The Walls Are Closing In: AI’s Impact on Emergency Management, Governance, and De-Skilling

Speaker A: So, um, generative AI has changed that completely. Now we see AI being leveraged for things that historically would have been thought as almost impossible to automate. So in the emergency management realm, some of the things that we see this task encroachment on currently are some of the basics like being able to develop emergency plans, being able to, uh, conduct exercises by the development of the scenario and capturing lessons learned. We see it being used to do risk analysis, being able to harness the raw compute power of these tools, to be able to pull in lots of risk information and put it into something that's meaningful for decision makers.

Speaker B: In a world filled with chaos and a myriad of risks, there is opportunity. You're listening to Riding the Wave Project Management for Emergency Managers, where we discuss how we adapt and rise above those rolling waves of hazards and threats we face and rise to the top. And now your host, the President of Pinnacle Performance Management, Andrew Boyarsky.

Speaker C: Justin Cates currently serves as Senior Business Continuity Advisor at UH WAWA Incorporated, where he's responsible for advancing enterprise wide business continuity and resilience for the company's convenience, retail and fuel operations. Before joining wawa, he was Director of Emergency Management for the cities of Somerville, Massachusetts and Nashua, New Hampshire and served with the Delaware Emergency Management Agency. He served as the President of IAEM M USA for the 2023, 2024 term and as Treasurer of the national association for Public Safety GIS Foundation. Um, he's also served on the FEMA National Advisory Council. And I'm very pleased to have you join the podcast.

Speaker A: It's great to be here, Andrew.

Speaker C: So you gave a presentation at the second, uh, annual Pracademic Emergency Management Homeland Security Summit on May 18, uh, earlier this year entitled the Walls are Closing, uh, AI and the Future of Emergency Management, uh, which seems, uh, pretty ominous. So in your, uh, presentation you talked about the four walls, uh, the first wall. And for folks who are interested, I'm going to include a link to this as well as some other links for Justin's bio and other publications, uh, and other, um, let's say digital assets that are on the web that you can read. So, um, the first wall is legitimacy and relevance as emergency managers. The second wall is the mounting demand for what emergency managers are expected to do, um, and I would say related to the third wall, which is we're expected to do more with less, uh, given the current financial constraints that we're facing. And that fourth wall, which we're going to talk about now, uh, that's coming at us like a little Bit like a deluge is artificial intelligence. So with that in mind, um, and you do make some recommendations, you refer to uh, Daniel Susskind's uh, framework on what will remain for people to do. Um, and again I'll conclude a link there. I um, want to start off with where is AI encroaching on tasks that emergency managers typically would do?

Speaker A: Yeah. So you know, the idea of task encroachment is one where we're looking at the variety of things that emergency managers do and try and understand what are the things that we were seen as the exclusive provider of that. Now automation and AI might be able to replace us. And we're seeing this across all parts of the knowledge work professions, um, where the things that you typically required a human to do now have um, this opportunity to be provided by an artificial intelligence tool. Um, historically there was sort of this thought that you wouldn't see automation and AI impact what we call those uh, non routine tasks. And these are tasks that um, have a lot of variety of them, that they don't um, operate the same way every single day. There's critical thinking needed, there's uh, different types of decisions that might be necessary. And there also uh, is a lot of challenges with being able to analyze those tasks. They are not easy to explain to another person. They're certainly not easy to explain to a computer so that it can replicate what you do. Automation was always seen as something that was really focused around routine tasks. Think of like an assembly line. It's the same process every day that needs to be performed. So um, generative AI has changed that completely. Now we see AI being leveraged for things that historically would have been thought as almost impossible to automate. So in the emergency management realm, uh, some of the things that we see this task encroachment on currently are some of the basics like being able to develop emergency plans, being able to uh, conduct exercises by the development of the scenario and capturing lessons learned. We see it being used to do risk analysis, um, being able to harness the raw compute, uh, power of these tools to be able to pull in lots of risk information, uh, and put it into something that's meaningful for decision makers. Uh, and then we're also seeing it being used for a lot of the administrative responsibilities that we have, whether it be writing grants, reviewing grants, managing all of the documentation and organization behind the scenes. So uh, AI is doing quite a bit of task encroachment in the emergency management space today.

Speaker C: What do you see as some of the dangers, uh, that you've either heard of uh, where emergency managers, um, you know, are. I um, would say misused. Um, so for example, we know uh, Google Maps for example. Everyone uses it for geolocation or not everyone, many people use it, it's commonly used. Um, now of course when you have wildfires and it's fast moving, they can basically possibly put people in danger. What looks like a clear road might be one that is an inferno. As a clear example, are there others that you've seen out there that you've seen are sort of clear dangers for how emergency managers are using them?

Speaker A: Yeah, uh, so maybe I'll break this up into sort of two different elements. So currently, um, when we think about some of the, the areas where there's some danger around what we're seeing AI used for, certainly AI does have challenges to this day with biases and uh, being able to uh, provide a realistic picture of something that we might need to make a decision on. So we know about that. Uh, we also see some of the dangers around uh, the scenarios where it's either got inaccurate information or it's got information gaps, the inputs to a model which uh, I think are kind of related to the example that you just described and it not being able to provide um, accurate, useful or intuitive information to the user because it's got those information gaps in them. So we know that those exist. We have seen improvements in that even just since um, when I think of uh, ChatGPT becoming a more commonplace name out there around 2023, that time frame. The models that we see today um, are much, much more improved than they were uh, just a few years ago. So the amount of change that we're seeing in this space is remarkable. And the amount of improvements to try and resolve some of those gaps are pretty impressive. I think. Um, currently some of the biggest, uh, poor uses that I'm seeing AI used for within the emergency management space align to this IDE of AI workslop, which I'm sure uh, many of the listeners have probably heard of, which is the use of these tools basically to um, maintain an existing way of working, whether it be creating long documents to make it seem like we have a lot of knowledge about something, or creating lots of presentations that people now have to uh, review and look through. I see a lot of that today within the emergency management space. People just simply using these AI tools, uh, to create the final deliverables of something. Just in the sense of, okay, well I've met the requirement checkbox and now it's completed. The challenge is that we've gotten so accustomed to that as emergency managers that we're now leveraging these AI tools for that purpose. I always tell people that, um, the mad lib style of emergency planning that we've always had within emergency management, where we rip off somebody's plan, fill in the blanks and then we call it our own, that was the original version of AI Workslop where we didn't go through the, uh, collaborative planning processes, we didn't do the work behind the final deliverable. So now we're just able to do that more quickly using AI tools. Uh, another area that has sort of been dangerous with, um, AI use in emergency management is in some of the processes and workflows that don't really provide a lot of value to emergency managers. One great example is around competitive grants. So many emergency managers know the challenges of having the right grant applications, align them to different, um, principles and priorities that other levels of government might have. Now what we're seeing is organizations using these AI tools to write their grant proposal, saving a lot of time. That was really useless from the beginning. And we see the grant, um, uh, agencies using AI to review those, uh, grant proposals. And so we sort of call it like Batbot are bot battles where now we have a computer talking to a computer, uh, to basically go through this meaningless process. So I think in the future we'll actually see AI leverage to replace those workflows that we're so accustomed to in emergency management and being able to leverage accurate data about vulnerability and social characteristics of jurisdictions, resource needs to be able to accurately identify where resources should be distributed to rather than the political and, uh, competitive nature that we see current grant processes. Um, and then I also, I think one of the biggest challenges where I see some misuse of AI, or maybe not use at all of AI, is we know some of these tools just in their current format, are able to really provide great capability to emergency managers in areas that we know as humans we don't do so well. One of the best examples being alerts and warnings. Every common AI tool that we know of out there, Whether it be ChatGPT, Gemini, uh, Claude, by Anthropic, they all have the functionality to be trained to serve as sort, um, of a tool for a specific purpose. And in this case being able to craft an effective alert for a jurisdiction. And we've got plenty of years of evidence that describes what that alert should look like, the format, the structure, uh, and then knowing the constraints that we have to deal with, how many characters it has to be to go into a, uh, wireless emergency alert, uh, format. And we know that this capability exists today. Yet we still see emergency managers try and pull this out of their rear end in the middle of an emergency and craft an alert that doesn't follow that format. So in some cases the misuse is actually the inability to use it at all. Uh, so I think that's definitely one of the big challenges with AI adoption currently within emergency management.

Speaker C: So I want to change gears a little bit, um, and step back and look at the governance aspect. And how do you see maintaining that level of governance, uh, of the use of AI in emergency management work?

Speaker A: Yeah, I mean this actually came up. I was just uh, in Bentonville, Arkansas this past week, uh, at an event that the U.S. chamber of Commerce foundation was uh, running called their Disaster Resilience Summit. One of the breakouts that we had during that session, uh, was around, uh, activation of AI tooling within the business continuity and emergency management and disaster relief spaces. And one of the common things that came out of those discussions was around accountability, uh, because it's difficult to uh, assign blame to a computer when something doesn't go right or wasn't done properly. There still needs to be a tie back to uh, a human, uh, in a lot of cases. And so, um, I think the first big thing around governance that people really need to consider is to look at the organization that they're operating in. What guardrails have they already put in place around artificial intelligence? So is there a policy in place? Are there specific tools that have been authorized for use that comply with whatever, uh, uh, data protections have been put in place for that organization? Uh, what we want to try and do is avoid the examples of shadow AI where people might be using their own personal accounts and tooling to leverage either proprietary data that we might be concerned about in the private sector, but in the case of uh, any organization, government, nonprofits or private sector, uh, sensitive data like personal, personal identifiable information, um, or sensitive, uh, data about critical infrastructure or anything like that. So that's the first step is really just understand what guardrails are currently in place by the organization. The second, uh, piece is really about, uh, training, awareness and upskilling of the staff who are involved in the use of these tools. So we want folks to understand how to use these tools safely, uh, what are the restrictions that have been put in place on how they're to be used, um, and making sure that they're aware of the limitations, the biases and how these tools actually work, where they might hallucinate and those types of things. So training is a big piece of the governance model as well. Now in the future I think there's another area that needs to be considered uh, that we don't see much of today which is really around emergency management's um, response responsibilities. Uh, so when we think of the four phases of emergency management, mitigation, preparedness, response, recovery, response is really the only one that we would consider like a high reliability type uh, of organization. And it's where critical decisions need to be made with not very much data. There's certainly bad consequences if we do the wrong thing during that phase. It's not as big of a concern if uh, the mitigation plan is a couple months late. But if we didn't open up a shelter in time the consequences are much more severe as we start to use these tools for decision support, making sure that we're familiar with how to fall back to manual processes if we're not able to leverage uh, an automation tool that now is sort of a critical part of our workflow. Um, and thinking of these tools from like a business continuity or continuity of, of operations perspective. If we now have these tools as sort of an essential part of performing day to day tasks, do we have the ability to continue those tasks when they're not available for some reason? Just in the same way that Microsoft 365 might go down or our enterprise resource management tools might go down. So those are things that I think also need to be considered as part of governance as we really integrate these things more and more into our day to day processes.

Speaker C: So I want to touch upon a similar issue which is how do we address uh, the deskilling risk? I think you talked a little bit, a bit about that, but if you could speak to that specifically.

Speaker A: Yeah, so to give the listeners some context here, deskilling sort of described in two different ways depending on what you read. I think one of the common uh, ways that it's described is around uh, the loss of um, our human ability to perform certain types of tasks just because now we're relying on a computer to do it and so um, we've just forgotten how to do it. Uh, sometimes we hear the idea of cognitive offloading where we're actually uh, starting to lose our ability to think critically and make tough decisions because we're putting a lot of that onto a computer. So that's definitely one element of it. But the area that I'm mostly concerned about and that I've um, talked a lot about from an emergency management context is the use of these tools in a very intuitive fashion in the same way that personnel might use Microsoft Office or now even GIS tools, which years and years ago relied on a very knowledgeable and, um, um, highly trained specialist to perform those activities. Now many GIS tools can be operated by the layperson, uh, and do some amazing things. So when we look at the variety of tasks within the emergency management space that relied on an expert, a professional emergency manager, now with AI tooling, a lot of those decisions, analysis that are done can be done by these tools. Emergency management has historically had this challenge of, uh, comparing professional emergency managers to those who are doing this as a collateral duty might be a fire chief, a police chief. Uh, in the private sector it could be like a facilities manager or, or a safety director or something like that. And now armed with these tools, they're able to perform many of the tasks that you required an expert emergency manager to perform. So this is, I think, one of the most problematic things about, um, the AI field when looking at emergency management. And you described earlier in the presentation that, uh, I did for the Pracademic Summit, I described sort of a framework that Daniel Susskind had outlined in an article that he had recently written about what he thinks will remain for humans to do. And I tried to apply that to the emergency management space. And there were three primary areas that I, uh, think may exist in the future for emergency managers. The first being this idea of general equilibrium limits. When we look at the slate of tasks that emergency managers currently perform, if we look at all of them and say, okay, what are the things where computers do the best job at currently? And then comparatively, where would humans be able to step in and do the remainder? Uh, in our case, I think in emergency management, those human relationships, the political nuances that we might have to deal with those types of actions and then leaving the detailed analysis and data collection and those types of things to computers to focus on, there's also this idea of preference limits. So, so in what cases do we prefer a human to do a task versus a computer? And even within that realm, there's sort of three different categories. One is a static limits. We prefer somebody to paint a painting for us rather than a computer to do it for us. There's also achievement. We would, uh, see in a variety of cases that we'd prefer to play chess with somebody and beat somebody else rather than the case of a computer playing against another computer. There's no real, uh, interest in that. And then the last is empathetic. We see this In a lot of like, care work, um, you might prefer to have a nurse take care of you when you're in the hospital rather than a robot. And I think that may be a remaining piece for emergency managers. We often see emergency managers after a disaster, uh, making that human connection about, uh, how they're providing support to their community or their organization and getting things back up and running. And then that final category is this idea of moral limits. And it's where we start thinking about things where there's some need for a human to be a part of the decision making process. Because of the accountability issues we talked about earlier, there might be legal requirements that require a human to intervene in something. And then also this idea of having a human in the loop around critical decisions, being able to stop something before it goes further in order to be that double check and make sure that, uh, we haven't done something that we would regret later. The biggest challenge though with these limits is around this idea that many of the things that I think will remain in the future are often tasks that are held by the politically appointed emergency manager or uh, the senior official who has emergency management within their organization, whether it be a fire chief who has the emergency management office embedded in their organization. Often they're the ones that are cultivating those political relationships. They're the ones at the podium during a disaster press conference serving as the face of the emergency. And they're the ones that are often called to make those critical moral decisions about evacuations and whether we're going to send out an alert and not so much the planners and analysts within the emergency management programs that uh, are probably going to be the first to go when we start to leverage these tools more. So, uh, definitely a challenge, this deskilling risk. And I think emergency managers need to think about how they can use these tools as a competitive advantage against some of those other collateral duty emergency managers and demonstrate their value that they provide to their communities or to the organizations that they work in.

Speaker C: And it seems to me, if I might just add to this, and that is there's also an institutional knowledge, uh, and expertise that might be missing or walking out the door with a younger generation coming on. If we're relying more on these tools as opposed to, you know, learning at, you know, someone's elbow, so to speak, in terms of doing the job. And you know, and as you said before, what happens when the systems go down? Uh, the other day, because I use three different, primarily three different, uh, AIs for some of the work that I do, I use tend to lean on

Speaker A: Claude a little bit more.

Speaker C: Well, Claude, AI went down or wasn't working fully uh, the other day. And so I was easy able to offload it on other AIs because I have subscriptions to them. But what happens if you're relying on that as one of your primary components for the systems that you use?

Speaker A: Well, and I think one of the pieces of that is there are other high reliability organizations that, that, that we can reference on how we should apply the use of automation within the emergency management response phase. Again, I highly believe that that's the area that we would need to focus on when it comes to being concerned about outages and disruptions and the loss of our skills. Um, I don't think there's as much of a concern if we weren't able to do a preparedness presentation for a community center. Uh, it's more on the how do we ensure that we can make critical decisions about moving resources around and sending out alerts and uh, those types of things. And when we look at those other high reliability organizations, take aviation as an example. There's a lot of automation in aviation. You think about every plane, commercial, uh, airliner that we fly on leverages autopilot significantly. Yet there's significant amounts of regulation and industry standards and requirements that say that we need to train those pilots for those novel situations where they might be confronted with a loss of automation. And so we would need to do the same thing in the emergency management response phase to ensure that emergency managers know how to do those very fundamental tasks. Getting an alert out, opening up a shelter, uh, convening an eoc. You know, those fundamental response capabilities that uh, we want to be able to do without a delay.

Speaker C: Where do project management skills and methods come into play to enhance the role of an emergency manager?

Speaker A: Yeah, this is, this is an interesting question. And um, when we think about a lot of the work we've done over the last couple, uh, of years, uh, I know you've done a lot of work in this space and there's a small cohort of emergency managers who really have been beating the project management drum. Especially when we think of how many of the tasks that we've talked about rely heavily on robust project management skills. Um, project management skills are one of the areas where I think automation, uh, and a lot of the modern AI tooling can really be an assistive resource. Um, for instance, if we look at Microsoft, uh, Planner, which is one of the common project management platforms that's out there to help organize a project and Keep us on track. They've just added um, within the last uh, year agent functionality that you can use as almost a project management assistant to help you go out and do research or to convene stakeholders, uh, to schedule a meeting. Um, and I think the thing that we need to ensure that those emergency managers uh, that are trying to become better project managers are considering how they can use these agents within ah, a project management tool to help um, provide some additional capacity. Uh, if they can use an agent to do some of the tasks that would have required them uh, to spend some of their time or that would have been manual toil, that doesn't really provide value back to the customer. That's I think a great way for them to reduce that burden. Uh, but even beyond the idea of these agents, which is sort of a modern um, uh feature that we've seen added into the AI world, uh, just using basic AI tools to improve their project management workflows is I think something that everybody can do. Uh, I know that again all the common platforms, Gemini, Claude, uh, chatgpt, um, they all have resources to be able to help with providing many of the artifacts that we have during a project, whether it be development of a Gantt chart, using some data that we might have. It'll take you a lot less time to use one of these tools than to manually do it, uh, using um, previous methodologies.

Speaker C: Um,

Speaker A: I think that there is a lot of integration there. I think it's definitely an additional area of education that project managers are going to have to consider to help do their job much more effectively in the future.

Speaker C: Yeah, I know not to get too, I guess wonky on this and that is there's a chapter in my book where I talk about the prospect of being able to use project management tool methods and tools that are not necessarily in the main. You know, we're familiar with critical path method, um, you know, program evaluation review technique, pert, which is more of a probabilistic method. And then there's conditional planning which there really isn't a lot talked about in your home state of Delaware. You know that was uh, actually created by I believe it was Dow Chemical in terms of developing new chemical plants, uh, way, way back when we're going back 50, 60 years ago plus. Um, and what's interesting is that when I think about those kinds of methods and how do you actually use that within our domain for uh, those types of variable planning? It becomes easier when you're able to use these types of tools. And so we can't not that, as you put it, we'd want to do it in those, um, high reliability types of operations, um, but we can test it in terms of, you know, given certain circumstances and really push the limits of what we're, uh, able to do there. So, Justin, I want to thank you very much for coming on the podcast, uh, and appreciate your time.

Speaker A: Absolutely. No, it's been a pleasure. And uh, again, I always appreciate the work that you do to try and get more emergency managers thinking outside of the box, uh, beyond just the traditional I need to write an emergency plan, I need to open up a shelter. And thinking about how they can leverage project management. And now, uh, looking at what they might do to, um, most effectively implement AI within their programs while not degrading their capacity to serve when they might fail or not be available. So thanks again.

Speaker C: Justin Cates, uh, currently serves as a Senior Business Continuity Advisor at WAWA Incorporated. Um, before joining wawa, he was Director of Emergency Management for the cities of Somerville, Massachusetts and Nashville, New Hampshire. Uh, and he also served as President of IAM usa, uh, as well as served on the FEMA National Advisory Council. It was great having you.

Speaker A: Thank you.

Speaker C: You've got mitigation projects, grant applications, training and exercises to deliver, but keeping all of it moving on time on budget with the team you have. That's where good intentions stall. Project Management for Emergency Managers workshop gives you the practical tools to manage, scope, schedule, budget and deliver. Built specifically for how you work. Walk in with your real projects, walk out with a real plan. Taught by practitioners who've led responses from 911 to Sandy to Covid workshop options from one to four days in person or online, it's time to move from planning to done. Visit pm4emm.com to learn more.

Speaker B: You've been listening to Riding the Wave, hosted by Andrew Boyarski, president of Pinnacle Performance Management and Clinical Associate professor in Emergency and Project Management at NYU and John Jay College.