How Alberta Wildfire uses AI to suppress fires and save lives​

Speaker A: Welcome to Pivotal. I'm um, Hayed Galou, corporate vice president for commercial solution areas at Microsoft. I work with customers around the globe to transform their business through technology. At, ah, the center of every transformation are people who give technology its purpose. And that doesn't change with the advent of AI. It's actually being accelerated. People spark visionary ideas for leveraging technology. The release of AI technology like ChatGPT this year is exciting, but it has led to big questions as we grapple with the best way to harness those tools to enhance and support the people behind the work. We like to talk about technology. I love to talk about, but we often forget that technology is most effective when it supports people with purpose. This season will demystify AI by talking to the innovators using new AI technology to uplift their industries and augment their people. From education to journalism to surfing. And it just illustrates what AI is about. Everybody thinks it's about tech. No, everybody's using AI and that's what we're going to show you on this season. 2023 was Canada's worst wildfire season on record. The fires forced thousands of people to flee their homes and burned 42 million acres. The wildfire damage drove a 24% increase in the loss of the world's tree cover. According to the World Resources Institute, Canada's tree cover loss hit 8 million hectares last year, up from just over 2 million the year prior. These are sobering statistics to keep up with this growing problem of wildfire. Some are actually turning to AI for innovative solutions. Wildfire management specialists are, uh, leveraging AI unique capability to analyze data to give them an edge in their fight. Since 2022, the province's Forest firefighting agency has been using a new AI tool to help support their strategic decision making and help direct their finite resources more effectively.

Speaker B: Ed Trenchard and I'm a provincial wildfire management specialist. Most people's journeys in Alberta start with uh, forestry. So as a kid I wanted to be a park ranger at university, took a degree in forest management and I worked in the forest industry for a number of years, dealing with trees, measuring trees, seeing if trees were growing. And then while, uh, I live in the forest, a job came up with the government for a wildfire ranger. So I applied on the job and I became a wildfire ranger. That was almost 20 years ago and uh, grew in my career to where I am now as a provincial wildfire management specialist.

Speaker A: Over the last 20 years, Ed has experienced firsthand the dramatic environmental shifts.

Speaker B: We do know that we are getting more I'll call it extreme, but the seasons are more lengthy. I'll say. In Alberta, our fire seasons start earlier and end later. The other thing that we are seeing is more impactful fires to communities and people. Um, I'm not sure if it's because we're getting more large, uh, scale fires or if it's because we have more people and communities in the forest to interact with those fires. So it's a little bit of both for sure. But we are seeing a shift in the length of the fire season, likely due to climate change. So we've adjusted our fire season timing so that's the biggest shift and the fact that, uh, we don't have the amount of resources to deal with all the fires on the landscape. So we have to prioritize, uh, where we put resources and how we fight fires.

Speaker A: At Alberta Wildfire, duty officers are on the front lines looking out for the latest developments that need attention. During fire season, every day for duty officers begins the same.

Speaker B: Get up, go to work, check the weather, check what happened overnight with all the fires in your forest area or the province. See if we had fire growing overnight, see what the forecasted conditions are through the day, uh, see if we have enough resources, do we need more resources today or for the next five days and start uh, figuring out where those resources are going to come from.

Speaker A: And then they sit through briefings from many different departments devoted to understanding and modeling wildfire behavior.

Speaker B: We have a number ah, of weather meteorologists on staff that uh, brief us on weather. We have fire behavior analysts, which I'm one of those as well, that will brief us on how the fuels are going to interact with the predicted fire behavior. And then uh, lots of fire modelers and the fire modelers are going to put all that together and grow fires across the landscape.

Speaker A: And then it's time to assign resources.

Speaker B: We assign helicopters to crews and dozer bosses to dozer equipment and ensure they're ready to respond to fires. For those waiting for the next fire, there'll be a bunch of people on fires putting out fires that have occurred in the past weeks or months or the day before. And then we have a bunch of resources that are just ready for the next fire. So you'll be briefing those crews through the day. Once you do have a detection from a fire tower or some people phone in, there's a fire in the forest, then we start sending resources. So we assess where's that location, what do we need to send, do we need to send air, tankers, helicopters, crews on the ground? Or all of it.

Speaker A: If the fire is in close proximity to the community, they send everything.

Speaker B: So everything goes. Obviously we keep stuff for the next ignition. By everything it goes, I mean that we'll send an air tanker, a helicopter and ground crews to deal with it. But we have more in reserve for the next ignition as well. So that's kind of the, uh, typical day. And you kind of rotate and it's like Groundhog Day. Every day is the same.

Speaker A: A lot is known about how a wildfire behaves once it started. And Ed and his team draw on their intuition and experience along with the daily briefings from fire modelers and meteorologists.

Speaker B: We know what's going to happen across the province because our weather forecasts are very accurate and our fire behavior staff are very good at their jobs. So we know what type of fuel, we know the topography at every location, so we can predict what's going to happen. What we can't predict is where the next match is going to drop. So that's what the tool. That's the problem that this tool is endeavoring to solve. Because we can have, uh, the highest, most extreme hazard, but if we don't have an ignition, it doesn't really matter. So if the hazard's super high and you have no ignition, no problem. But it's kind of a critical piece for our business going forward is to know where the next fire is going to start. So we can place resources close to that ignition or bring more resources in because we won't be able to manage the number of ignitions.

Speaker A: That's where Alta ML comes in.

Speaker C: Graham Erickson, senior lead machine learning developer at AltaML.

Speaker A: The idea for the AI tool actually came from Alberta Wildfire Management themselves through a general intake form. It's actually interesting with AI, what we're seeing right now is the ideas actually come from the businesses themselves. It's tech that is going to enable it. But at the end of the day, the businesses have to come up with the ideas because they are the ones who are going to need the outcome that comes from the projects. This is one of the biggest tension right now because you need businesses to come up with the ideas, but then they may not be equipped for the technology. So it's the notion of distribution, of innovation, but centralized technology and how do you make that work? It's creating a lot of questions for a customer, their operating model and how they think about it. That's one of the biggest thing I see. So it's interesting because if you're wildfire expert, you know, Wildfire and you know what you need doesn't mean you're an expert at technology. So how do you make that work? That's what we're seeing right now in every customer, every scenario.

Speaker C: We were involved in part uh, of an AI strategy contract with the government of Alberta. And as part of it we were developing PoCs like proof of concepts to show the power of AI in different departments. Um, so different departments were allowed to fill out these intake forms and basically propose, you know, we have a problem that's, that's valuable. And then we would take those and do a feasibility assessment where we break it down into some of the core pillars that are needed for AI. So basically what is the thing that you're predicting? Does the data exist to be able to learn from, um, would predictions be able to generate a value in some sort of operation? And uh, assessing those together, um, you know we got this intake form from the Wildfire Management branch, went through that process and it scored really well. So that's when we moved it into the next phase where we started working on it and uh, created kind of the foundations for what the tool is now.

Speaker A: So Altai, Mel and Alberta Wildfire worked together on a four month proof of concept projects. The data prediction, mathematics and design interface all came together very quickly. After a workshop with Alberta Wildfire duty officers, they moved to operationalize it with a soft launch in 2022.

Speaker C: Now we did a relaunch in 2023 that was more official, I guess more of a production release. And uh, had used sort of our feedback from the 2022 season to enhance it. And that's kind of led to today

Speaker A: Alta, ML and Alberta Wildfire have worked together to develop a tool that predicts where the next fire will start.

Speaker C: We've divided the province of Alberta into 10, uh, forest management areas of 10 forest areas. And those areas have administrative significance. So basically they're the areas that a duty officer um, is supposed to manage. The resources that are used, uh, this is specifically for an activity called pre suppression planning. This is about fires that haven't started yet. So every day for pre suppression planning, duty officers, uh, in the afternoon make a plan about what resources are needed for their forest area the next day. The resources could be a human cruise. Um, but where the real costs come from are heavy equipment like bulldozers and uh, aircraft that, that is where a whole lot of costs come in. Historically Welfare Management Branch had observed that there's quite a lot of money spent on pre suppression where they don't end up fighting fires. So this would be planes that idle because they're there for risk management and not actually fighting a fire. So their goal was hey, we're being too conservative with our risk assessments. We're losing a lot of money in pre suppression planning. Can we have a tool that allows us to more specifically understand the fire occurrence risk for that next day so that we can plan uh, more effectively and, and use kind of our budgets more effectively for the pre suppression uh, outlook. So we've got those 10 forest areas every day. Then a prediction is made saying the likelihood of fires uh, for the next day. And we've broken it down into morning likelihoods and afternoon likelihoods so that they can plan shifts around that as well.

Speaker A: So each of Alberta's 10 forest areas has an independent duty officer and each of those duty officers has access to this predictive tool. These officers use this tool and are uh, providing real time feedback for improvement.

Speaker B: They'll look at the tool every day and they'll see is there a likelihood of fire, uh, ignition in the morning, afternoon or no ignition probability. And they may or may not adjust their resourcing based on that. Alta ML and there's 10 wildfire management specialists in the province that they track the, I'll call it the success of the tool. How well is this tool performing over time on a daily basis? So there is a dashboard or a tool for duty officers and there's a tool for the specialists so they can look at what's going into the prediction. And if we are seeing the tool uh, potentially predicting wrong. So it's predicting you're going to get a fire tomorrow and no fire happens, you're going to get a fire tomorrow and no fire happens consistently. Then the specialist will go into the background and kind of explain why that's happening to the duty officers and they'll provide feedback to me to provide to alta mal if they're seeing.

Speaker A: Because we're dealing with wildfire and the stakes are high. ALTAML has trained this model to be very risk averse.

Speaker B: We don't want the model to predict no fire and fire happens. So we're really tuned the model to be a risk adverse model. So we understand we're going to get more false positives that way and we're okay with that because we'd rather a false positive or not. Last fire season was historic. We had extreme hazard for a large uh, majority of the fire season. So the utility of the tool at those really high extreme values is it's not as useful as you may think it could be. And the reason is at the very high hazards you need everything to respond to fires. You can't decide should I hire this helicopter or not? You're hiring every single helicopter that's available and should I hire this heavy equipment or not? You're hiring all the heavy equipment. So to really test the model on an extreme flight, fire season is probably not the best utility for it. Definitely there is some days within that fire season where you're at the moderate hazard levels that you can use the tool for deciding on helicopters. Yes, no, or heavy equipment. Yes, no.

Speaker A: This makes a lot of sense. On extreme hazard days you don't need an AI tool to validate the assumptions you're seeing and feeling. Where the tool becomes really valuable is on the days where there is a moderate hazard, where the recommendations from experts like fire modelers or M meteorologists, along with human intuition, leaves you as the duty officer feeling undecided. Again, the AI tool is here to complement, not replace. It's augmenting the people, not replacing them. A predictive AI tool could also keep this extreme hazard situation where you need all available resources from happening in the first place by helping forecast and suppress large wildfires before they even start. So what is the data and the tool using to iterate past the point of indecision?

Speaker C: It's trained on historical features. A uh, bulk of the features are typical fire weather severity, weather predictions. So this is kind of an established system in Canada called the fire Weather Index. And this is largely related to like if you're driving into a national park or something like that, you'll see a sign that says, you know, how, how light, how severe, uh, the weather is for contributing to a fire that day that is aligned with that information. So the model is working off of the core pieces that that kind of anyone has access to for weather. Um, we look at a bunch of temporal features which is kind of trying to factor in human component. Um, so this has to do with stuff like is it a weekday versus a weekend, is it a holiday weekend? Because those change the factors and likelihoods of uh, of recreational fires. Um, uh, we've got a global CO2 emission and this allows us to extrapolate uh, two more extreme fire seasons. So we're learning on years and years of data different fire patterns existed in the past. So it uses that CO2 data to extrapolate and then we've also got like kind of more immediate fire patterns. So it looks at two week rolling windows like if there were fires recently, that changes the likelihoods of fires. Um, and then there's historical uh indicators for the region specifically so that it can learn kind of um the more personal uh patterns of fire in that area. Like for example if the same severity ratings are forecasted in the north of Alberta ah as they are in the south, an experienced duty officer knows that that means very different things to fire likelihood. The north has uh, has a lot is a lot more prone to large um and out of control fires. The south is more prone to like recreational fires. So these things come out in different likelihoods. So our model has those uh has the ability to be learning kind of more personal predictions for the different regions and being able to take the weather then predictions into account with those personalizations.

Speaker A: Taking into account the human knowledge of a region is critical to an accurate modeling for this tool whose goal is to accurately predict the point of ignition. So once the tool is factored in the human component with the expensive data processing it offers prediction for the experts to consider and act on. Um, you might have heard the phrase all horsepower, no steering wheel when it comes to AI but this tool gets smarter the more you teach it and it learns over time to make better predictions.

Speaker C: A lot of the innovations lie actually is in um cloud technology for um scaling and this would be scaling not just uh, uh the solution itself but sort of how we're developing um how we're developing it. So we are using Azure Microsoft um Azure Cloud services for everything for this project basically so we use um a number of its data storage stuff uh to be collecting um the data daily when it kind of hits and to back up data we are using something called uh Azure Functions which is like a serverless deployment mechanism. We're using that to actually process data daily and make predictions daily but in an economical and responsible way. Um because Azure Functions allows us to, to not have machines sitting spun up when they'd be idling. Uh it only picks uh up resources at the times that they're needed. We use uh Azure machine learning workspaces for all of our experiment management and development. Um Azure manage endpoints for the actual predictions. Um and then we use uh Azure uh hosted uh and managed Cosmos DB for collecting our output. And then Power BI is where all of the uh front end um platform kind of stuff sits. Um so end to end it's Azure Stack. We have everything automated for deployments so that it can be replicated and tested um in development, staging and production environments. Um yeah and Microsoft directly has helped and advised uh some of that development.

Speaker B: I'm not a technical guy.

Speaker C: Hi.

Speaker B: It's on my phone it's on Power bi, whatever that is. There's a link where the duty officers hit a link on their computer. Most people don't have it on their phone. I do but uh, yeah it can be on a phone but it's not a mobile app right now. It's a web based app.

Speaker A: It's exciting to see developers working together with government entities, harnessing new technology to iterate and experiment with solutions and embracing these offerings as the way to broaden their scope for problem solving at scale, reaching people, including people who are on the tech guys where they need with a simple ux. For Graham, the latest innovation in cloud technology have had a huge impact on experimentation. The fact that you can simulate within the model and experiment before things even happen is super powerful.

Speaker C: Their cloud offerings are checking a lot of boxes like especially in the experimentation cycle. I think it's making experimenting in a way that is going to make taking your work to production and turning it into a solution uh, easier than ever before. An example of this would be the Azure Machine Learning uh, Workspace has integration uh, with a toolkit called ML Flow. And basically this lets you uh, track your experiments and register your models, um, all kind of while you're developing. And then from there there's automation for turning that into a deployed model. And this allows you to see kind of through the life cycle what data was used in what experiment, what were those results and what is the attached model to go to deployment. And it's that sort of lineage that really leans machine learning development more into like a development operations world where you've got control over the different pieces that are at uh, play and you're managing your risks throughout the experiment process. And not as an afterthought uh before these sort of breakthroughs. The problem would be that someone who is uh, really good at the machine learning technology would be very rapidly uh, typing up experiments, uh, they'd be isolated in a notebook and there wouldn't be a clear road for okay, how do we actually turn this into a solution? And these sort of toolkits are allowing us to manage that experimentation process in a similar way you would software development, um, which is important for being able to take these uh, projects, these experimental projects and actually turn them into stuff people are going to use.

Speaker A: Implementing a new tool, especially an AI predictive tool has taken some adjustment.

Speaker B: One of the biggest challenge and I think the Alta ML and the government of Alberta team realized from the get go on this was going to be the change management. It's a very well it's completely new to Wildfire Alberta to be using an artificial intelligence tool. The one thing that we did was, uh, altaml, they developed, uh, duty officer Personas. So I think they had four or five different Personas. A brand new duty officer, an old, old grizzled duty officer, probably someone like me. And then kind of in between duty officers to kind of figure out what would these, uh, people and how would they react using the tool. So we had duty officer workshop where we ran the tool and we ran them through days and we asked them to make what's called a pre suppression plan, where they hire resources, put resources in certain work hours and, uh, come up with a solution for what their plan is the next day. And what we found was a lot of the older folks, I'll say, they're like, well, I knew that was going to happen and that's perfect. In my mind, that's great. Well, if you knew it was going to happen, then the tool's actually working pretty well. So it is, uh, lining up with their intuition for what's going to happen.

Speaker A: As we've heard in other episodes, this notion of Persona is so critical for success. You have to really understand who you are targeting, who you're trying to help, what outcome you're trying to drive. And when you do it, you get the best results. And this is also how you get the feedback loop with the humans giving you feedback, feeding the model, and then the model gets better. Developing Persona is also going to help you gauge how people will feel about using the AI tool. Identifying people's fears is really crucial to an effective change management. They're actually doing everything they need to do. It's actually pretty impressive. And the feedback loop is so great. They're really listening to their users and designing around users and needs. And that is what makes the tech better and relevant. Ed collects this feedback from the other duty officers and brings it to Graham and the Alta ML team. This is also helpful for accelerating the onboarding phase and helping the newbies.

Speaker C: They found that the more experienced duty officers, it sort of confirmed things they already knew. And for the more rookie duty officers, it came with in with more novel insights. This was actually validating because that meant that we could position this tool as, as a way to sort of enhance and, uh, stabilize sort of expert knowledge, like distribute, um, expert knowledge. And there's a lot of fields that deal with this problem of, uh, what do you do when experts are retiring? If you've got an aging sort of population in this expert field, um, how do you bring on new people. How do you democratize that knowledge? How do you get them involved? Um, this seems to be actually picking up on some of those intuitions that duty officers would have if they worked for like 20 years doing this. Um, but it would let a more junior officer pick up those intuitions much quicker.

Speaker A: Despite the natural change management process, REM actually thinks the Wildfire Management Forestry team is a prime group for experimenting and adopting these new technologies.

Speaker C: They've actually been, and maybe I'm stereotyping, kind of forestry as analytics people right now, but they've been working with uh, analytics, statistical solutions for a long time. And uh, they're very direct and they kind of believe in what they're working on. So they're, they're great for adopting these, these technologies. Right. They, they uh, don't have a lot of patience for kind of the marketing fluff that goes on. They're direct to the point. If something has value, they'll use it. If it doesn't, they won't. Um, and they're used to working with statistical measures. Like the FWI system I mentioned was developed in like the 70s, I think. So shifting to AI, but delivering them with predictions and uh, likelihoods and that sort of thing. It's very natural fit for them and isn't too much of uh, a heavy change management process.

Speaker A: At the end of the day with change management. And here we're seeing it, it's all about the value you bring to people. If it's fluff, they won't take it. If it's concrete and it's really helping their day to day, they'll adopt it. They'll give feedback, There'll be friction, but they'll adopt it. Even with a team that is more adaptable, a busy fire season means more government resources and more hires. But there's still an experience gap.

Speaker B: So we have a ton of new staff, uh, that they don't have the, I guess the experiential base built up to know what I call it a bad fire day. I walk outside, I live on an acreage, so I live in the forest. And I never walk on my path to my car. Like my paved path. I always walk through my grass. If I walk through my grass and my shoes stay dry at 6 in the morning when I leave, it's going to be a bad fire day because the humidity has not come up overnight. So that's kind of my first indication. So it's not a computer program, it's just grass. So just those little things we're Trying to bridge, I call it bridging the knowledge gap. And then younger duty officers or newer duty officers or inexperienced duty officers can, can look at the tool and make a decision based on the tool and hopefully build that gut feel as well. So they're going to see the tools predicting it. They'll look outside at the weather and they'll understand the relationships between the, between the two. When I started, lots of the time when we thought of ignition we'd look at very windy days. That's a bad ignition day for Alberta. We get a lot of power line fires. So if it's windy ignitions or if it's a long weekend, humid ignitions, beyond that it was really hard to know when we were going to get the human ignition. So this is kind of uh, just a way. It's another tool in the toolbox to assess. Okay, we have bad weather conditions, our fuels are dry and we're getting ignition. Let's be ultra prepared today for Ed.

Speaker A: This AI tools has shown great promise thanks to the simple fact that people are using it.

Speaker B: So it doesn't sound like a big success but just getting folks uh, to look at the tool is a huge success in my mind that they're looking at the tool. Um, some duty officers have reported that they've changed um, change some of their pre suppression plans based on it especially in those mid range hazards. So should I hire a helicopter, yes or no? Uh, and the tool said no fire so they didn't hire a helicopter. So not hiring a helicopter saves the government wolverine tens of thousands of dollars. So you multiply that by 10 forest areas by multiple helicopters, the savings can be in the tens of millions of dollars over a fire season.

Speaker A: Throughout the two seasons of our conversation on pivotal, we've discussed the need to overcome the push factor and really generate a uh, pull towards the new tech. And in this case it sounds like they're already there. Duty officers are reaching for this tech, referencing the tech and the feedback is being fed back to Graham and the developers at Alta ML so they can refine, iterate and experiment. All is happening while the tool is in use. As Alberta Wildfire and Alta ML look towards fire season 2024 there are a few items that are top of mind.

Speaker B: One of the challenges we found is actually we didn't find a source that saves their forecast. So the forecast for Environment Canada, the multi day forecast, they save the actual weather but they don't save the forecasted weather. They save the actuals for record keeping and stuff but the actual forecast that they made isn't saved and that's what we want to base a prediction tool on. So some of those interesting little data issues become a real uh, challenge moving forward. So obviously like we'll, we're trying to attain weather forecast information, we save all our weather forecasts in the province but we just forecast out to tomorrow. We don't forecast out multi days. So our, our meteorologists for Environment Canada, we're working with them to try and get those longer term forecasts saved so that we can test the tool and uh, train the tool on forecasts. We don't want to train the tool on actuals because we need to predict into the future. We don't need to predict what's happening right now.

Speaker C: If we could at least get to three days out that would be much more flexible. So duty officers do make predictions for one day out. But what's tricky is navigating at uh, things like planes. Like if you have one day that says high risk and the next day says low and then the next day says high, that's way too much of a yo yo for you to be throwing planes around the province. So that three day sort of average forecast is what we'll probably need to move to for them to pull it into planning kind of to its highest uh, degree of success.

Speaker A: It is surprising to me that forecasted weather data isn't recorded. And I'm reminded to the fact that 90% of the world's data was actually created in the last two years. And every two years the volume of data across the world doubles in size. So as we look ahead with that in mind, I'm confident weather forecasts, not just the actuals, will be captured so that the data can be harnessed soon. Aside from capturing weather forecasts to help their model, Alberta Wildfire and altaml are working on building in more granularity.

Speaker C: We know that the next steps for this model are to make predictions more spatially granular. The 10 regions are massive and while this aligns with their current planning uh, philosophy, this does still have some open ended questions. For example there's some, there's some places in the north, there's some regions that could basically be cut into two areas but they have very different sizes. They could be cut into an area where a lot of people live and they could be cut into an area where not that many people live. And ah, managing those two kind of situations is very different and the risks that you would kind of take into account are very different.

Speaker A: Alta ML is exploring breaking up this large area into areas that are defined by something more significant than just administration.

Speaker C: Another thing would be like power lines. Like we would want to be able to overlay pretty nice maps of uh, power line and power line assets. I know there's different companies now looking at say vegetation and power line risks, like for um, power companies to be doing their own vegetation management. So hopefully there'll be breakthroughs kind of from that uh, world that we could, we could borrow.

Speaker A: It's impressive to see how they're already thinking about what's next. They're already so refined they really get what the model is and how you need to think about data sets for more accurate predictions. They know what they're doing and they're already onto the next thing. They're thinking very pragmatically about how to match operations with data. A more specially granular tool will have far reaching downstream impact. More specific ignition prediction will lead to more accurate fire growth models which will all lead to faster response time to contain and prevent wildfires. In the end, There are still 10 duty officers at 10 forest locations throughout Alberta province looking at all the information needed to make a decision to send or not to send critical resources to a, uh, location before a fire starts. Their decision will always center around trust, trusting the tools, the data and their experience.

Speaker B: So we kind of already are, we spread our resources out across the entire province to make sure that we have kind of average coverage, I guess. But once we incorporate the tool more at a granular level, it won't be at an average level. It will be based on values and, and potential for ignition and impact, I'll say. So then we shift. So instead of having three crews here and three crews there, we might have five crews here and just one crew in the other location where there's less likelihood of fires. So the risk of that is the trust in the tool. So that becomes when you trust the tool's outcome or not, if you have low trust, you're likely going to just keep your crews average. If you have high trust, you'll move your crews based on the tool's outcome. We're not there yet, but that's where we want to get is putting people in places where ignition's happening right now. We move crews around to common locations. Like we'll always put them at this location if there's hazard out in this corner of the map. But we might actually put them right next to um, an ignition that hasn't happened. So they're there in the morning and then it's predicting an ignition to happen in close proximity in the afternoon, they're already there waiting for the ignition. So it is a little strange to do that and risky. But, uh, it's more about trust in the tool and just change management and being okay with making the wrong decision too. So if all the tools are aligning and supporting the decision of the duty officer and they make that decision, as long as the organization understands, uh, why all those decisions were made and backs up the individual with the decision making,

Speaker A: looking forward, there is a huge potential for this tool to extend beyond Alberta.

Speaker C: We do think like it would be a natural fit and pretty easy fit for any other Canadian jurisdiction. Um, but just because that's the FWI system, but, uh, it's not restricted to Canadian. It would need some modification and customization. But that's definitely, I mean that's what we do is custom solutions. So, um, yeah, we would love to uh, bring it to other jurisdictions if people are interested.

Speaker A: Graham reflects on the power of working with the public sector.

Speaker C: I think government sometimes gets a bad rap of being kind of behind on the times or slow to adopt. Um, and I'm sure there are ways where there is a slowness to adopt due to due diligence and that sort of thing. Um, but the welfare management branch is not slow to adopt. It's fairly cutting edge. Um, and they're working on problems that do have this kind of net benefit for all Albertans and if we can bring this to other areas, like large groups of people. So I believe that that AI in public sector makes an awful lot of sense because this is a way to distribute the value of AI to, uh, entrepreneurs that have a societal or even social benefit. Um, for us as a people, it is not, uh, solely for kind of individual commercial benefit. We're working in a domain where the value and the benefit goes, has kind of large impact for uh, our societies. Um, and I think that applies to an awful lot of, uh, public sector problems. And so there's real value in being able to work on public sector problems with that.

Speaker A: Regardless, last year's devastating wildfires in Canada are just one example of the impacts of climate change. We need to use every tool at our disposal to face this crisis, including AI. In our previous episode with Biodiversity, it was all about a, uh, government trying to understand the composition of land. And we see exciting projects emerge when the public sector partners with developers working with the latest tech. AI offers an amazing opportunity to scale those efforts. This should be something that every government wants to leverage and to use, whether it's climate or their societal, uh, agendas. That they may have tech can really help. And this is a fantastic example of it in this case. It's amazing to see how Alberta has structured their innovation. They prioritize, they score, they implement. And they have a dedicated and experienced duty officer and a sharp software engineering team together, uh, creating and refining a tool that can save money, wildlife, and perhaps even lives. It's a great example and a fantastic story, and I hope we get to see many, many more of those. So this is the season two finale. What an exciting season. I mean, we started the year thinking about how AI can really unearth new possibilities for the world. And we also started by saying we know there is a lot of anxiety in the season system and is it good to use AI or is it bad? And I think what we've learned is there is no good and bad. There is really identifying an outcome that you're looking to achieve and then using the technology on behalf of whatever it is that you're trying to do. And when you do it, make sure you complement the human. It's not about replacing the human. And I think we saw that consistently, the human needs to be in a loop and we need to make sure that all the drudgery work gets away and they can focus on the most important things. And we've seen that over and over again. And I think we're going to see more and more of this. This is very exciting. Thank you for listening to Pivotal. I'd love to hear your story and your pivotal moment, so don't hesitate to follow me and share on LinkedIn. Audience information is also available in the show notes. Our show is produced by LARGE Media. That's LARJ Media. Special thanks to Lin Yang and, uh, our partners at WE Communications.