037 - Human behavior misconceptions that lead to (mis)modelling with Enrico Ronchi

Hello, and welcome to fire science show session 37. Great to have you here again. today's guest is professor Enrico Ronchi from Lund University. And you may remember him from episode 16, where I've tapped into his mind and professors. Ruggiero Lovreglio's mind to understand what the future has for evacuation modeling. I've invited Enrico after a passionate, discussion on Twitter about, protecting evacuation modeling field from bad science and how badly some models are use. And. How it creates a parallel literature scheme. And it's quite funny that, just today in the morning, There was a really nice, Webinar by human behavior group of IFSS Where the terminology used in evacuation science was discussed. And a lot of this is so relatable to my today's discussion to say, In a way, the webinar in the morning has shaped the discussion in the afternoon. And that makes me very happy because I think it gave the talk another dimension. And I hope you like how we've ventured with Enrico, through some concepts that are common in evacuation modeling and maybe they should not. And some concepts that are not common and maybe they should be a little more so. , yeah. whether if you are engineer or researcher, I think there's a lot of useful things to take away from this episode. I hope you enjoy. So let's not prolong this. And let's jump into the episode. Hello today I'm here with professor Enrico Ronchi from Lund university. Hello, Enrico. Great to have you. back in the show,

Hi. Hi Wojciech. Very nice to be back,

Please, don't panic. Uh, yes, it's, just 15 seconds into the episode and we've already used a forbidden word and It's an important part of this episode. Just earlier this morning. I, saw a great webinar, in, um, human behavior in fire work group of IFSS with, , Anne Templeton and Milad Haghani who did, discuss the P word in depth and. You were moderator of that. It was great. Will it be available? online by the way?

we have actually YouTube channels. So we record all our webinars and, yeah, so I will share around the LinkedIn subscribe to our YouTube channel.

That's so cool so two for a price of one, that you ended up listening to these episodes, go and immediately learn something about, how bad. we Use terminology in evacuation, modeling, but it has, consequences far beyond the, Hm. Cleanliness of the language, it's not just using the terminology, because the vocabulary, says so so it has a deeper meaning that leads to misconceptions , and issues with that. do you think it's really. such an important thing to have workshops on that and, really battle.

I think in our subfield of evacuation modeling and human behavior fire, really flagged this as a very big problem because noticed that there is, uh, an even increasing, literature modeling papers. They try to model something that has very little link to the real world. And, If we look at the studies for instance, that both Anne have been doing into the use of a terminology, this seems to have spread more and more following up, these seminal papers, which make an incorrect use Of those words. So for instance, there is a, let's say one of the key papers in our field is the paper from, Helbing and some of his colleagues published in Nature in the middle nineties, which presents the social force model. Probably that's one of the most common modeling approaches in a evacuation simulations. it talks about modeling panic And the fingers that actually the model is a beautiful model mathematically, because it's very simple and it's very flexible in representing people movement, it is actually not modeling panic. And what happened is that people literally the words as they are in the, paper. And they start making modeling efforts, which let's say derail from theoriginal purpose of that paper, which was actually about the modeling people movement. And we see a lot of modeling efforts, which, , don't make too much physical sense. And, and it's a really a pity because, as community, we should try to intervene to do something into stopping this waste of energies. I will argue the property. This is not

Yes.

as we think. Did we have that may become from slightly different domains from ours, that approach our domain of, fire and evacuation modeling and basically, they are very good at maths. Probably. They're very good into, coding and implementing models, but they probably did not spend enough time reading the literature. Of the what they're meant to model, but they just look at the modeling literature and these created this very, huge bulk of modeling literature with.

Jeez.

Unfortunately a very poor, for the fire engineers or for any users of evacuation models, because they mostly attempt to model something that doesn't happen in real life or hyper very rarely, or that it is not the assumptions that we use in design. So this disconnection between the real world and the real world. And the modeling world unfortunately creates a lot of waste in the research community. So there needs to be effort to try to avoid this somehow. And I mean, that was the motivation for one of the motivation for our groups, on human behaviour and fires of the IFSS was indeed put ourselves together, hit some hot topics that we thought could be interesting both within the community, but also outside the community so that if someone is approaching our field, at least can have a general overview let's say, on the common mistakes the common problems that we see when a new people, approach, this field.

Okay. Let's try and figure out what's really the issue is. So in my understanding, you can use term panic or whatever other way you would like to name it. But you can use this term as a proxy of a very complex human behavior. Like you say, okay, this group of people or this person has panicked which means there is a new rationale for their behavior. They will act like this. They will do this. And you also used the term That it's contagious. And I think that it was how we. Previously described it can spread from person to person. So Suddenly you have a fairly easy way to model behavior of large groups of people. Right. And in reality, you have evacuation scientists looking through through real cases of fires, looking through data, carrying your experiments. You do not see this simple negative behavior. I had the Erica in the podcast and she was mentioning the complexity of modeling behavior So in one way, putting this in, in this panic route or, or this simplistic behavior route gives you the benefit of modeling. And, you can claim that you've modeled behavior, but in world you did not. You just shown the proxy, which actually is incorrect.

it's like, you're building a video game, but that's nothing to do with with reality because it's much easier. Let's say model social influence as a contagion, this spreads with contact with physical contact like, okay, like I touch you, you become a red dot, you wear black dot and so on, but this is not how it works. And that's the problem that we face. We face the situation in which the leadership. Unfortunately, it's kind of split, into a part of, of scientists that to model the real world phenomenon. And this, as you said, and as several other colleagues say, like Erika Kuligowski many others. , it is not easy. You can either approach this with, let's say more macroscopic approach or microscopic approach. Try to look at each individual person. But in any case, it's not simple because you will have to factor in many variables. One of we know is social influence, which not based in relying on who we have, around us, but also in who we are in the simulation, so that the actual profiling each individual person. It's also important all these models, they are already, there are a few very good models technically for modeling these things. But the problem is how to calibrate them real world scenarios real world data. Knowing that even to get hold of these data is very difficult. And even if we have dates. Of this kind, sometimes the, either come from observational studies in which you don't have level of detail let's say, what is the inner decision-making process that people do all the other side? The problem is when you have like a. Uh, instead a stated preference studies in which actually people claim what they will do, they have problems with validity because then you may have people claiming they will do something. But, , we do not know how often, how well this relate would actually what will happen in the real world. So of these type of data carry their own uncertainties and their own limitations. So. The solution is not just to make everything simple, overly simple, and just say, okay, we do a spread of contagion model and we have solved the problem. and we can use this for design that the solution is to try to, first of all, as scientists to understand what are these fundamental variables that affect decision-making, as modeler. Just let's stick to what we know from the real world. So go ahead with, the knowledge that we have. Let's not try to go one step forward into try to model things that we don't know because. That's not gonna work. That's not gonna work. Any model will not be meaningful for fire engineers it's not validated and if it's not properly calibrated. And that's what I'm saying. There are, there is for instance, very interesting work that will. A couple of years ago, also by Erica, Steve Gwynne Michael Kinsey a couple of colleagues there, Max Kinateder, they were looking at so-called behavioral statements. So they were saying, okay, we have a

Yeah.

statements we can produce, which in very short sentences, they tell us what we know about human behavior and fire. So what we actually can say with a certain degree of certainty that is going to happen. So one classic example is. Humans tend to satisfy rather than optimize. So, humans will not necessarily pick the exit that will give us the short this time, but they will pick the exit in the building that will make them arrive safely. They don't care if they're riding. 10 seconds before or 10 seconds later. But you seen instead this kind of flawed of optimization models, which in one sense, I understand why they are there because okay. You want to first know what is the condition, the optimal condition in order to strive for that. But they see much more problems, for instance, understanding in understanding how people comply to what they're told to do rather than, knowing what is the optimal. So much more. At least my view and I'm, and I'm very opinionated about

Um,

much more important to put efforts into understanding what bring to comply, to information that are given either, the building itself. So. Environmental features by intervention of the, of building managers, than putting a huge effort in studying the exact optimum evacuation strategy a building, we need something that works. again, if it takes 10 seconds, less or 10 seconds more to an individual person will not make a difference. , so I think there is a. Wrong allocation of resources in researching these terms. And that probably comes from the fact that it's much easier to build an optimization model rather than doing an experiment.

Yeah, think in the contrast to what you say about this complex modeling of behavior and even the route choice. In contrast to that most of the currently used models would just have some shortest path algorithm to decide which path you would take. Maybe some social force model to, um, push the agents from each other, maybe, maybe some model of attractiveness of exits, where you could physically alter the, how attractive one exit is from another. But from my experience in this practical modeling, I would say I use it mostly for engineering not for science, so I have never ventured into writing my own models or. I've never really went too deep beyond what my contract required. me to do. So ma You know, I'm an engineer. I, it's not my job to figure out the best way when the tools are on the market. And they are advertised are the industry standard. right? So you have this contrast of what it can be with modeling behavior and what it is mostly, which is route optimization or. path optimization And just movement calculations and there's nothing wrong with movement. calculations. It's just, it's not modeling of complex behavior of people. And Do you think this, creates this, let's say dualism in the world of evacuation science?

I think in general terms, the key is flexibility. This is the key word. So when you use evacuation models, and I will say many commercial model developers have understood that it's much more important for a user to be able to control. the behavior of agents in order to be able to represent different "what if" scenarios then to have the most optimized algorithm for route choice, let's say, or for a people movement and so on. So I think that's something that. Many people in the commercial, uh, evacuation model world, have well, implemented and understood. The problem is actually with researchers. And I say this with a lot of pain in my heart, but there is a full, uh, body of literature, digs a lot into modeling things that are not really needed for engineers, they are actually starting from wrong assumptions and that's the real

that's why

why a lot, the efforts for instance, that a modelers do they start with the model and then they jump into. Data and try to validate it with data all the way around. data driven models in which they start with

Yeah.

and then they try to build them all. That's a very good starting point for an evacuation model. But don't see these. I mean, I see these, but I don't see as often as I would like it to see, because there is still of the literature of people that have basically never done. an evacuation experiments in their life and they build models or they, and even worse because that's in principle is okay that you are a specialized modeler and not into experiments, but it also, and never probably read behavior And the actual, experimental data. And that's the problem. And that's where you see this mismatch, where that's, where you see, these publications that unfortunately even get published in scientific journals are very little link, to the real ward and unfortunately, very little use, to fire engineers or in general, to any users of evacuation models, because seem like, uh, exercises done often, I don't know, a graduate students or thesis students that have to publish something. PA

Um,

to publish. it's mathematically beautiful. I don't question this. Maybe some people a really good at that. Maybe better than many of us, but it's very little use dilutes the literature in the field and he also has very little use for engineers. And again, I'm not arguing that all literature should be a linked to the practical ward, but you should not mislead at least engineers. You should not lead them into doing, design mistakes. And this is unfortunately what I see because this literature is so big of those kinds of papers, trying to modeling panic or contagion or, , whatever. Let's say behavior that has been, uh, deconstructed and, dismounted in the literature, in the field of social psychology or similar fields it is actually warring at this point because we, we are not able to stop this, , and they cite each other and, they continue doing this.

I can make an example.

example. A nice chat at some point with an applied mathematician. And he told me something that I always suspect is a crowd modeler. And the Emiliano Christiana is from Rome. And anytime, you know, there are some people that really fall in love, with certain mathematical equations and they will

Yeah.

in our field to squeeze those everywhere. So they say, oh, I can use this also for crowd dynamics and. And, for those that instead, try to link this to the real world and, and have a use of this. This is painful. is painful because you see such a waste of resources. So these people could be

Yeah.

brilliant solving, certain problems with evacuation modeling, but they putting said so much emphasis on something else that is not needed. So that's what we both were saying. Okay. We wished that we will have, the right effort. Putting the right problems, in our community.

I think this dilution is not only, typical for evacuation science. I see that a lot in tunneling example, there's a group of researchers who will, determine that back layering distance or critical velocity, The, the fourth, , significant digit when, as an engineer is just double it And then make it safe. You don't really care that much. and as as you mentioned, this creates this powerful ecosystem. I mean, when it's just one paper that does that or one researcher who does it it can be an error It can be in mistake. It can be just a honest misjudgment of a scientist, but when it forms an ecosystem where this is continuously repeated, repeated, repeated, you Start building on the parallel paradigm. You need to build science you need to have paradigms. And if, in your case modeling, assuming some certain behaviors some certain extreme behaviors that do not have, Proof in really evacuation studies, a real fires. if that suddenly becomes a paradigm for. a part of your field, it, this despite the field is genuinely lost. And as you mentioned, that is waste of resources that is waste of human time that this waste of work sometimes, brilliant scientists who are really good at what they're doing. For example, applying this mathematical problems to solve something. And we are not a big field. We don't have people to lose. We we need all hands on deck. solving important problems, of our discipline. Do you think may be sometimes entering the spiral can be just a conscious choice because it may be somewhat easier or

I honestly don't think so. I think it's generally a genuine lack of understanding of the literature, which leads to basically sticking to a certain type of terminology and a certain type of assumptions that let's say if we are split and we are, , let's say a percentage of people that claim that we should modeling panic, a contagion and so on, and let's say. Part of the community is trying to explain over and over and over in the last 30, 40 years. I mean, this morning we were discussing it that the first papers discussing panic misconceptions, all the come from the seventies or things like this. So it's still that we're battling. I think there is a genuinely honest problem with the community is not able to intervene to spread this misconception. And in the, in our case, this is also very much fed by media he's fed, by, you know, everything that you watch outside the scientific world because that's the problem that we have as well, maybe other domain, , modeling of fires. It is not such a big problem, in our domain, the thing is that when you turn on the TV or open a newspaper online, always this tendency to have a sensational titles and so on. So they want to use kind of extreme. wording and depict this extreme behavior, even if they didn't happen because their goal is often to sell newspapers rather than, actually, telling what, what happened. But I said, I don't think there is a need an agenda in a way it's more like a genuine mistake made by a portion of the community that is. comes another world I mentioned the world of applied mathematics, but it can be applied physics or anything that is not to the core literature or in the field of psychology, social psychology. And so on that explained us this mechanism, of human behavior and fire, work. so this is what I think is the, is the problem nowadays. And to be honest with you, I'm even more worried now with the new trends of science and machine learning and so on, because these are fantastic tools, but they, often start from another premise that we don't want to understand the fundamental mechanisms, we want to identify patterns in what happens. And often this leads to the blind trusting the data, which is the biggest problem that we have in this case. And which I think we should be very careful about because there is a huge risk that we will end up now with a whole bunch of new literature, , in the domain of evacuation using these kind of data science approaches with machine learning and so on, may really start from data that have problems. And instead they take them as good by default and come up with wrong conclusions. And as I say, don't get me wrong because. Big, a big enthusiast, myself of this kind of data science approach and machine learning. But I just have to put down a warning there for our field specifically because there is this risk, to data, and get, start getting models that. Down conclusions that do not make too much sense just because we are not putting efforts in understanding the fundamentals what is happening, but just trying to identify patterns into what happens

so in my field using machine learning or artificial and any type of big data tool actually is, let's say straightforward because I'm working with fundamental physics, the conservation of energy in the mass is always the same the way how, I don't know, pressure or density changes with temperature. I can describe this with an equation, even if I have very, very complex relation between the group of variables and my outcome. It is in the physical low and here by data, I assume you mean the common tools for evacuation modelings, which would be surveys, which would be data collected from, evacuation drills and real incidents. It would be a mathematical data describing the movement of people, but mathematical data cannot describe the behavior of people or their choices, or at least not in such a straightforward way as I can describe the motion of smoke or heat transfer through a partition in the building. And so you are really like Your data by assumption has this level of There is that you first need to understand that the variability in your data set and you first need to understand what your data represents before you drop it into a black book seeking for hidden relations. Is that what you meant?

I think this is exactly what I mean. And sometime I make decent example. When I talk about this, I don't know how many of you are chess players, but you've probably heard the story of alpha zero, this super software that has been built learning from millions of chess games and the game and actually having one of the main grand master number one in one ofMagnus Carlssen and learning the software that use this kind of, algorithm carying just about the final outcome, not about the chest theories on everything that we w we know from the theory. in that context, this works really well because assumptions to start with, so the movement of the pieces are what they are. So there is no, let's say a M B

We showed them

a straight line. They move diagonally. That's it. So instead. What we see in our field is that there are also aspects and assumptions that are really, questionable sometimes, and are they're really object of, debates how and why certain behavior, occur. and also they are depending on the local conditions that you have, because if you have a very large data set, I mean, I spoke with a very brilliant researchers. With the massive data sets of, pedestrian movement. Like for instance, I think that the group that has the largest collection of data sets on this, it's a university of Eindhoven with Alessandra Corvette and Federico Tasco. These guys, they put sensors in train station, they collect millions of trajectories. So it's really fascinating what they do. But imagine even with this kind of huge. This will still be a very specific population in a very specific context, in a very specific environment. So to generalize findings, eh, he needs another step he needs. Like I, I see those two things as complimentary. I'm really happy that these heifers are hip are happening I think they're really good, but we cannot just rely on those models try to identify patterns in these big data. We also need to try to dig up. Into understanding what are the fundamental processes? And theories that can explain we see. That's why hope that many of the people that work with these methods probably wouldn't agree with me. don't think that those data science methods should substitute what we have been doing so far. They should compliment it, but we cannot have this a subsidy. our current, theoretical methods, when it comes to those types of data, for which we have a lot of uncertainty to begin with. And as I say, we may argue for hours, what domains of fire science. They have those kinds of uncertainties, which ones they have not. heard a lot of discussion for instance, and I'm not expert in that, but I've

I've heard a little discussion. on Why

use deterministic models a lot infire modelling than, much

watch more

models. And there,

there, you have a

model, mostly because you have a physical phenomenon, but even in that physical phenomenon, there are so many things that can impact it. So many boundary conditions and, and all sorts of other variables that we cannot fully control. In our subfield of evacuation this is evident to me that we are not yet at the stage of science understanding of which we can completely control, , and fully understand what leads to certain behavior. So we need to be really cautious when we use, data science approaches, because they often. Take the data as they are. They don't question them. So

the next day.

and I think there are groups that are trying to do this in also to try to understand how

Oh,

treat these kinds of data in order to understand, because sensor, any device, anything can have fault or can, there can be a inner problem with the data so that's why I think we should be very cautious with this. And, and maybe I don't want to sound over negative, but I think these are very good tools, but they should be seen as complimentary, not as a substitution to everything that we've been doing so far regarding fundamental theories.

I have a challenging thought for you. if I look Well, in the years of development of the tools we use to predict, evacuation processes and buildings. whether you go back to like seventies, eighties, you use graphs to solve it on a piece of paper, or you use the first spreadsheets to calculate it in computer, or you start going into computer models which were it's a glorified spreadsheets, or you go into models that model that go with, you know, movement trajectories and calculate parts, social forces, and so on. Or even you jump into. machine learning, You still have the exact same constraint related to your boundary of the human decision. process. like in every single of these tools, you are constrained by the exact same limitation of not knowing the, principles behind human behaviors. So maybe if that's the case for all from the simplest, tool to the advance or a modern tool you have, maybe that's a Place where you should push your heard of evacuation scientists to solve. maybe we need a standardized model. of a human behavior that could be just applied in research instead of, trying to figure it out, from scratch. what would be this, if you had the power to just solve it, instantly which path would you, would you go and, what would you take to deliver that to the community?

I mean, there have been efforts in this direction, at least to understand some of the key variables and a key mechanism. So you wouldn't be able to fire. And as I said, I think exercise that all models. Do is to take indeed what I mentioned before this behavioral statements, are a list of statements that describe what happens with humans, during a fire emergency what do they do? And try to violate their assumptions again, against those. So to see there are assumptions, hold against this

That's good.

now as a researcher and also fortunately, as mentors of researchers, what they tell them. , every time that you write a research paper in which you are investigating a specific population or a specific, scenario and so on, to create these behavioral statements. So try to create simple statement that concisely. Describe, what actually happened in that given situation so that a modeler can take this statement and use it as a benchmark the assumptions that are used in the model. Because as I said, we do this exercise, if we take this knowledge that we have nowadays, and we compare it, with models, we will definitely find some models that violate through their assumptions, some of these statements. as I say, the trends that I see, especially in the commercial world, and that's a, a bit strange because I would have expected more researchers to keep up with. Actually see it much more in the commercial world. model developers in the evacuation world are trying to actually give flexibility to their model so that they, at least you can make sure that you can, calibrate the model in a way that doesn't violate certain assumptions, that you have. But as I said, in, in the research world unfortunately, it happens often that want to reinvent the wheel from scratch and that's not probably the right approach. I mean, if I will never imagistic and decide what to do with the research funding in the vacation world, in the world, I would basically put 90% of my money on data collection and theory and

The,

the modeling be done by the

um,

model developers in companies. They probably are much better writing codes than all of us. is at least how I will, how we'll do it,

so these rules could be, Or can be applied as some sort of sanity check if you're modeling makes sense. Right?

Yeah, that's the idea. So I mean to, try to benchmark the assumptions that you use against athose simple rules and, you know, there have been simple rules defined in different domains of the evacuation world that have been these, be able to statements that come from our own community for fire safety science. for instance, there are a lot of, efforts done in the pedestrian, evacuation dynamics community, instance, in identifying, rules of, interactions between people in a crowd. So at a more microscopic level, see, how does a lane of people get formed? Brilliant researching this or how actually queuing happens. So, trying to extract those rules from the individual papers, boil them down in simple statements and then benchmark the models that we have out there against those. see, does this get violated? To begin with, is this inbuilt in the defaults of the models? we modify the default so that at least we can tune the model in a way that these assumptions are not violated any more? So this is the kind of benchmarking work that I would like to see more and more. And I mean, it's all linked to the efforts that have been done over the years in and validating, evacuation models. You know, this

Y,

of my. main, life mission as a researcher to develop standardized testing for over evacuation models.

But as I said,

as I said, we need to have another step the community trying to make their simple enough to be directly implemented into models.

I have one more challenging. One asset. so I think in most of these approaches, we focus very much on the behavior of a single agent or a single person, that evacuate, but Do we truly need this microscopic approach to simulate the population, you know, to see large crowd patterns to see queuing, is, this the only way, because we started this with the issues with papers and his part of the science that goes in a slightly wrong way, because they misunderstand some of the assumptions and, to do things that are putting a lot of effort in researching things. that makes no make no sense. basically. so maybe if we could break the product that you modeled an agent, maybe there are waste, to skip that individual behavior and just focus on like group behavior or something. bigger. And how do you do wild land interface, fire? Do you, do still stick to agent or.

I think this is a, another very good point of discussion. I had several discussions about this with different researchers that work in different domain, both in the WUI fire world and in the crowd, evacuation world, I think there is not a type of modeling approach that works for everything. I think it really depends. Problem that you need to solve because some cases it is true to, you could actually use a very simple, say flow calculation or a very simple microscopic model and get a fairly decent approximation of the result that you need. in some other contexts is actually probably not enough, especially when you, the impact of an individual, a behavior or a small group of individuals can, uh, can be huge on your total evacuation time. Then you can adjust. The big picture you may get these diluted and maybe let's say overlook certain aspects. So discussion came up a lot, for instance, when it comes to WUI fires because there, there is even a lot of people arguing that we don't even have to

Yes.

pedestrian execution modeling within with our modeling because of the scale of the traffic. When you have scenarios, we, traffic evacuation is much bigger. but again, I will not the answer like yes we need or no, we don't. The answer is, it really depends on which scenario you're looking at and what conditions you're looking at. Both for buildings and WUI fires. we should be able to have in the available. a pool of tools then select the most appropriate one, given problems that we need to solve in the design issues that we have. So will not argue, I will not recommend to just abandon a microscopic modeling. That's definitely not a good idea. Uh, personally, I think, nor abandoned macroscopic modeling, because that is also useful

Okay.

many instances, but more like choosing. What is the right path that we need to take, depending on what applications we have and have those two tracks developing in parallel. So maybe microscopic models can learn something from microscopic models and vice versa, , because you know, there've been also efforts trying to combine those and the using,, hybrid, approaches and so on, but that's not a trivial,

would have to decide when a person changes from, , a person into a population. It's this, this sharp edge when If you would agree that that's some point that the population scale, these individual decisions have less and less impact because essentially they become statistics. if you care about the movement of a 1 million people. the decision of of a single person ends up being a statistics that's it. So, finding , this. Edge between them and applying correct approaches. to each, maybe an important, aspect of modeling overall And, I think in the same way, it's when you go from modeling agent movement into agent behavior, it's also a, a, shop interface where you literally stopped doing something very simple and jumped into a field where more complicated things, dictate the outcome. and like, honestly, from your experience, do you think in I don't know, next five 10 years we will have reliable behavioral models to use in our modeling, like I go today, I start up computer software of my choice to model. my, building, I pop in the pre evacuation distribution times and, I run the simulations. Will I be able in like five to 10 years, be able to just put default human behavior, model and then just run the same simulations without this.

I think in five years, no, probably I don't want to be over optimistic because that is something that we have been looking around already for quite some time. And, and we are digging into this, but I will argue the tools that we have today and as mentioned several of them, they are becoming more and more flexible in a way that have as users, as solid understanding. Of what are the scenarios and the behaviors that we want to represent. can, we actually have the design tools to do our design nowadays. So is missing here is a more research towards of everything that has to do with those outliers scenarios that we have. Understanding and control on because we know that in certain situations they may be a small group or of people doing a certain behavior that might dramatically affect our results. And, you know, when we design a performance based deisgn fire engineering, we actually look at RSET. So we look at the actual, , tail of, our evacuation time curves. And that's where I would like to. more research too, because now we can do it, but we have a, quite a large uncertainty in this. So I hope, and I really think that we will doing 10 years from now is to basically work more on, making a small. gap what is the result that we expect and the uncertainty that we

Amazingly.

to basically, uh, have anRSET because, you know, when we work with RSET ASET, sometimes actually we recommend to have these probabilistic approaches to. To make sure that in 10 years from now, this uncertainty is reduced. So span of possible results become smaller, which means that we have more confidence in our, RSET when we do our design. So this is what I see happening 10 years from now. I don't think we're going to be able to pin down to one RSET for each very specific scenario because also. That's not really how we work and how are. It's very, there's so many variables that can play a role that it's very hard to isolate them all, especially in the timeframe of five or 10 years from now, but we can improve our understanding of this uncertainty and decrease that uncertainty that we actually have a better understanding of what these rset. This requires safety egress time is.

I just can't stop myself from asking you for predictions for the future and for the new listeners of the podcast, there's like a whole episode with two Italians rambling on how the future of evacuation. modeling will look like. And I I highly recommend that, , the out it was a great discussion and I would also love to recommend your episode with Erica who. Explains how she's modeling behavior in wild land fires. And that is, that is truly beautiful. on the, how this aspects of human behavior, are being modeled and what it takes to actually model them. Maybe not the buildings outside of buildings, but still it's, Similar problem. And I absolutely love how she presented it and it really, shows the complexity of the problem. And I think through a resources like this, true research like that we, truly bring. more people closer to understanding. What is the big problem in here? what is the big picture in here? And if to understand that they have to move back to fundamentals and by moving back, to fundamentals, they move away from doing this misjudge conceptions that, you can simplify a human behavior into, , a phenomenon of panic Or whatever else. And, uh, I guess th that could be like a good. Summary on where should we go educate people on the complexity? And what it really means, to, to model behavior, to stop this overflow of bad science that in the end influences our discipline.

Correct. And I think this is a good take home message. So, , if you are, a user of models even if you are a researcher, back and start questioning, the assumptions, both that the model use. And that you are using because this is one of the key aspects that can lead, to mistakes and problems in your simulation world So I think, these should be the take home message, when it comes to, representation of such a complex problem, like human behavior and fire.

Fantastic. that's a great, summary. I I link all the resources into show notes, and I hope we've helped some scientists to get on the correct path and others to, to help us. , fight or battle of, bad science, thank you very much Enrico for coming to the show for the second time, it was a huge pleasure to have you here. And, uh, I have a feeling we will see each other, not so far from now. Again, maybe touching on the RSET concert. Oh man. That would be, that would be fun.

thanks. Wojciech was very nice to be back and, thanks a lot for all the efforts and time you put into this podcast because it's really fantastic. We really needed it.

Oh, thank you. I'm so happy. I didn't know what, I didn't know. we needed it. until I started it and now I just can't stop. It's really cool experience. Thank you.

Thanks. Bye-bye.

I hope you've enjoyed this episode. And if there's one takeaway that you can take from it, I guess a sanity check approach for testing your evacuation model would be the thing to take. Enrico mentioned the paper and I found it, the paper is

Guidance for the model developer on representing human behavior in egress models and it was published in Fire and Technology journal. It's authored by a Gwynne, Hulse and Kinsey. And I link. It in the episode description, obviously. Enrico mentioned the paper contains statements. And in fact it does. Statements that, are, let's say the things that we know about the behavior of people when they evacuate. And this can be used to cross check your analysis, whether you break this statements or not. And if you do well, then most likely there is some error or misconception or the model is used in an incorrect way. And. As we don't have yet the perfect models as we don't. Have you had the ability to just easily model the. Evacuation behavior of people. Actually cross-checking our simulations with such a sanity check is, sounds like a pretty powerful. Tool that can be used to make sure that. And modeling is of high quality, and I would recommend it to engineers and I would highly recommend it to scientists who research in this field. And as mentioned in the start of the episode, I would also highly recommend listening to that. IAFSS webinar. There was a channel on YouTube, uh, with, these webinars did this one was the third one. And they're all great. So I hope this is another. resource that is interesting to you. And, I hope you'll reach out for, for that. Thank you for being here with me. Thank you for another week of podcasting. I look forward to the next week. So see you again next Wednesday. Thank you. Bye.