had to be decontaminated before they could leave. So again, it all comes down to communicating in a crisis. Invacuation, in Austin, Texas, when they had an active shooter on the streets, they need to get people out of the line of sight, move people away from that, and then they protect people in the buildings. So they now triangulate quickly where the shooter might be, move people out of that line of sight, and then protect them away from it. So invacuation is another strategic response. And then finally, stay put, lockdown. We've got an external threat, and it's safer to keep people where they are. So it could be a security alert, it could be suspicious activity, it could be a CBRN type attack, or an active shooter, or any form of malicious intent. But then it's not just get them away from the threat, it's also keeping communicating, when is it all clear, what's happening, where is the location of the threat, other evacuation options.

So again, this is where scenario analysis comes into play. If we can analyze the site and look at where the different threats might be and what our different responses are, we can work out a crisis communication technique. We can communicate with the crowd. So it's a question of methodically and rigorously working through your options and then considering what the risk matrix looks like. Down the side are different types of threats. Along the top are different strategies for evacuation. And if it's fire, well, okay, the fire alarm suffices. If it's a directed evacuation, say a suspect package, then I need to consider all the locations a threat might be, all of the evacuation options. And again, this is where AI is extremely powerful at walking through all of those options. We've been developing and designing these type of systems. I've even got tools that allow me to import a building plan and work that out very, very quickly. What are my strategic options? Now, consider a building that might have four exits. I could have any one of those four could be out of order, out of commission or not viable in terms of an emergency. I could have two out of any of the four. I could have three out of any of the four that's not viable. And then finally, I might need to lock people in place because the threat's external and we're safer keeping them where they are. So understanding what your threat scenario matrix is, that's what this tool is most useful for. Understanding what kind of communication we need to put into play, that's what this tool is used for.

Now, this is not new. We've been doing this for decades. This was the Sydney Olympics. It took me about 14 days to build the model for this. I knew the game schedule, I knew the security and screening process, I knew what the transport capacity was, I knew how long it would take to get from Sydney to the site, and therefore I was able to instruct the crowd, give yourself at least 4 hours from your house to your seat. That includes travel time, queuing and congestion, security screening, entry and access points, site location, walking to your various area of the site. Now this site you could only really get to with trains. The train had a capacity of some 30,000 an hour. There were buses, there was also a water taxi. Lower capacity than the trains, but 30,000 an hour meant that it would take me two, three hours to fill the main stadium, which had 110,000 capacity. And there are events going on all over the place. Round about 11 to 14 million people, round about 500,000 to 600,000 a day. So hours and hours of transport capacity. So what we did is we built the model, showing them what we're expecting. And there it is in the center of the control room. That's the model. Every hour on the hour, it would flip to show you what we were expecting. They knew what the train schedule was because that's on that screen. They also had a weather radar so that it could determine when any weather fronts were coming across. They could look at any monitor where they'd see congestion, and they could pull it up on the main screen. But because we'd also designed the site, it meant that the crowd management plan, the CMP, was simple: hold and release. Whenever there was congestion, hold the crowd back. And it all ran extremely smoothly because they had a model. They knew what they were expecting and therefore anything that was out of the ordinary, any crisis that might have been developed, whether it's crowd congestion or otherwise, was very easy to identify and act upon.

Now, it doesn't require a huge amount of sophistication in computer models. We've got people around the world. This is Happy Valley Stadium in Hong Kong. And they can see what the crowd density looks like using green, amber and red indicators. And when it reaches a certain level with any part of the ground, they can shut gates, they can divert crowds. And even a site operator with only 30,000 people on it, when density reaches a concerning level, he needs to shut the gates. So again, a very simple method of understanding just how we can assess the risk, determine if there's a crisis developing and fix it.

This is one from Edinburgh City Council. They have a number of major events running through Edinburgh and one area was considered extremely high risk. And what they would do was be monitoring that very closely. So we put the graphics up beside it and they said, actually, no, it's only three people per square metre. It's only about 500 people. It's not a high risk because the duration is only a few minutes. So they then downgraded the risks once they'd had an identifier. Now the missing element on that was being able to directly communicate with the crowd. And again, this is where tools that we've developed, the Crowd Risk Analysis Toolkit, which consists of the Dermice model, crowd dynamics, ramp analysis, risk and congestion mapping, and decision support matrices, come in. Once you have all of those tools in the control room, designing and building complex sites, such as the work that we did for the Taylor Swift egress at Murrayfield, becomes manageable. They have a stadium very close to a railway infrastructure. And we needed a queuing system. We needed information to be able to communicate with that crowd, to tell them to go to the right place at the right time. And they have existing PAs, but that's where the Audiban NUC comes in. I can just integrate it with an existing system. It's a plug and play communication device. So again, for crisis communication, all of this needs to be thought through beforehand using matrix analysis.

That's the app which sits in your phone and provided you've got security coding, it will get you to a particular NUC computer. Then depending on what language you want to use, what message you want to use, whether you've got pre-programmed messages, whether you want to speak into your telephone and have that communicate in a different language. So for instance, at Murrayfield, when they're playing an overseas game for say Italy against Scotland, they also want to be able to talk and communicate in Italian. And a police officer who's not trained in Italian can do that because AI can do the translation for you. So again, it can all be pre-recorded information, different zones, different information simultaneously. So imagine you've got an incident where you have to send out seven different bits of information at the same time in a cool, calm and collected manner in multiple languages. That just wasn't feasible before these tools became available. So now with artificial intelligence, the ability to communicate in a crisis is vast. A huge amount of information in a very short space of time.

Now science isn't finished until we communicate it and our job here is to get an audience to understand what we're doing in terms of crowd science. There are now a lot of tools available. We have got one that allows us to take a crowd, manipulate it in real time, put it into an environment and say, this looks like a high risk. And I've got the ability to put in a 3D object there that might be a threat in order to test and assess our learners as to whether or not that's a risk. We run a course on this, offering Level 3 or Level 5 qualifications. We run for the police the Public Event Commanders Course, mandatory training for all UK public event commanders. And I run train the trainer programmes on this. In 2018, we had 59 police trainers, training every single police commander to understand risk. Crisis communication is now the latest module that we've added into it. Mandatory training, updated in 2024 to include AI in mass communications. And subsequently, we're doing the same around the world with our police trainers, in the UK, Australia, Canada, including Ottawa, Calgary, Toronto, Royal Canadian Mounted Police, Parliament Protective Services, crossing Hollywood in Belgium, India and Saudi Arabia and the USA. So we've got these tools to allow us to put crowd science around the world. AI really gives us this huge advantage. We've got smarter insights, safer choices. So that's in effect what I wanted to talk about. And now we've got about 10, 15 minutes or so for any questions and answers. So Andrea, I'll pass it back to you if you want to filter any questions through.

Thank you so much, Keith. I think we can give an applause to Keith. Thank you. I think, Keith, you touch upon a lot of concepts we have seen with structural risk communication, the importance of understanding people's psychology. And I think it's also a type of crisis communication or risk communication by authorities, from the authorities in a very specific situation. Any questions?

I had a question about the models you were using and the simulations you were running. What type of models you used for simulation and the programs you ran them in?

So you were asking about models and simulations. What kind of tools do we use? We developed about 20 years ago a spatial analysis technique that uses basically a CAD plan and I create from that a lithograph, a black area where people can walk. And then I analyze the space. We developed an algorithm that allows us to look at the geometry of what people do. They'll look for the straightest line, the shortest path, and then what people do in a crowd, they try to avoid banging into each other predominantly. And if you couple those two things together and look at all of the exits and then superimpose that on top, you've got a map of what's called space utilization. Now that's a very powerful technique in terms of architecture because it gives us the optimal design and I can iterate between how people would use it and what I want. Now we do this on exhibition sites. The key areas are entry and exit points, toilets, merchandising and where the key attractions are. If you map that and run an analysis tool across all of it, what you're left with is those areas that are underutilized and therefore economies can be gained from it because an underutilized space you might want to charge less, an overutilized space you might want to redesign to prevent congestion. So it's a rather unique tool. It's on the website. If you look at gkstill.com and under spatial analysis tools, there's a better description of it and some graphics there. We actually run a sort of a 15-minute course on how to use that tool, and we're currently upgrading it to be web-based to make it more accessible.

I have also developed agent-based models, where I can get little artificial intelligence creatures running around a site doing things. They become quite complex and quite difficult to understand what's called the computational tractability. How many options have you got available? What happens if they're all feeling in a bad mood? What happens if it's raining? And all of those psychological elements give you a much greater set of things you need to test and you run out of time to do that. So models and simulations can be extremely useful, but there's a huge distinction between a model which distills down to the fundamentals of a site and a simulation which tries to model everything. And if you've ever worked with computer simulations of crowds, you'll realize it could be a lot of fun, but what does it tell you in the end of it? I would always model dumb people doing dumb things, because that would tell me how the space would be used. If you model smart people doing smart things, it never tells you how things might fail. So a lot of experience in models and simulations, but we tend more towards a modeling technique. What are the fundamental analysis of that environment? What do I need to test? And the more you define the problem, the easier it is to apply the right modeling tool as opposed to jumping straight in with a crowd simulation and then not being able to understand the output. Does that help answer your question? Yeah, thank you.

Any other questions? I have a question. You mentioned a lot of times the use of AI, and I think AI is kind of changing a bit the way you work with crowd behavior. Can you just elaborate on how you're using it? How do you see the future of crowd management with AI? How much AI can help you in managing crowd or crowd crash?

In terms of communication, it's available now. It's extremely useful. When I first saw it, I was blown away from the perspective of this is the sort of problem that we've been facing for 35 years. How do we best communicate with the crowd? Things like a fire bell, well, you've all experienced this. A fire alarm goes off or a car alarm goes off and you ignore it. Because that's part of everyday life. We hear these things all the time. We were running a lecture demonstrating this, and the fire alarm went and everybody thought, oh, it must be a false alarm, it must be a test, and ignored it, until the organizer of the event came in and said you've got to leave. And everybody gathered up their belongings and left by the nearest exit. No idea what the threat was, no idea where the threat was. So this is just the reliance on old technology like fire alarms.

When you start to couple it with AI, we can get a lot more intelligence gathering of the information, process it a lot faster, and then send out the most appropriate message, like everybody needs to leave by the south exit. Everybody needs to go to zone 3. Now, rather than have to rely on that information going to a human being, who then needs to have the presence of mind and the accessibility of being able to talk through a PA system, and then to deliver the message in a clear and concise manner, bearing in mind that individual might also be under threat, I can automate it all on the app. So I think about mapping the site, analyzing the site, using the various tools that we've got to analyze the threat matrix, and now in the event of an emergency, a human being again needs to make that decision. I'll need to force an evacuation, here's my threat, and then the computer can do the rest. The artificial intelligence can communicate the right message to the right zone in the right tone of voice simultaneously. That's not possible using human beings and microphones. I might not be able to get to the microphone, but the AI system now allows us to do that communication from a mobile app. And because they use Starlink to communicate, you've always got access to it. Some of these big events, you don't have a GPS network signal when you can't access the Wi-Fi. But using Starlink, direct communication from the computer, from your mobile app to the computer for the processing to occur. And even looking at the redundancy of having localized mapping and zoning in those spaces, that's where AI suddenly starts to become extremely useful and applicable. We cloned my voice the other week and I was having a conversation with myself about crowd science, pros and cons with it, using AI. It was weird and creepy and fun at the same time. But this is where the AI tools now come seriously into play, is the ability to take and process options and to look at what your alternatives are. But again, it requires you to sit down and have some kind of tool of mapping and modeling the site in order to put the right kind of communication strategy in play. So whether it's any one of those five strategic responses, AI can now deliver a clear, concise, humanized message simultaneously to multiple zones that might all be different. That hasn't been possible before. And that's where AI really does cut in and help facilitate quicker, faster, more efficient and clearer evacuation strategies.

How likely do you think it is that people would listen to the AI and actually perform what the AI asks them to do, and what potential pitfalls do you see with the implementation of AI in this sense?

Very good question, which is why we were looking at the DJ, the television announcer. It's coming across the same PA system in that voice. You can't tell if it was me speaking or if it was the computer synthesizing my voice. I have for the last two years or so been creating vocalization using AI Studio and Synth V. These are voice clones, computer generalized voices. My wife's a trained musician and we've now got the choir performing a couple of tracks. They can't tell that it's not a real performer. It's that good now that I can communicate in my voice. So I'm the DJ at a particular event. People are used to hearing that voice. And now AI is speaking with that DJ's voice and delivering an emergency evacuation message. You don't notice the transition, but it's coming across as a voice, an authority figure or an empathy figure. Again, this is all about how do we understand crisis communication? How do we program the system to deliver it in that tone of voice? But you can't tell whether that was a computer or whether it was a real person speaking it. And that's where AI has now crossed the threshold where we can't tell the difference anymore.

So you talked a lot about crowd management in concerts or public spaces. Do you also work in a situation where the crowd might be hostile towards the authorities, such as a demonstration? And if yes, do you adapt your approach?

Yes, that's a very good question. That's what the police were asking, is how would we best utilise this in a hostile environment? And this is where the advantage comes. You have a police officer stepping into a crowd, becomes a target. You have a PA system talking with a voice of authority or even a local celebrity's voice. You're at a football match, violence is starting to come out and you're talking with the voice of one of their players telling people to calm down. It's now a mechanical device. I don't take the same aggression against that as I would have against a physical human being, a police officer in a uniform there. They thought this was a huge advantage because the device itself, whether it's using these big speaker units or through the existing PAs, it's the tone of the voice that's delivering the message, the level of authority with it, but it's no longer putting an officer out into the crowd to communicate. It's utilizing the AI system, but now it's mechanical, it's physical, rather than being a human being there. So that's one of the advantages they thought were that people are less likely to attack a loudspeaker than they are to take it out on a police officer that might be stepping into a hostile or threatening situation. And again, the importance here is how do we deliver it in the right tone? Is this an authority figure tone? Is this an empathy figure tone? Look at some of the work that the psychologists John Durie, Chris Cocking, Clifford Stott have done on analysing crowd behaviour and best messaging and responses to authority figures. They understand what the problem is and understand what the communication needs to be. But now we've got the technology to be able to deliver it.

Have you ever encountered a situation where AI could not be utilized due to maybe some technical difficulties, for example some kind of arena was too technologically outdated for AI to be effective, and how do you then proceed?

Yeah, very good point. Obviously, it is a system, and you need to think about resilience and backup to it. But what we are talking about is that so long as I've got a loudspeaker system, and nearly all of these places must have it by law, a method to communicate with the crowd, then I'm plugging it into an existing PA. The compatibility there is, does it have a USB-C cable? Does it have a 3.5mm TRS jack? So long as I've got the ability to plug it in, I can now utilize existing PA speakers. But if you do have complete failure, for instance, you've lost power on the site, then again, you can operate it through mobile units. The existing units that the Audiban team have are all battery powered, will last four days in the field without losing power, and have got solar power attachments to them if you want to use them remotely. So they've considered the resilience of their systems and their equipment very carefully. But in all instances, if that fails, then you still have the strategy of mapping out the site and the ability to communicate in whatever means necessary. So I would never say that this replaces a human being. A human being needs to make that decision. So you need to have somebody in the loop. But what it means is that I can now make multiple simultaneous decisions and then start the communication process while I'm dealing with the incident.

So you've got to consider the pros and the cons, the benefits, the technology, the potential failures. All of that should be wrapped up in your risk assessment. Is this the right kind of technology we should be using in this environment or is it better suited for other environments? If we take something like deploying it to a street riot, that might be more technically challenging than, say, utilizing it in an existing stadium or arena. But in all instances, I consider what happens if things fail. If we can't get the speakers working, which is why we run tests beforehand, we run the test again on the day, make sure the technology works before you start going live. So again, everything has got its potential risk. That's part of your risk assessment process.

Right, again, I think we can wrap up. Keith, thank you so much again for being here. Can we give another round of applause to Keith? Thank you for listening. Keith, there are a lot of nice training options there. I'm actually planning to take one of your courses later when I have time. It's a great area. There are a lot of events that now require crowd management. It's a quite important aspect. Keith, thank you so much for everything. Thank you. And I'll send you a copy of the recording. Thank you, please. Bye.