This article was taken from the paper “AI & Water management — What utilities need to know now.” You’ll find the full paper here.

The future of chatbots: Imagining a more intelligent customer service interface for water customers

You can see how ChatGPT, limitations aside, could have helpful applications when it comes to more intelligent customer service chatbots in the water sector. Most customer service chatbots today start with computer generated responses to customer questions. When the computer can’t answer the customer’s question drawing on its available database of preset questions and answers, it will pass the customer onto a human. But these kinds of chatbots often feel very much like you’re talking to a computer, and often aren’t very helpful.

If something like ChatGPT, however, had access to information from a particular organization like a water company, it could produce much more relevant answers to customer questions by drawing on billing data, customer data, and even the water system’s data.

The AI may be able to get much further answering concerns and questions like “Water trickles out of my faucet,” “My water is brown,” or “Why is my bill so high?” before passing the customer on to a human. 

Imagine, for example, a customer complains to an AI-assisted chat with powerful capabilities like ChatGPT that their water is coming slowly out of the faucet (due to low pressure). If it had access to the right data, it could do sentiment analysis to see if there are other reports of problems in the same area. Then it could alert crews and formulate a helpful response, such as, “Yes, we appear to have a problem in that area, and we have a crew heading there. They should be done in two hours and your service will be restored.”

That’s the type of impact an AI like ChatGPT could have on improving customer service chat bots in the water sector, if it had access to the organization’s data and its other limitations, which we listed in the first section of this paper, were resolved.

Example conversation of a utility’s customer with an advanced chatbot

Evolutionary computation for solving complex water distribution system design problems (in half the time)

One promising area of AI in the water sector is evolutionary computation for discovering optimal designs for water distribution systems. This would be especially relevant to planning initiatives in mid-sized water distribution systems, such as cities of 50,000 to 100,000 people. It could be used, for example, for long-term master planning initiatives to optimize future expansion, maintenance, and repair costs.

Evolutionary computation is a subfield of artificial intelligence that solves highly complex problems with too many variables for traditional algorithms. As the name implies, the field is inspired by the fundamentals of natural evolution. Evolutionary computation uses genetic algorithms that draw on principles like inheritance and natural selection — the standard that organisms better suited to their environment will propagate their genetic material and reproduce to become more numerous in their environment.

In the water sector, evolutionary computation solves the problem of testing the almost infinite possibilities there are to achieve an optimal water system design. Even with 100,000 feet of pipes, for example, if you had to change each and every pipe in turn to find the most optimal design in terms of efficiency and cost, there’s no computing time in the world available to do that. It certainly can’t be done by a human. Most engineers will only have time to consider a handful of design solutions for a project.

The idea is that the AI can continuously test small changes across large complex systems, adopting the strong solutions and changing the weak ones, until the system reaches what’s called “a near optimal solution.“

In fact, evolutionary computing has been found to save over half the cost of, and produce better solutions than, manually produced designs. AI based on genetic algorithms can bring us solutions to water distribution system design problems that our sector could not have tackled in the past without AI.

AI for everyday operations: machine learning for fault detection and computer vision

Fault & anomaly detection

One of the most relevant applications of AI in a utility’s daily processes is fault detection. This would be an AI system that uses machine learning to detect faults in a water system before it causes a problem for the customer. The way it works is the AI learns what is normal based on the historical performance of the system. It recognizes normal patterns based on time of day, temperature, season, and other variables, and then it makes a prediction for what will happen next. If the prediction is of equal or similar value to what it would expect, then there’s no problem. But if there is a large discrepancy, then it can alert operators.

For example, imagine a pipe bursts somewhere in your distribution system in the middle of the night, and your AI system detects a problem. It could then alert a crew to go and repair it before it causes problems for the customer. In the morning, the customer turns on the shower without even knowing there was a problem.

Combine this kind of AI-driven analysis and alert system with a generative natural language model like ChatGPT, and utilities with limited resources, especially smaller utilities who may not have the in-house analytics department that a larger utility often has, could see a huge benefit.

We see this kind of machine learning and natural language processing capabilities emerging in the water management industry with tools like Qatium, for example, which has a digital assistant “Q” that provides alerts for anomalies like abnormal pressure zones, tank levels, flows, and other important daily operations.

Computer vision

As discussed earlier, a deep learning AI application we increasingly see in the water sector today is computer vision. Computer vision uses deep learning to analyze video and photographs to detect anomalies, leaks, and faults much faster and with more accuracy than traditionally human-manned analysis.

Deep learning AI is extremely good, and fast, at learning visual characteristics of pipes. It can characterize imperfections in a methodical way that humans simply don’t have the bandwidth to do. Image-based deep learning can draw on its immense dataset to be able to identify whether, for example, the footage is showing an accumulation of debris or whether it’s actually a break or slippage in a pipe connection. Even if humans didn’t make mistakes, after two or three hours of looking at this kind of footage, it loses all meaning. But deep machine learning doesn’t have that problem, making it a very good application for Closed-Circuit Television (CCTV) data. There are even recent applications where image-based deep learning models use CCTV footage to analyze the density of raindrops to recognize the intensity of rainfall, which has important implications for quick flood management responses.

What’s next for AI applications in the water sector?

Robotics have a clear future in the water management industry. Our industry deals with infrastructure that is difficult and sometimes dangerous to monitor, access, inspect, and understand by virtue of it being buried underground. And while robots play a key role in other industries where it’s too difficult, dangerous, or disruptive for humans to travel to collect data, the water sector hasn’t successfully fully leveraged robotics yet.

Still, we’ve seen some movements towards robotics in the water industry. For example, there have been devices developed that people drop into pipes to collect data for leak detection. The devices don’t stay in the pipes; they’re dropped in at one location and then picked up at another location, so the data can be harvested and analyzed.

But in the future, we should see more autonomous robots that stay in a water system indefinitely, crawling through a pipe network and continuously collecting and sending data to be analyzed with AI analytics for information on a network.

Pipebots aim to revolutionize buried pipe infrastructure management with the development of microrobots, designed to work in underground pipe networks.

As mentioned earlier, advancements in water sector AI are constrained because, one, we’re not a high-profit industry and, two, we’re risk averse because our water systems are essential to human health. So even though robotics and AI are a logical future for water management solutions, our progress in that direction is slower than in other industries. Still, we should be optimistic about the exciting potential applications of future AI solutions in our sector.


Qatium is co-created with experts and thought leaders from the water industry. We create content to help utilities of all sizes to face current & future challenges.

Dragan Savic, CEO at KWR Water Research Institute, is a Qatium advisor.

You might also like...