Description
Key Learnings
- Learn about the various levels of AI maturity.
- Apply technology applications to benefits of AI use.
- Discover key points in building the case to apply AI within your organization.
JAMES P. COOPER: I'm here to talk about The Future of Water-- Harnessing the Promise of AI within the water sector. I'll be talking through what is the promise of AI? How can that be used across the water sector, looking at adoption of AI, going through a number of research studies that we've participated in over the past year, what is the current state of adoption. Then dive into some of the more detailed topics like laws and ethics of using AI for something like public health, which is the water sector, and wrap it up with harnessing the promise. What are the actions we can take today and should be taking to better leverage AI?
So I'm going to start with what is the promise of AI, thinking through what are the examples, what are the use cases. There's a lot of hype around the topic. Is it worth it?
Before I start, I think it's really important to at least present one definition of AI. I think if I ask everyone what is your definition of AI, every single person is going to have a different definition. So I just want you to have an idea of what I'm thinking about when I talk about AI through this presentation.
So I define AI as the ability to sense, reason, engage, and learn in a manner that seems intelligent. There's a lot to that. But I think the key thing you want to take away is, to me, you need to have all of these together. It's an and not an or in order to truly have artificial intelligence.
For example, machine learning is fantastic. It's certainly a component of AI, but I would not call machine learning itself AI-- all of that through the context of it needs to be a computer doing something in a manner that humans seem, or feel, or sense is intelligent.
I think of AI today as the internet of the '90s. There were certain people that were using it very well. There was a significant promise around it, but it's not something we were leveraging every day and certainly not something that most water professionals were using in their jobs. Whereas today, if you think about the internet, it's hard to identify a single role across the entire water sector of millions of people, millions of workers that don't use the internet in some way for their job.
Even if you're out digging up a water main or replacing a lead service line, you're logging the work order, you're documenting everything, you're using the internet very significantly throughout your role. And I see that as where we are and where AI will come over the next handful of years. That timeline is dependent on advancement of AI.
Speaking of advancement of AI, there's all kinds of opinions of how quickly AI will advance. In fact, I was first starting to use AI with water utilities back in the mid-2000. I have this slide, and I have the dates on here because I presented this slide-- this exact slide-- in 2015.
At that time and, in my opinion, still today we are focused on artificial narrow intelligence. So what does that mean? That's isolated, project-specific examples. The AI algorithms that we create to, for example, predict what pipe will fail next in a pipe network is not the same AI and the same algorithm you're going to use to optimize chemical dosing at a water treatment facility-- different AI, different applications-- whereas that same human could certainly look at data and predict what pipe would fail and also look at treatment plant data and identify what chemical dosing. That's really more towards general intelligence.
Also, many applications are considered a black box. There's uncertainty around what is actually happening behind the computer screen that makes those decisions.
Now, moving forward, we see an advancement towards artificial general intelligence and then superintelligence. General intelligence is really more learning based on small data instead of massive data sets, so we might think about today, which large language models and other things are trained on.
So there's so many examples across the complete water cycle of where AI can be used. I could probably talk for two or three hours just on applications of AI from source water management, and looking at groundwater and runoff, to customer use trends, and everything in between. I'm going to just briefly highlight a couple specific examples but just know that I see the applications of AI truly is constantly evolving. I think whatever the latest need is in the water industry, whatever the latest trend or regulation is, AI is going to be there to help meet that need.
A couple of quick examples. Two very big trends across the water industry are replacement and identification of lead service lines for customers and PFAS. Both of those have come about over the past few years due to pretty significant regulations. And both of those have seen a pretty significant benefit of using AI, where in the past that tool was simply not available for us to use.
If we look at a water distribution system, for example, there's many applications where AI can be used and can be used effectively to make decisions and support the workforce, which I'll be talking about in a little bit, as well as the customers-- ultimately improving customer service for the public and for everyone that uses water within a city or in a water system.
Maybe if we zoom in very specifically on a pump station, for example-- I think that to truly enable and unlock the value of what AI can do for us within the water industry, we need to think bigger than our traditional thinking. So, for example, if you just look at a simple pump station-- typically, today, when I ask people, well, what data are you using to simulate that pump station or to design that pump station?
It's very technical data. It's hydraulic data such as flow rates and pressures. It's energy data. How much energy is that pump consuming for the amount of water that it's pumping? Maybe there's some status data like is the pump on or off? How many hours in a day did that pump run?
Well, if we start to think about what do operators actually experience when they open the door and step into that pump station, we might be able to truly unlock some of the value that AI can bring. An operator when it walks into the pump station, the first things it's going to notice are, is there an unusual noise or sound? Is the temperature different? Is it much hotter than it usually is for some reason? Is there a smell? Is there something burning like a wear or bearing or something?
All three of those are sensors that we could very efficiently put into that building and then use AI, use machine learning, use computer vision, use some automatic algorithms that will constantly, in the background, detect if there's a change in noise or sound. We'll look at a thermal image camera and detect if there's a change in temperature and where that is, whether it's a pump, or bearing, or certain part. So these are applications-- when we start to think about the data we collect within the water industry, if we expand that a bit, we can probably leverage AI much more than we're even thinking about today.
Switching gears and thinking maybe on the water treatment side, so many applications, as I mentioned, about ways that you can really use AI to make a meaningful difference. And I'll talk about what that means here actually on the next section of my presentation.
But things like process optimization-- optimizing chemical dosing, very simple things like optimizing work orders-- making sure you don't have three trucks show up to the same address to do different things-- all about process optimization across the utility-- as a-- as the water utility but also as a business-- all the way down to operator training, having virtual pilots, having flight simulators for treatment plants and all the different aspects that you can have when you allow AI, and machine learning, and sensing, and reasoning-- remember, going back to that definition-- all to be embedded within a computer and allow people to interact with that. It's certainly an exciting time to think about what we can do in the water sector for AI.
So I'm going to switch gears a little bit. I'm going to talk about where we are today-- What is the current adoption of artificial intelligence within the water industry?-- then continue on to talk some about the laws and regulations. Maybe, to some, those are considered more of the headwinds, perhaps to adoption of AI and then wrap up with some key actions.
So I'm going to highlight briefly four different research studies that have been performed in 2024 that I had the honor to participate in. The first is a research study funded by the Water Research Foundation, and that's called project number 5189, Quantifying the Impact of AI and ML-based Approaches to Utility Performance. I will then be talking a bit on AI for water, evaluating the impact of the water workforce, as well as some additional research that's been both performed by Arcadis as well as led by others, such as through American Water Works Association, AWWA.
So jumping right into the results for the first study that I talked about, looking at what are the approaches to utility performance use today, most applications of AI from the results of this survey were looking at forecasting or predicting. There are some there that are optimizing and some focused on reporting, but for the large majority-- over half-- of the applications evaluated, it was on forecasts and predictions that then can be used in operations.
Looking at the different aspects of how it can be used for the different types of work within the water industry, asset management was currently the largest use case and that's more focused on failure identification, failure prediction, when is the pipe going to break, why is it going to break, and which pipe is going to break. But it's relatively good representation across many different aspects of what a water utility does.
And when I say water utility, I should clarify. I mean, the water industry and the water sector, so not only drinking water, but stormwater, wastewater, drinking water, et cetera.
Rainfall prediction is a great example. Predicting and reporting water quality, looking at treatment plant optimization and reporting, and water demand allocation. Forecasting how much water each individual customer will use are all applications that were identified.
Actually, looking across the water cycle, there were examples of AI pretty strongly used across the complete water cycle. One exception was that not much use-- at least within this survey-- that was identified for water reuse, even though we've certainly seen some applications of water reuse where you're using AI.
Looking at what was the headwinds or what are the potential challenges for the use of AI and it's really coming down to the people aspect. It's-- the lack of time is identified as the greatest reason why AI is not being used today. And that's a pretty remarkable shift. I've been doing this for, oh, geez, at least over a decade-- applying AI to projects-- and early on in that period, most of the challenges or most of the reasons why the water industry would say we don't want to use something like AI, or intelligent water, or digital twins is because of lack of confidence in the technology. Can it actually do it?
And you see here, that's actually the lowest reason why people are identifying that they're not using AI. They're saying low confidence in tech is the lowest reason. So that's telling me the technology is there, the confidence in the technology is there, so we're getting much closer to, I think, an opening of the floodgates, so to speak, of using AI across the water industry.
So diving in a bit to the technical details. When I say AI-- within this research study, we looked at-- and this was looking at over 100 different applications and uses-- we identified what type of algorithms were actually being used.
So on the left side of your screen here, you can see-- actually neural networks was the most prominent use, a regression was another example, decision trees, heuristic, many different applications. I'm actually a fan of Bayesian, so I'm surprised that was not used more, but certainly a number of different applications.
Taking those algorithms, we then really dove into them. So the pie chart you see is all the different types of neural networks, specifically, that we looked at, from backpropagation to multilayer perceptron to fuzzy neural networks.
And then within those, just as an example, we further dove into what are the specifics of those. So in the one breakout I show there, it's actually looking at the fuzzy neural networks. And then looking at what type of fuzzy logic was used. Was it fuzzy logic, or was it adaptive neuro inference, and many different applications. So this is a great research study to see truly, for the practitioners of AI, what type of AI is being used where within the water industry. Similar to neural networks, just looking at the different examples of regression and showing heuristic methods here and which of those are most prominent, which of those are maybe not used much yet today.
So switching over from some of the algorithms and the details of AI to the impact and looking at the AI workforce within water. So this is the project that was completed between Arcadis and Bluefield Research. And this looked at 61 projects of AI. About 60% of those were on the operational side, or opex-focused, and 40% were on the capex side, or capital-focused. And you can see here the breakdown of the different types of projects and different use cases of AI within those 61 projects.
So similar to the previous study, we're seeing a large presence in the use of asset management, network management as well. So I see that more as on the operational side. How are you operating and maintaining that network?
One of the results of this research was that there's actually quite a few proven results of AI in the water industry using these projects. And this is just showing some of the examples, everything from increasing meter revenue capture-- a pretty significant opportunity since meters are technically really the source of income for the water utilities all the way down to energy cost savings, so thinking how you can work more efficiently, and more effectively, and more sustainably overall by using AI.
Going back to the people component, I think it's really important to connect technology with the people that it's impacting. The water industry is very much a people-driven industry. It's all about interacting with the public and it's all about people doing work, doing manual labor, doing administrative work to perform those tasks. As much as I think it would be neat to have a robot go out there and build a new pipeline or repair a main break, in reality, that's very much a person doing that job.
And we see that here, looking at this data set. This is a representation of all the water, wastewater workforce employees within the US, specifically. It's looking at almost 2 million employees and 60% of those are field operators and maintenance staff. Another 20% of those are facility operators and maintenance staff. So we've got the network-- the field is the network-- or facility is more the treatment plant, inside the fence. That represents 80% of the workforce.
So as we think about where is the greatest opportunity to use AI, I think it's important to think through the lens of, well, where are the majority of people within the water industry and what are they doing? Because if we can impact them in a certain way, in a positive way, and improve what they're doing through AI, then there's a great opportunity there.
Again, looking at some of the data and looking at employment trends. So this is looking at percent employment change forecasted between the next, oh, eight to 10 years. And generally speaking, we're seeing-- actually somewhat concerning-- a decrease in water and wastewater treatment operators and supervisors. Meter readers, a pretty significant increase. That's primarily due to the implementation of advanced metering technology, AMI and AMR.
But we're seeing a large increase in management analysts, accountants, budget analysts. So there's certainly some industry-- interesting trends here that we see happening that I see as certainly an opportunity for AI and I'd like to expand on that a little bit here.
This is looking at percent of employment increase, again, forecasting over the next eight to 10 years. And we're seeing data scientists, information security analysts, software developers-- all around technology is the greatest growth opportunities because of this growing skills need, opportunity, application of technology, of AI to positively influence and truly make life easier for that water workforce.
These are some of the skills within those roles that are needed. Again, this research is something that Arcadis and Bloomfield Research have worked on together and we've been putting that-- putting the results out there, publishing those results out there so that the water workforce can use this and start to look for and build in these skills within their job requisitions and within the opportunities.
Now, one question I often get-- well, two questions-- I guess, one, is does AI only apply to the largest, most complex systems out there? And certainly that's not the case, I'll talk about that in a minute. The other question is, well, maybe AI is something that the younger generation will use and the folks that are more senior are looking towards their retirement dates. They might just think, well, we can kick the can down further and the next generation will be the ones that actually use AI. And that's really not true.
In all the data and research we've looked at, there is not a direct correlation between adoption and interest in AI and the age of the water industry. So it's really important to think through what and how can AI be used across the different ages within the workforce.
So this is a graph looking at age ranges and the percent of the US water utility workforce within those age ranges. And the table below it-- we're looking at OK, well in certain age ranges it's really focused on recruitment. In the middle age ranges, we're looking at upskilling, so providing training through programs, facilitating mentoring. You can start to segment this out and identify how AI can best be used to support these different groups of people.
Now, you can also do that looking at specific operator roles across the water industry. So this table is looking at, well, what is the specific operations categories and technologies used, whether it's looking at a water treatment process, or a lab process, or even regulations and administrative duties? You can break down, OK, here's all the different types of tasks that can be done. And then all the way on the right, you see the various applications of AI within those specific tasks. Again, we can talk about the promise of AI all day long. There's so many examples that can be and are being used in the water industry.
If we dive into a specific role, this is a role of an individual that's a meter reader. You can go through and look at, well, what are their specific day-to-day functions? Do a day-in-the-life exercise. Document and identify exactly what they do in a day. And then from there, translate that to, OK, how could AI automate, or improve, or simplify some of those processes? Shortly I'm going to talk about AI potentially replacing people and is that a thing. It's a question I get all the time. So stay tuned for a discussion on that here in a few minutes.
One of the last surveys I want to talk about is this AWWA survey. Now, this was focused on water distribution systems, so keep that in mind. This survey occurred earlier this year and had about 500 responses and was represented by 10 countries. And you can see here a wide range of utility connections and customers, some small, some medium, some large-- also similar on their system demand from very small to some very large.
A couple of interesting takeaways from this survey. I have so many results. I'd love to spend another hour just talking about these results, but I'll focus only on a couple of key points here. 36% do not have any data governance procedures or policies in place today and about 25% have central data storage across all their systems. However, over half of them still maintain that data within each individual system.
I think one of the promises of digital transformation was easy access to data. And I think the reality-- the practicality-- reality is that as we brought on more digital systems, we probably created more digital silos than what we wanted to achieve, which was open access to data. But we're seeing a bit of a shift there.
Thinking about the easy access to data, we asked all the respondents to identify which data is easy to access and which is hard to access. And you see here that a good amount of data, which should be quite important-- such as water quality, and lab data, and customer complaint data-- is on the very extreme hard to access for most individuals. So certainly an opportunity for AI, certainly an opportunity to allow some data processing within utility networks, within the walls of the systems of the utility to be able to scrub and get access to that data.
Probably one of the most interesting survey responses that I saw was asking the survey respondents what related information systems or technology do you wish you had but is inaccessible to you? And almost half of the respondents-- actually it was the top response-- said artificial intelligence. So this is quite remarkable to me. A couple of years ago, it would have been optimization. It would have been cost-focused, how do we improve things, how do we have a better GIS? And here the number one response is we want AI and it's not accessible to us yet.
So it makes me wonder if there's this driver, if there's this interest in it, if there's all these opportunities and promise, why are we not using AI more at water and wastewater utilities? And there's some research for that. And this is the last slide I have on research, I promise, before we talk about some other topics. But this research study looked at what are the barriers to using AI specifically within the water industry. And the number one answer was personnel or skills, not having staff or consultants who know how to use or manage AI.
So we have an education component that is the number one, according to the survey respondents, reason that we're not leveraging AI and using it. Again, an example here, this is not talking about lack of technology or lack of confidence. This is looking at we don't really know what we don't, essentially, and how we can use AI? Wow. Very interesting topics. Very interesting things. I'm going to switch gears a little bit now and talk about laws and ethics.
I will start by saying, I am an engineer and operator by background. I am not a lawyer, but think these are really important things for everyone to understand as we think about AI and some really important regulations that we should be aware of that are happening out there. So I'm going to start with some of the ethical dilemmas actually around AI, and talk about a few regulations, and then we'll wrap up the presentation.
So you've probably heard about ethics and AI. It's in the news pretty frequently. It's a rather polarizing topic, if I'm honest, because there is so much unknowns around AI. And, for example, this is just a short list, but bringing in bias and the lack of transparency. And will it replace jobs? And is it safe and fair? So I think about that and think, well, what is the role of human judgment in the water industry. Let's think about that.
I have an example to show. It's an example application where AI can be used and is used very frequently today and a bit of a caution on using it.
So this is just an example. This is looking at Seattle Public Utilities and the graph on the left is looking at concentration or quantity of claims within different geographies. The overall map that you see is the service area for Seattle Public Utilities. The graph in the middle with the green dots is looking at customer complaint data or customer feedback data on water quality. And you might be able to see the list there. It talks about brown water, color, taste and odor, et cetera.
So based on what you're seeing here, if we were to identify where we may need to make some capital improvements in our water distribution or our network, we'd look at the concentration of where those are and say, you know what, we think this is where capital improvements are probably needed. This is the greatest concentration.
Now, the folks at Seattle Public Utilities are very smart and looked at a bit more data before they made capital investment decisions. And I'll explain what I mean by that.
So the next two graphs-- the right two graphs on your screen are looking at two different things. First off, it's looking at number of accounts that are delinquent in paying their utility bills. So the red area that you see is the higher concentration and the yellow is essentially a no or very low with blue concentration.
The graph on the far right is looking at people of color within their service area. And what they found is that most of the complaints is not coming from the areas where they are delinquent on payments or where people of color live and reside. And what they found was that if someone is having financial hardship and is late to pay their water utility bill, the last thing they're going to want to do is call up a utility and complain about water quality. They don't want to talk or interact with that utility at all. However, they may have some of the worst challenges from an infrastructure perspective in that area. So it's really important and extremely important to understand your training data and make sure you use the right training data within your AI to make the right decisions.
As I mentioned, the folks at Seattle were wise enough to recognize this. And when they actually have capital improvements in their plan, you see it as spread across the whole geography. And there is a decent amount of capital improvements in the areas where there was not complaints because they identified those were needs.
Back to ethical considerations and the use of AI for water. So I asked what is the role of human judgment in the water industry? For those that are in the water industry, you know that you have to be a certified operator to-- and there has to be an operator responsible for operating and-- responsible for the water quality, whether it's going into the environment or going out into the public in the pipe network. At the end of the day, when there's issues with that with that facility, it's the responsibility of that person, that human, that operator in those. So think about these questions is an AI-based decision defensible? Who is at fault if an accident occurs or a permit violation occurs through the use of AI?
And I also like to think about are all decisions made by AI of equal risk or ethical considerations? Certainly, in my opinion, it's a different risk to identify what pipe is going to fail next and maybe that's not the correct decision than it is to have AI automate a chemical dosing and it put the wrong chemical dosing in water treatment.
So thinking about that, I think there's many different tiers or levels of AI that can be used and is beginning to be used across the water industry. Today, I would say we're almost primarily focused on cost-focused AI and a little bit on performance-driven AI. I think we can move forward. And as we advance in our maturity of using AI, we can do things more that are more human-centered and more socially responsible and, I think, more ethical in the ways that we can use AI within the water industry.
I wanted to get back to a question that I get asked all the time. Is it ethical for AI to replace people in the water industry? A couple of things to chat about here. I think with 100% certainty AI will replace job functions within every organization in the water sector. Now, it's up to the organization leaders to determine if the people in those job functions are actually replaced by AI.
And I have a picture here, great example. It was not that long ago when the primary amount of time that engineers spent in a day was taking designs, and calculations, and thoughts, and getting them onto paper, onto drafting tables that you see here in this picture. Today, that does not happen. We have AutoCAD that came along. Now we have folks like Transcend that do automated design. These things that used to take months to do now happens in days or minutes.
Now, engineers are still here. Engineers didn't go away. Engineers were not replaced by technology. But what they do with their time is much different today than it was in that time. And I think that's going to be true across all of the potential roles within the water industry that are impacted by AI.
Switch gears to some thoughts on laws and regulations. By far, the most well-publicized and impactful regulation is the European Union's AI Act, Artificial Intelligence Act, that went into full force beginning August 1st of 2024. And, most importantly, it applies a risk-based approach to the use of AI. Now, this covers any industry, any sector, but they do identify different sectors based on risk. And that's why I'm mentioning this here in the future of water presentation, because this is very important.
This identifies four levels of risk. The top level is essentially prohibited. You cannot use AI. You cannot use AI for biometric identification of people, for social scoring, for cognitive manipulation. Right under that is high risk, which is regulated use of AI. And very specifically within that, it includes the management and operation of critical infrastructure, which, of course, is water infrastructure and everything that I'm talking about here.
Well, what does that mean? That means if you are using AI within the European Union, you have these requirements to effectively use AI in a high risk system. And there's a long list of requirements here. First, you need to register with the EU's brand new AI office that they've just established and have been working to staff, all the way down to having a risk management system and quality management system for the use of AI, having appropriate cybersecurity, enabling human oversight. I refer to that as having a human in the loop. Now, there are some exceptions, but, be aware, there's a lot there around this European Union AI Act that I'll go back to here in a minute.
In the US, there is not a federal regulation on the use of AI. Today, it's been primarily delegated to the 50 states. However, there are some frameworks and some playbooks through NIST that are-- across the US-- that are great resources, great tools if you want to look at some guidelines around use of AI.
So back up to 2018 and there was one state with proposed legislation on the use of AI. And I'm just going to show you-- fast-forward two years, that legislation was enacted, few more states added. 2022, which, boy, that doesn't seem like that long ago, yesterday-- there's a few more states. And as of about June of this year, wow, all of a sudden, the map lit up. So right now there's over 30 states across the US with proposed or enacted legislation on the use of AI.
Now, I wonder how many water utility professionals today are using AI and are aware of the laws and regulations in their state? Something to certainly look into, and be aware of, and coordinate with the experts so that you're aware.
Going beyond legislation to thinking specific about the water industry-- Arcadis performed some research across the regulatory agencies for water utilities, EPA, the state regulators, the DEPs, et cetera to understand what was their pulse on the use of AI. And certainly varying perspectives from everything to, hey, we really don't address the use of AI to if it's going to have any potential to impact the water quality, like going through a treatment process, then perhaps that requires engineering plan review and a submittal to the agency to determine if you should use it or not. So be aware of that. Be cautious of that. However, important to understand the promise of AI and what's out there.
So as I begin to wrap up the presentation, let's think about what are those actions we should be taking today, knowing what's here, and what's to come, and that we're only just beginning with the use of AI. First off, I think it's most important to know I'm not suggesting I is a red light by any means. I'm a strong proponent of AI. But I also think it's, at this point, probably not possible to completely avoid it if you wanted to.
However, uncontrolled use of AI is potentially dangerous, could be unethical, and is possibly illegal, depending on the regulations in your state. So it's not a solid green light, go full speed ahead on the use of AI. It is very much, in my opinion, a yellow caution light. Have a plan, understand the risks, develop guide rails. What are those considerations that you should have within your organizations to be-- to effectively leverage and use AI.
As the regulations and the ethical considerations evolve, it's almost like a trend line of, OK, there's going to be more and more headwinds and more and more guide rails over time. But, at the same time, the maturity of AI, and the capabilities of what's possible, and the benefits we can get are also growing and outpacing the headwinds. So you need to be aware of both and bring that into your consideration.
Things like ethics, transparency, justice, and fairness, privacy-- all important in the use of AI. So all of this-- is it a cause for a pause? Should we not be using AI? Absolutely not. I think it's fantastic. There's so many opportunities and use cases that I started the presentation with. We just need to be cautious. And I'll end with this quote because I love it. "The economic anxiety over AI and automation is real and shouldn't be dismissed. But there's no reversing technological progress. The key is to implement measures that enable everybody to benefit from those transformative technologies and turn AI and automation into forces for shared prosperity."
At the end of the day, what this is saying is you when people have concerns and cautions, they pause about the use of AI. That's really important. We should not be dismissing those. We need to evaluate those, and consider those, and have those voices heard, and incorporate their needs and opinions into what we're thinking. But, at the end of the day, those voices, those cautions and concerns do not, in any way, outweigh the benefits that are coming and could be used from AI.
With that, I thank you for the opportunity and appreciate everyone's time and attention. There is plenty of research out there, as I mentioned. You can do some Google searches and find out more and the latest information on the use of AI. Thank you very much.