Name: Bridging the Gap in Corporate Water Stewardship: The Role of AI
Uploaded: 2025-02-20T17:16:09.845Z
Duration: 59 min
Description: Bridging the Gap in Corporate Water Stewardship: The Role of AI

Transcript for "Bridging the Gap in Corporate Water Stewardship: The Role of AI": Hello. Hello. Hey, Super excited to discover you here. How are you doing? I'm doing great, Jose. How are you doing? It was, it's it's I'm also super excited for this webinar. I think it's, it's a topic very close to my heart. Thank you for inviting me. 100%. It's a pleasure. How are you doing, Ben? Doing great, Jose. Thanks for having me today. I'm excited about this. Me me as well. Super excited to have you both here. So we can wait. I see that people are joining. So maybe we can wait a couple of a couple of minutes, and then we can start with this amazing webinar that we put together, which is about reaching the gap in corporate water stewardship, the role of AI. Right? So we have, some amazing speakers today, that we're gonna go to introductions in a second. But maybe while people are joining, we can start with some, housekeeping items as we always do. Right? So for the audience, as we always do, we are gonna record the webinar. Right? And the webinar recording is gonna be published to the Wireban website. So you're gonna be able to access that after we finish. However, we won't be sharing the slides. The duration of the webinar is gonna be about fifty minutes to sixty minutes, I would say. Just the cameras and the microphones of the speakers, in this case, Keban, Ben and me are gonna be available this time. And then in terms of q and a, we really expect your questions. So feel free to put your questions on the chat at any point or there's a q and a feature on the platform. You can put it there, and then we have some time at the end of the webinar to answer those questions. Or, you know, Ben and and Kevin can decide to answer that on the spot, whatever they want. Right? And last but not least, we are constantly improving the series of webinars that we are doing. So we'd love to have your feedback. You're gonna send this survey after we finish. We'd love to have your ideas for new topics that you would like for us to cover, new people to invite, anything that you believe that we can improve. We are super looking forward for your feedback there. Alright. So maybe with that, we're gonna start with some of the amazing interactions. Ben, it's an honor to have you here. Thank you so much for joining. We'd love to have your to start with your interaction. K. Thank you. Could could could you could you do like a quick interaction of yourself, please? Oh, sorry. I'm not sure who's confusing you. There you go. Off to a great start. Yeah. So Ben Zajac, I am, the chair of the Department of Planetary Sciences at Johns Hopkins University. And I do a lot of work on climate analysis that involves really understanding and find resolution, how climate variability and extreme events emerge and how they have impacts on water systems and other social goods that are a value to us all. That's that's amazing. And, the this this I would say, like, this webinar is really close, I would say, to my heart. Right? One's because, obviously, I I love water, and I'm really, really interested in about AI, but also to have the opportunity of doing this with you. Right? Like, Ben has been, like, one of the, initial kind of, like, supporters of of the mission of Waterfin, and, I'm really grateful for that. So, like, it's a pleasure to have you here, Ben. And, Hemant, it was all for you if you can continue with your amazing interaction. Super excited to have you here as well. Thank you so much, Jose. Thanks, Belk, for joining. It's a it's an honor for me as well, alongside Jose, to be present in the same webinar as you. So, as Jose asked, like, hello, everyone. I'm, Heman Servia. I'm working as senior water scientist and science leader at Water Plan. I have spent over thirteen years in the water resources management field, with the past four years dedicated to developing and advancing AI driven solutions at WaterPlan. Based in The Netherlands, my passion lies in understanding and mitigating, water related risks, particularly in the face of climate change. And today, I'm very excited to be part of this discussion on how, AI driven scientific data can, drive better water stewardship decisions and shape public disclosures. It's great to be here again with, Jose, Ben, and a lovely audience like yourselves. Looking forward to a good discussion. Thank you, Jose. Over to you. Thank you, Herman. That's great. Yeah. And let me, like, just do my my interactions. I'm Jose. I'm the CEO and cofounder of WaterPlant. WaterPlant is software to help companies that use water in the production process supply chain to quantify and mitigate water risk. So a big focus on water risk and and everything related to to water. My background is in software engineering. I have been working in the, software and and water space for several years now. I'm based out of San Francisco, and I'm super, super excited to share this, amazing panel with, Ben and, Hemant. So maybe we can start, off the conversation with the role of AI in water stewardship. Right? So maybe, Ben, we'd love to have your perspective first. Like, if we start by introducing can we start by introducing, AI tools to the wider world? Thinking from the perspective of scalability managers, how can a wider scientist contribute when looking into wider risk? I would love to have your perspective then and then, Keman, also your comments. Yeah. Great. Thanks. I'm I'm gonna be very interested to hear him on spots on this as well. I think anytime we start a conversation about AI, it's important to do some level setting. Right? What what do we all mean by AI? And so I would say that for this conversation, I'm I'm entering this with a very general definition. Right? So that artificial intelligence is that field of science that takes on, you know, how do you build computers and machines that we say can reason, can learn, can act in a way that you would normally require human intelligence, to to act. And so that involves data often that at a scale that exceeds what humans can analyze, on their own in a reasonable amount of time. So just to, again, to get the term the terms here. So the way I'm gonna use them, machine learning then would be a subset of AI. Right? And then deep learning or natural language processing, which we hear so much about these days are subsets of machine learning that are really talking about ways of analyzing complex and large datasets. So in water resource analysis then, I would say machine learning algorithms have a pretty clear and well demonstrated role to play. So in a formal sense, even simple analysis like various kinds of regression that we might use to to analyze data, make a prediction, that's a kind of machine learning. And then as we get to more advanced algorithms, getting to different forms of neural networks and and deep learning algorithms. That's just a way to really take that same logic but apply it to larger and messier datasets with the same general objective of of making better predictions. And now the ability to create, AIs that then are taking some of that machine learning algorithm, integrating sense data, right, from some sensors within a water system, maybe some remotely sense data, and doing that in real time, that's adding additional power. And so that's all in that world of like really, really powerful data analytics. Now as we move from those kind of technically difficult, but I would say easily framed problems, like how do we make a better prediction, to a broader intelligence problem like adaptive learning or optimization of solutions, multi criteria problems. This is where we enter an area where artificial intelligence algorithms, I think we need to think of them as partners in our decision making rather than just, you know, a sledgehammer. And so this gets into this world of questions of like when is it appropriate for an AI to dictate what the right decision is? And that's interesting and then it's kind of becoming almost an urgent philosophical questions for us. But, I think it's not right now the central narrative for water stewardship. I think it's something that we're moving towards. What is I think central now is emerging as central in the worlds that I operate in is thinking about how adaptive learning or optimization AIs can make sense of messy and uncertain data in a way that allows a human operator, a human decision maker, to have better information at hand. And so in any of these applications, I think we have to think of AI as a powerful tool, but also a powerful tool that depends on the quality of the input data and on the interpretation, that human operators are able to make. So just like in any field, I think the use of AI in water resources and water stewardship, still very much requires human expertise and judgment, both in how you train the AI and how you use the information. It's really I love the way you put it, Ben, when you say like AI as partners in decision making. Right? I believe that resonates like a lot. Right? Like, how we can use these tools to enhance the research and the work that we're doing. So I believe, like, that framework makes a lot of sense. Right? And the way you put it partners of decision making, For sure, it's, I I totally resonate with that. Eman, would you like to have your perspective as well? Yeah. I think, Jose, like, what, Ben explained on the role of AI as general and what ex how you explain AI in the field of water stewardship, I think I have, nothing to add to it. But I think I can take the other part of the the question. I think the broader question was, what's the role of water scientists in whole in all this schema? Right? So I think, like, water scientists can offer a very unique and invaluable perspective to analyzing water risks as whole because they have a they have a ability to combine their deep scientific expertise with practical insights, you know. And this helps businesses address complex challenges that are on their doorstep now, such as water scarcity, flooding. And I think, both of us have mentioned about climate variability and their ability to use, such hydrological tools and clubbing it with climate data. It ensures a holistic assessment of risks that considers not only, the immediate threats, but also the root causes and cascading effects. Right? For example, like, I I have a very, like, a very common example that we we usually discuss is, like, while flooding might appear as a local issue, a water scientist can uncover some broader drivers like, upstream, deforestation or unplanned urbanization, which actually can amplify the risk. Right? And this needs to be, tracked soon to enable a more efficient and sustainable mitigation strategy. And equally important is their ability to, to to bridge the gap between local and global data. I think we will be discussing this in detail, but this is very important because many businesses these days, as we see, as we deal on every day, they rely on global datasets for for their risk assessments. But this data actually lacks the granularity that is needed for a site specific insight. So, for example, a facility which has been flagged or maybe flagged as a high level risk in terms of flooding, it may be found safer based on local topography. So this nuanced approach, ensures that resources are allocated efficiently and risk are addressed, accurately, actually. So I would say, like, to summarize, by combining the water scientist technical insights with the sustainability managers, because I think those are the people, like, who actually implement those things. Their understanding of operational priorities for the organization can align their risk mitigation efforts. And this integrated approach ensures, which is very important, that the water stewardship strategies are not only scientifically sound, but they are operationally viable as well, which is the later part is always sometimes missing. And with this club approach, it can be ensured. That makes a lot of sense, command. And I believe that you put it, like, really, really nicely. Right? But you mentioned a couple of examples when you were talking. Right? So maybe we can switch for that. We'd love for you to talk a little bit, like, about, like, AI tools for what are risk assessments so we can, bring it down a little bit more to to the day to day kind of, like, use case. So maybe we can start with you, and then then we'd love to have your perspective as well. Could you share some examples of how we can use AI tools to reach the gap, when we don't have water scientists in our team. Right? So that they discovered stuff. Right? We there's a lot of organizations. It's really, really important from my perspective to have that scientist perspective when we're doing wide risk analysis and water stewardship gap plans, but not everyone can afford or they they can have, like, a water scientist in the team. Right? So give us a couple of examples how we believe we can leverage these tools to bridge that gap and, yeah, we'd love to hear your your perspective, Camand, and then Ben as well. I think that's a very interesting question, Jose. I think, as you said, we have seen this scenario quite often and even happening with very big corporations. Right? But I suppose, like, in today's world, there is absolutely nothing to be worried about. Even when water scientists are not available in a given team, I think scientific AI solutions I I again say scientific AI solutions guided and developed with the help of science can help address some of these very complex challenges. Right? And these AI driven tools can be designed to not only, fill these expert gaps, but they can be also scientifically robust and which can be, adhering to the latest latest research and principles to ensure that the the outputs are credible and accurate. I would give a very small example. I think, the things that you see on your screen for the audience. So this is a a research that we did in Japan and US. So we developed country scale deep learning models, to deep to predict stream flows in some pseudo on gauged basins. And this is a area which is traditionally hindered by data limitations and data resources and accessibility to data. So by combining the metrological inputs, from, different, sources, which were local sources, which were remotely sent sources, and combining them with the catchment characteristics, we offered reliable streamflow predictions across diverse regions. And you will be surprised to know that with over 90% of our validation results showing very high accuracy, this approach provided actionable insights for managing water resources in, these data scarce regions. And just for everybody's, reference, we also presented this research at the twenty twenty four European Geosciences Union, which we usually call EGU, in Vienna. And you can find more details on water plants, resource, section on on the website. And even as we speak, it's it's interesting to know that even as we speak, several of these models are still running, actively running on our cloud, continuously generating these predictions to support water management decisions in one or the other regions in the world. So this is, like, how it is evolving, how dynamic it is, how near real time it is. That that is a great example, Hemant. I'm super happy that you shared that. By the way, I shared on the on the chat that analysis or that report that we that is on our website that I believe is quite quite exciting. Yeah. For me, the the possibility of using data from a specific base to train this AI, model, you know, neural network and extrapolate this matter for that in place, you know, with the level of accuracy that we achieved, you know, specifically for Cisco. That's amazing. Right? And I understand that this kind of usage of AI is is kind of new in hydrology. Right? So Ben will not have your your perspective as well. Yeah. I agree. That's a great example. And I I think you said great example in both the techniques that were used and also in the model of how it was developed. Right? The the management model where maybe water experts are involved in a conversation about defining needs, implementing a system, but then the system can operate. And so you don't not necessarily need full time water experts checking on all the time. Just to add a couple examples, you know, maybe complimentary. One thing that we know these algorithms are very, very good at is detecting anomalous behavior. Right? Like they they can figure out when something's looks different, even when the data are complex, messy, large. And so having system if your concern is active management of something like water quality, something like leak detection, or something like inefficiency within the system, that's where we're getting something, a system like this set up for operational use can then really be be, of value to the team and risk managers and operations facilities folks, even in the absence of, you know, active engagement with with the water experts. So I think that's the place that we're seeing a lot of a lot of value. And similarly, not only can they find when something's wrong, but just kind of classifying, like understanding patterns of behavior and patterns of, of how a system is operating. And so one challenge that we often have, you know, in the data sciences world independent of water, is just how do you deal when you've got deal with situations where you've got many correlated processes going on. You're trying to understand what's driving what, what's similar, what's different. And so implementing a system, again, probably in collaboration with some subject matter experts to figure out exactly, okay, let's pose a question, let's identify the right tools. But something that's gonna be, both doing initial analysis and then perhaps ongoing monitoring of different types of processes and types of behavior. So like in a water management system that could obviously have something to do with what's happening with different times of year or at different facilities with respect to water usage and water availability. It's a really good area where that doesn't require a lot of hands on partnership. And the last one I was mentioned is in terms of if you're in a situation where you want to design measurement systems. Right? You know that you wanna be keeping track of something with respect to your water quality, your water amount indicators, or water usage indicators. AI systems are great saying, well, here are the places in your system that you need to have better measurement. Here's where we have uncertainty, because we don't have the data we need. And that can maybe help you decide, okay, I'm gonna implement some more monitoring in order to enhance my ability to, to steward the resource. I agree that those are great examples, Ben. Like, it's really interesting to think about, like, the first one that you mentioned, like, the one about, like, hey. Now we can really understand, like, any anomaly in, like, large, I would say, like, sources of data. Right? So, like, the use case of peak detection, water quality, anomalies, etcetera. I think that that makes a lot of sense. And you can see, for example, just thinking about LLMs, how the context windows of these larger group models is going, like, bigger and larger and larger and larger. Right? So So we like, that opportunity for using for that use case, I believe that it makes a lot of sense. And and it's really interesting that you're already, like, thinking and using kind of these tools. Right? So let's shift now a little bit to the role of technology. Right? So, maybe, Ben, you can start with this one and then, Kevan, I would love to have your comments as well. That that is my question is, in what ways AI tools improve water risk assessments and a scenario model? Right? Like, thinking specifically about ways to assessing, you know, water risk and also to do a scenario model. Right? And Ben, we'd love to have your your perspective here. Yeah. Thanks. And so I I think probably my in my research world and a lot of the kind of translational research we're doing, largely with municipalities, I would say, and large utilities more than small companies, but I think it applies. This is a real focus for us, particularly the scenario side of it. So one thing is just a practical element. So in a world of uncertain change, right? So we know we've got climate change, of course, but also other kinds of environmental change going on, other developments in technology, where it's difficult to foresee the future, but you might be making investments or trying to at least make plans for investment over longer time horizons. How do you do that? And a lot of these water management modeling scenario tools are very computationally intensive. And that's a real problem because when you have deep uncertainty, if you can only ask a couple questions like, well, what if this happens or what if the second or third thing happens, you're really under sampling the space and you really do not have a full perspective of what your downside risks are or your upside opportunities. And so if you wanna be making decisions that are robust, kind of this whole world of robust decision making, you need lots and lots and lots of potential storylines, lots of scenarios. And computationally, that's difficult. Unless you have AI that you can then train and make these surrogate models, right? Where maybe you start out by looking at a really sophisticated physics based model and then you say, Okay, well, now that I can train an AI to that, I can then propagate huge numbers of potential scenarios, really explore the decision space. And that's gonna allow me to have a better sense of my uncertainty and be able to make decisions based on a robust view of what's happening in the future and not just the couple of scenarios that I was able to run through my model. So that that use of AI as a computational efficiency for surrogate models, I think, is really changing the way we're able to apply our scientific understanding to planning. And the second kind of related thing to that that I've found recently has been interesting to me. I've learned this, others might have already known it. But so we're doing a lot of this work now looking at things like flood protection and the future scenarios, the coastal floods and inland floods. And part of what I thought would be like, well, it really matters what assumptions we make about future climate. So are we assuming high emission scenario in which climate goes very rapid and large scale change or a more stable solution and you wanna train your model and do your optimization based on one of those potential futures. But then what we found was that when you train in an adaptive learning algorithm scenario, so basically an algorithm that is learning to create decision rules rather than actual implementation choices. In our experience, they were very robust to what you train them on. So we could take this model and say, okay, let's assume that climate change isn't gonna be so bad. And so we trained it and we said, okay, well now let's hit it with a surprise of of rapid climate change and watch it fail. And it didn't fail because it was able to use the adaptive learning algorithm. Now I'm not gonna say that's always true, right? You always wanna say this, but in this case, it was like, wow. You could actually have an AI that creates an adaptive learning process that says as if we hold to this over the coming decades and learn as new data comes in and say, oh, wow. Climate is changing faster than we had anticipated. The learning modules are able to adjust to that if they were properly, trained at the beginning. And so I think that's a really important area, that again, the specific specific kind of AI engaged in multi criteria adaptive learning is gonna help us make much better decisions. That is super, super interesting, like both cases, I would say that you mentioned for the scenario modeling example that you mentioned. Right? Like, I clearly see, like, the opportunity, right, of not just, like, creating, like, two scenarios or three scenarios, but doing really, really explore, like, different scenarios. And I believe that's really, really exciting, I would say. How how did you have like a like an example of any research that you did, I would say, when you did that? I would say like like did like several scenarios of the same? It's super super cool use. Right? Yeah. Yeah. And so pretty much everything we do well, not everything. Everything we do in that world now we're doing. And so this whole world of robust decision making kind of demands that large sampling of the space, right? And so we are working, for example, on climate adaptation plans with the city of Baltimore, where my university is located. And so in that instance, what we do is we work with communities, with city officials, with industry to say, okay, what are the metrics of success, right? What will we be trying to achieve? And so then once you define those metrics and you say, well, what are the levers? What are the kinds of things we might be willing to invest in to avoid it, to avoid the impacts that we're trying to prevent? And then you say, Okay, now we've kind of built this framework, this decision making framework, and then we just hit it with these large, large number of scenarios. And where the AI comes in there is that we say, okay, let's say that what you wanna do is limit your flood risk along a particular stream in the city and you have these various options for putting in green infrastructure, gray infrastructure, blue infrastructure to do this. What's the response? And that's actually really tricky, right? Like you say, okay, I can put in different interventions in different places in a watershed and somehow under different scenarios of climate change or other impacts, they're gonna be a response for the metrics I care about. That response function, if I were gonna model that in a spatially explicit physically based model every time, it would crush me. Like there's no way I could do that. But if I could train an emulator or a surrogate model, that is able to mimic that behavior, then I can do it tens of thousands of times, right, in a really small space. So that that that's one way in which we're using the surrogate models. Wow. That that's an amazing amazing example. Maybe we can do that in the future just like what you're talking about that. Right? That's a really fun part. Idea. Really, really exciting. Kehman, we'd love to have your your perspective as well. Right? Like, how you behave, like or have you seen AI tools that improve scenario modeling or what are these assessments? Yeah. I think, like Jose, Ben already explained many things which are, like, at the forefront. I I would like to take a step back. And I think, like, at the first place where I AI can really help us enhance water risk assessments and scenario modeling when we say I think Ben mentioned that a lot of data goes into it. So so, actually, by improving the accuracy, scalability, and integration of different datasets that goes into it is the first place where, like, you know, AI can significantly contribute. Advanced data analytics and machine learning, these allows businesses to move beyond traditional methods, providing dynamic and real time insights, which are actually, should be there for the complex problems of what are risks. And it can help bridge gaps between this global and local datasets as we are again and again highlighting, which because refining global risk evaluations requires localized analysis. And this approach helps us to prevent any false negatives, false positives, and, therefore, it it ensures that the resources are allocated efficiently. And, additionally, I think, when we are talking about the sustainability managers, it can help them align, streamline the data collection and analysis process overall and automating these complex processes because, otherwise, they spend a lot of time or I would say, in a way, they they spend 80% of their time in collecting and analyzing the data, but then it would free them to analyze on or to focus on main, focus regions, which should be analyzing the strategies and making the stakeholders engagement. And so that is why I feel like AI also empowers, like, scenario modeling by simulating future conditions under different climate conditions, land use conditions, and operational scenarios. So as discussed in the case of Japan that Japan and US that we discussed, so AI based streamflow models can be used to predict streamflows or flooding impacts, allowing sustainability managers, which are our target audience or, which which should be, like, you know, be empowered, to assess the resilience of these facilities under different risk scenarios as Ben mentioned. So I think, like, they should be more resilient against any extreme weather events, any supply chain disruptions, or any regulatory changes, which can be done with the help of these AI tools that we're talking about. That that makes a lot of sense, Kamal, that I totally resonate with what you're saying. Right? For example, what what was we're working on, I don't think that gather local data parts. So we know that doing, like, a a local accurate granular data for each one of the locations in which, like, you know, like, people have factories or, you know, like, any production site or any supplier is extremely important because we know that, you know, like, water issues are really local issues. Right? So to be able to have these, really good models, I would say, like, getting a lot of different information from, like, local measurements, you know, like, the scientific reports from the area, etcetera, and to be able to select the right evidences. Right? Cross check those evidences with other sources of information such as road sensing, hydrologic modeling, and be able to not just, like, filter those, right, and select the the most relevant ones, but also to do the scoring. That is quite amazing. So we are seeing, like, the the amazing, potential of doing that. And, obviously, we have, like, internal, you know, like, data retention tool in which we have, like, a water expert kind of, like, validating the data because we know that we want to ensure a specific level of what we're seeing. Right? But I'm seeing how these models are, you know, like, really, really improving, like, almost every week, I would say. Right? Like or not almost, like, every week. Right? So I I really see, like, 2025, like, like, potentially an inflection point, I would say. And and I believe that it's really exciting to work in water management at this time because this is, like, a really concrete and useful, application of LLM's AI, and that is, like, quite, quite exciting. Right? So so yeah. We have a before we go to the next part of the conversation, we have a poll that we just line launched for all the audience. We'd love to have your your perspective as well. So the question is, how is your company currently leveraging AI for water risk management? We'd love to have your your perspective there as well. Right? So shifting a little bit to talking about, AI tools specifically for water risk mitigation. Right? I have a couple of questions that I would love to have your perspective then and and Eman. Right? So my question is if you can provide some examples of how AI has been helpful for predict and mitigate specifically, mitigating water risk in real world projects. Right? And do you think that it can help businesses optimize the water reuse efficiency and be better prepared for for future risk scenarios? So we have already spoken it and, like, I talked a little bit about this event, but we'd love maybe if you can dive a little bit deeper, specifically thinking about, like, mitigate mitigating water risk in a real real world. Right? So we'd love to have your your perspective there, and then, Ken, I would love to hear you out as well. Yeah. Thanks. I think, you know, from from my view of the field, some of the most exciting kind of moving into actual operations applications have to do with forecasting water use and demand as a function of weather, human behavior patterns, other factors. And there are lots of examples of this. I mean, Singapore has started working on this in the municipal sense, and also there are some irrigation examples. I won't dig into those because I I I might defer that to Hemant because I know that it seems a lot of pool work in that space. So I won't dig into those specific examples. Instead, maybe I'll try to complement with some things that are maybe at this point, it's real world, but it's still pretty high level. And what I'm excited about in the coming years is how it's gonna get more specific. And that refers back to your last question, Jose. Some of these issues of scenario analysis. And so if you look at larger river basins, you know, so in The United States here, the Colorado River Basin has been an issue forever and it still is on the radio this morning or people talking about, you know, some of the states, the river runs across several states and then into Mexico and now apparently some of the states are lawyering up getting ready for the next the next round of negotiations because it's always so controversial. And so that's an area where a piece of really big actors at play in long term negotiations, this use of multi objective robust decision making, with AI kind of under the hood, you know, as we discussed to allow you to explore a lot of scenario space is starting to see real application. And I think it's entering the sector from those kind of large institution long term decision making processes. And that's where you need to get multiple perspectives around the table, train the AI for these diverse objectives, iterate with the human users, right, to make sure that people are comfortable with the way in which the algorithm is interpreting the metrics and coming up with various ways forward. And then the algorithm doesn't give you one answer, it gives you a number of potential pathways you could follow, each of which is vetted with certain robustness characteristics. And then the human users can say, well, okay, at least we can agree upon the data, we can agree upon the analysis. Now let's argue about which one to do. Right? But that's a huge, huge place to get to. And sometimes maybe the arguments maybe maybe it doesn't solve the argument, maybe sometimes it does resolve some because, people had fears and uncertainties that they now have more confidence that they don't have to worry about certain outcomes. And so I've seen that become really useful in those exercises. Outside the water sector, national security negotiations are already using this a lot, right? And so I think there's a good heritage being established there. And what I'm excited about is as we talk about individual companies, smaller communities, how do we then learn from what's been what's now becoming operational as of that large scale and bring it down, to a local facility, to a local community, to a local group of actors and allow them to do this. And I think that as the AI, as you're saying, is advancing every week, as is our ability to work in intelligent ways with it. So as we have informed teams and good algorithms, this could become a standard for all kinds of decision making, not just you know, a multi decadal negotiation about a river basin. So I think that's it. That's one area, where there's a lot that can be done. That's really, really, really exciting. Right? And, I I couldn't thought much about that because, like, we mostly I would say, like, collaborate with the private sector, I would say. It's, like, really, really interesting to think about, like, how large scale applications, you know, we can have about that. So, like, that's for sure. It's another topic for another webinar, and we'd love to hear more about that. Kemal, we'd love to have your your perspective about the about these questions as well. Right? Like, what are the what kind of examples you can go with that you can give specifically for for mitigating water risk in real world projects? Yeah. I I think I have a first disclaimer, Jose, because I think I'm really stunned to see how AI is helping our team at water plants, so I may be a bit biased. So this is a disclaimer. So I would say, like, you know, I I think AI has proven to be a game changer in predicting and mitigating water risks by by integrating different datasets, automating different data collection, and in uncovering insights that traditional approaches have missed or might miss. A high level example is how AI enhances our regulatory and reputational risk assessments by tracking some local regulations and identifying water related concerns that are not being captured in any of the standardized global datasets as of date. For example, I think you already, broke the suspense, Jose, but I think, with the help of our amazing data scientist team, we have developed WaterPlan has developed internally something which we refer to as a local data collection pipeline or score data collection pipeline, which leverages AI to systematically scrape scientific reports, local, even, local insights, local news, providing businesses with real time insights. This enables companies to stay ahead of the regulatory changes, and it also helps them to optimize their water use efficiency and refine water standard ship strategies based on high quality and context specific data and not a global data which may which may not hold any context in their local setting. I know this was, maybe a high level example, but if you look at a very practical case of AI driven risk mitigation, at one of our client site, of course, not disclosing the names, but it that site used to, extract water as of date even from a nearby lake. Historically, there were no indications of disruptions in water supply, water availability. But when AI powered, simulations combined with some local data, it flagged some potential supply disruptions by analyzing climate changes and the water usage patterns in the local local setting. This led to us, creating a simulation scenario, where we we estimated a 510% reduction in lake levels. And then we were able to predict the number of days the water, withdrawals could be halted. And you know what? This actually happened with the site, and they were, like, already prepared. They were mentally prepared. They were physically prepared to handle that that, extreme event. So, therefore, I think by proactively, modeling these scenarios as Ben and we all have been discussing, AI enables better planning for water scarcity risks and help develop strategies to mitigate future supply disruptions. So this is overall, like, all the future scenarios that can be covered for any corporation ranging from public to private sectors. That's, that's a really good point as well. And, I wanted to add it to that, comment, Evan. Like, that has to do with the the question that Renee asked, which is, like, how can we know if AI has been properly trained in the beginning? Right? So okay, Renee. Thank you for that for that question. How are you doing? I can tell you what we are doing in WaterPlan. Right? So specifically for WaterPlan, we are building, like, the first AI water expert. This is the way we think about that, specifically starting with, finding and doing, like, a, like, a accurate scoring with local information for each one of the locations in which companies operate. Right? For doing that, we develop, like, a pipeline of different agents that collect information, cross stitch information, filter information, and do the scoring. Right? And the the way we think about accuracy is, hey. We have to make sure that the AI is, who points to the best expert doing a specific one risk assessment in this specific side. Right? So the way you think about the accuracy is you can think of it in two ways. Right? You have, like, real time accuracy that has to do with, hey. We have an internal validation tool. And every time that we have a new location, we do a validation. Right? And we want we we should make sure that each time that, for example, an expert, like a human expert, do a correction, like, the accuracy goes down. Right? So you you want to achieve to a place where the human expert, they don't have to do, like, any correction. Right? Because, like, the the AI is really accurate, I would say. The other way to do it is, like, with that static accuracy that has to do with, hey. We have a specific data set of amazing actors, the best in the world doing specific one risk assessment in these specific locations. Right? And then we run the AI and we compare to that that static data set. Right? This can be, like, more nuanced. Right? Because, like, you can have, like, overall accuracy and then you can have a recall, and then you can have, like, availability of the information that you have. Right? So that might depends specifically in the use case. Right? But this is the way we're thinking about that. And, yeah, we have, like, different ways in which we can continuously improving the accuracy that has to do with the way we do evaluation, the way we do the prompting, the way we improve the models that we have, the the new models that we try, etcetera. And, I believe that's quite exciting. Right? Because, like, the way the the more the better that the AI becomes, the more accurate, like, all the local data becomes. Right? And, and that is, like, quite quite exciting, I would say, in that use use case. Right? So, I don't know, like, Keimant or or Ben, if you want to talk a little bit about, like, that question as well. If not, we can go to the to the next, slide. I think that was a great answer. So I just wanna emphasize what you said. Like, Renee's question is exactly the critical question to be asking. It's a huge consideration. Totally. Let's go now a little bit to talk about challenges in AI adoption for water risk management specifically. We hope to have your first your input and then Ben and then Raymond. The question is, what are the biggest challenges industries and businesses face when implementing AI for water risk management And how can this be addressed? Ben, let's have your perspective first. Yeah. Thanks. And, again, I think this does, it very much relates to to the question about proper training of the of the AI, which is, you know, having trust and transparency in the use of AI is really, really critical. And that that applies in various levels. Okay. So one thing is just knowing that the algorithm is behaving appropriately, that's well trained, and and understanding how it would be expected to perform as you project out to the future. Right? So there's these are classic data science problems, right, of of overfitting, of extrapolation from parameter space with different algorithms. And so I think that I think I think people have and organizations have a very well founded skepticism about some of these things, especially when they're black boxed and saying, well, I don't necessarily understand the math behind this. And very few of us do, right? So that's okay. But if you don't, if you're not gonna dig into the actual like math of it and this behavior, then say, okay, well, how do I know that what this is really doing and that that's gonna be reliable for my particular application? So that's on the algorithm side. And then there's the broader question, which I think is a big, it's where again, the term AI is I don't wanna say it's unhelpful, but sometimes it gets us into conversations where we're talking past each other. Because there's a general question, I think for society right now, like, do we trust AI? So what does that mean? Right? So there's that that specific question of are the algorithms, reliable? But I think people when they say that are often thinking about a bigger question like, do I want this decision made without my direct, you know, my hand on the scale? And can I explain this? And so I think that, you know, I'll let him speak to private sector because he's got better experience from that. But you know, I work a lot with the public sector, local governments and things like this. And there, it's a huge issue, right? Because I'm like, I mean, if a decision is gonna be made, and then whether it succeeds or fails, then you have to explain that to the voters, the taxpayers, right, to your own citizens, and say, you know, well, here's we did this because the machine told us to. Right? Well, that's not that's not a great answer. And so really, I think a big challenge is in how do you collectively formulate the questions, collectively evaluate various outcomes with the AI kind of in the middle there, but have a comfort level saying, you know what, that we are still in charge of our own destiny. And that these kinds of sensitive resource decisions are gonna be made by communities that by the proper stakeholders and that the AI is empowering those stakeholders rather than in somehow being viewed as this external threat that's gonna tell them they're wrong and change the way that the water is managed. And so that's, I think as we all get more comfortable assuming we do it right. Right? I think as we all get comfortable with the use of AI as, again, as that partner in the process, that will get better. But right now, it's a huge, I think, it's a huge barrier to to adoption that we're seeing. That's super interesting. Right? Because it's a it's a matter of accuracy, obviously. Right? But also it's a matter of, like, even if it's, like, super accuracy, like, getting used to that. Right? And then it's a matter of, like, also, like, I think in Ben what you described, like, user experience. Right? Because, like, maybe you want, like, the suggestion, but then the last click and the decision, you want, like, the the human to to make to make. Right? So that's interesting to think about, like, these three components, how they come together. Right? So really, really, really interesting. Hey, Manuel. Let's have your your perspective as well. I think, Ben really highlighted the the challenges that we face with the public sector because, I had a long history of working with the public sector. But I think in the last four years working in the private sector, I think, Ben, the challenges are not much different. The trust issues are still there. And I think, actually, AI models sometimes how they are structured, they can act as black boxes. And it makes an it makes it hard for businesses to fully trust their outputs. And I think addresses addressing the this requires the use of explainable AI models, like what the world hydrological world especially shifting to, and continuous validation against the real world data. And balance between AI insights and human expertise also needs to be identified. So I think this is one of the challenges and one of the how we can address them. And I think some things that we are we are actually facing a lot with the public sector is their integration with the legacy systems. Many businesses struggle to fit their AI tools into their existing workflows, existing systems, and IT infrastructure. So I think overcoming this challenge, requires them to be early adopters as we do have a lot of early adopters with water plan with whom we are working. And I think we are setting the space for that, for the new adopters. And another thing is to start investing in AI ready platforms and providing training to ensure that teams can effectively use AI driven insights. I think Jose and, Benny both mentioned about the last click. So I think that is where, like, you know, it it needs to be realized that it's an aiding tool. And that is where the last challenge comes is, like, the change in management and adoption is something a big challenge, actually. So even with the right AI tools, successful implementation still requires a shift in mindset. You know? So many, organizations face internal resistance, as you can imagine, due to concerns over job securities, due to, a mindset that overreliance on AI may disrupt the the whole system. And the key is to position, AI as a decision support tool rather than a replacement of the human expertise. So I think this is how, like, you know, this this, this has to go hand in hand, and not something which is replacing the human expertise. So this is how, like, you know, these are the some challenges and I think the way forward to addressing those challenges. Yeah. I I totally agree, Eman. And and my perspective is, like, as AI, like, continues to evolve and gets better, the expertise from the human is more important. Right? And and the thing behind that is because you have to train the, you have to tell them how to do things. You know? And, and and and this is, like, dynamic. Right? Specifically for what are we management, which is, like, pretty interesting. Right? Like, the other day, I was talking with with one team member of OpenAI, and they were saying that one of the main things that OpenAI is thinking right now is how they can start working on dynamic benchmarks. What's the meaning of that? Because, like, static benchmarks such as, for example, a SAP kind of, like, exam or, like, a sign that that is static. AI is a really saturating those. Like, everything really, really good. Right? But when you have something dynamic, right, this continues evolving. Right? So that is a much, much more harder, I would say, like, challenge. And I believe, like, it's really, really interesting, like, to to think about that. Right? So so yeah. Definitely. Yeah. Maybe kinda like we're running out of time, but I have, like, another question for for you, Ben. So first of all, congratulations of being, president-elect from, American Geophysical Unit at AGU. That's amazing. My question for you is, with this new role, let's say, I like AGU. Right? How do you see AI tools evolving in water stewardship space over the next five to ten years? We'd love to have your your perspective there. Yeah. Thanks, Jose. So it is exciting for me to be in this role partly because I get to be I have a seat at the table of some of these conversations and AI is central. I will say that. I think that probably every scientific society and large institution for right now, AI is just, you know, comes up at practically every meeting and we're having a lot of focus sessions on it. In terms of what's most exciting, I mean, we could talk all week, right, about all the different things that are happening. So maybe I'll just say one thing that I'm interested in, which is, you know, we talk about AI's as black boxes, trying to understand the reliability, trying to sometimes understand the interpretability. You know, they're really good at optimizing, for, for for performance, for predictive skill, but sometimes you might not understand why that is. So in my world, this world of physics informed AI is really exciting. And I think that a lot of people in the AGU are excited about this because it it takes what has been attention. Right? You've got your kind of people who wanna do physics based analysis and then they're saying, well, what is this AI black box magic that doesn't actually explain anything? And it turns attention into an opportunity. Right? Saying, okay. Well, the like, the AI are remarkably good at at discovering physical forms, right, at discovering relationships and things and identifying patterns. And if we can then integrate that to a physics informed knowledge framework, then we can have the best of both, right, where we say we're getting great predictions, where it's those predictions are in fact informing our physics understanding to make better hypotheses, to make better broad scale understanding of potential futures, that do require an understanding of the system and not just a statistically performing algorithm. Right? So I think that that's one area that I've really gotten my eye on, and then I think that the AGU, we're gonna see a lot of science coming out of that, institute that that organization, in that area. That's quite exciting. I would love to to see, like, in or especially more of that, but that's a really, really interesting kind of plan, I guess, what you think about that. Right? Like, not just AI, but how you can integrate that with the physical models, I would say, for precision making. So I believe that that that makes a lot of sense. I have a question for you before we we wrap wrap up, which has to do with reporting. Right? So you have a lot of experience working on this. Right? So what's what is the role that do you believe that AI will play in helping companies align the one stewardship efforts with global reporting standards. Right? We have been talking a lot about, like, reviews, webinars, like, CSRD, the CDP, and TFID TCFD. So we'd love to have your your perspective there as well. Well, I think that's a very interesting, question, Jose, because in a way, it is the summary of, like, what we discussed the whole, in the whole webinar for the last fifty minutes. As a matter of fact, I'm working on this project these days. And I think, like, in my opinion, how I'm seeing things evolving is, like, AI can play a transformative role. I'm I I again repeat it's a transformative role because, it can help companies align their water stewardship efforts with global reporting frameworks. Like, some of them are there, but I would say, like, CDP, TCFD, ESRS are a few to to mention. And with I would say, like, with I what what I have been seeing in the last one year is, like, these reporting requirements are becoming stringent with each passing day. And there there is where, AI can simplify the process because collecting this data and, reducing the reliance on manual efforts is something that every company is looking for because these reports require a lot of data gathering, lot of data collection, lot of data analysis, and then getting insights of it. So it can actually, like with the help of AI, what we have been doing is that we are trying to automate the data gathering and process processing of different data points and then enabling which enables us faster and more accurate compliance with the reporting standards. And I think even, moving beyond efficiency, AI sorry. AI adds value by identifying patterns and insights that may not be visible through traditional approaches. It can integrate different datasets, providing the necessary context to produce comprehensive and critical, sorry, credible disclosures. So I would say, like, I I said that, in the, like, previously as well, but I would like to reiterate is that for sustainability managers or anybody working in this domain, this means less time spent on labor intensive tasks and more focus on strategic decision making. By enabling faster, more accurate, and better contextualized reporting, AI helps AI can help businesses meet global standards while showcasing their commitment to water stewardship in a meaningful way where where it actually matters. Totally like, I believe that's spot on and totally, like, aligned with that. Right? So there's a huge opportunity, I would say, like, then how can we get, like, what our experts and, you know, we need people to, spend less time reporting and building the reports and more time, like, taking action and making the right decisions, which I believe is what everyone wants. Right? So there's a huge potential of leveraging AI, not for just for the data direction part, but also for completely streamlining kind of the reporting and make sure we have this as as, you know, we we mentioned before, like, be this, enhancer for making decisions and really really focus on that and really focus on creating different scenarios and make sure that we are doing, like, the best decisions for each one of the the the things that we have to to achieve. Right? So, we are running out of time. It has been a amazing discussion. Right? I I really love this topic, and, I believe, like I I feel, like, super grateful to have the opportunity for all of us to work in, water, in this AI moment of the world. I believe that there's an amazing, amazing opportunity ahead of us. So thank you so so much, Ben and Jiman, for for joining. And, yeah, maybe just before we, wrap up, just wanted to mention a couple of things. Right? We'll share with every one of you the recording of this amazing session. Right? You it's also gonna be present in the Water Plan website. Right? So you can enter the air, you can download it. And there's a couple of new, webinars that are coming up in the next few weeks that I believe would be really interesting. We have one specifically talking about factories in the context of of, overwater stewardship and one specifically talking about water scarcity risk. Right? And, the official you read it, that it it it's gonna be, like, an amazing amazing conversation, as well. Maybe, like, just like the last question that I have just, like, because, like, I'm super curious about that, that Ben and Aman will not have your your your point of view is so we have been thinking a lot about, like, the AI water expert. Right? Specifically for the data collection and and gathering, like, the local information. Right? There's a lot of conversation and and not just in water. Right? But, hoverable, about, like, creating, like, the AI scientist. Right? So not just, like, collecting information, but also be able to achieve one moment in which we have AI that can that can help us create, like, new new new knowledge. You know? So I'm I'm super curious how you think about this. Right? And specifically for water management, how you know, close or far we are from that. Right? And I would love to have your your perspective more than thinking about, like, science and academia specifically for that. Right? So so yeah. I I think that, you know, one of the questions that's most interesting to me right now is philosophically is do we already have it? Or like like when do how do we know? Right? The old, you know, the new versions of the Turing test because, I mean, in some sense, these these approaches, once you're at a place where the analysis suggests the next analysis and then you have an algorithm that can say, okay, well, based on the error characteristics of my results here, I'm gonna do this other thing. You already have this this kind of learning and implicit hypothesis generation going on. So in some ways, it's like how do we package that? Like chat CBT style where it's like, oh, look, now it feels like I'm talking to a scientist. Versus the deeper question you might be asking is, which is like, when does it really change our approaches to knowledge? And that's something I don't have an answer on, but I think it's it's just a fascinating Yeah. Think about Yeah. It's like when when it creates, like, new knowledge. Right? Yeah. And then we but that that is wow. Like, you know, like, when we are achieve that. Right? So, hey, man, we don't have your perspective as well. I think it's it's like, I I fully resonate with Ben, and I think sometimes, even even, hydrologist myself, I feel like, talking to chat GPT, actually, it creates a new knowledge at that moment in time because how you define new knowledge is also something you'd be need to define more rigorously. But sometimes, like, you miss some very basic things and, talking to these AI bots, which you yourself have trained based on different documents, based on different inputs, it gives you some insights which are, like, really an eye opener for you on that moment. So I think, like, Jose, it really, for me, I think I won't say, like, as a new knowledge, but I would say, like, some missing knowledge or some knowledge that we tend to forget or we tend to, like, you know it's a it's a human tendency these days to move towards more complex things and, you know, forget the basic ones. So I think it it's still, like, you know, reminds you of those basic stuff that still exist. And I think those are sometimes, like, really the eye opening moments for me and my team. So I think, like, I feel like that moment is already there, and we are very fortunate to be in that moment, in that right moment, I would say. It's amazing. Thank you so much again for joining. I I feel that we can go be talking for two hours more. So for sure, we're we're gonna have more conversations, talking about this. But thank you so much for joining. It's been a amazing discussion, and thank you for the audience for for joining as well. Looking forward to seeing you all soon. Bye bye. Thank you so much, Jose. Thank you, Ben. Bye bye. Cheers.