What Would Jerry Do? Bringing Generative AI to Water Utilities

“Generative AI, because it uses language that humans can understand, is very accessible and very affordable.”

Gigi Karmous-Edwards is a water technology consultant and advisor who serves on the boards of several technology organizations. Ahead of The Water Council’s ReFRESH Summit, she spoke with waterloop founder Travis Loop about what distinguishes generative AI from traditional AI, how utilities are already testing it, and why data governance, security and workforce training must be part of its adoption.

This conversation has been edited for length and clarity.

What is generative AI, and how is it different from traditional AI?

AI is an area of science that focuses on machines behaving like humans and doing things that humans usually do. It has been around since the late 1940s. However, what we had all been talking about and thinking about prior to the end of 2022, before ChatGPT came out, was machine learning and deep learning.

Both of those use algorithms that look at large data sets and detect patterns. Once they detect those patterns, they can use software to help predict what comes next with another set of data. Deep learning is a more complex version of that using neural networks and layers of complex data sets.

Now enter natural language processing, which has been worked on for the last couple of decades, and you start to understand what generative AI is. It uses natural language processing, so it understands human language. It has also been trained on the internet, textbooks, law books and medical books. Even though it is a probabilistic type of machine, it can help predict things very well—even though it is predicting the next word or the next pixel to create new content, which is text, video or images.

All of this is very probabilistic, so we cannot rely on it the way we rely on machine learning or deep learning. Machine learning and deep learning usually require software developers or data scientists to really get benefits from them. When a utility uses them, it usually works with an engineering firm or has in-house software developers and data scientists.

Generative AI uses language that humans can understand, so it is very accessible and very affordable. That is one way I try to differentiate between generative AI and traditional AI.

What did the new research examine, and how are water utilities using generative AI?

This was a one-year research project titled The Role of Generative AI for the Global Water Sector. We had about 21 water utilities and conducted about 22 pilots, or experiments, across those different utilities. These were utility-led experiments: How do we use generative AI to help solve a problem?

One that is very famous now is “What Would Jerry Do?” We had an 80-year-old gentleman who worked in the City of Carrollton in Georgia. He has been around water utilities and led water utilities for over 40 years. He is the go-to guy. He is the one everyone calls—not only in that water utility, but at surrounding rural and small water utilities. He is very generous with his time. He always takes a phone call and helps his fellow mates out.

We said, he is so busy helping other people, and yet he is still running the whole City of Carrollton. What if we create a chatbot called “What Would Jerry Do?”

Since his knowledge expands across the whole water utility, we decided to focus on chlorine. We came up with 25 of the questions Jerry is asked most often about chlorine and put them on a sheet of paper. I flew down there and interviewed him just like you are interviewing me. We captured that transcript and made it part of a chatbot that answers all kinds of questions about chlorine. It also includes EPA documents, SOPs and local water-quality information regarding chlorine.

Now users in that area can ask the chatbot and get an answer very similar to what Jerry would have provided. There is probably a Jerry in almost every utility across America.

What lessons from the research will you discuss at the ReFRESH Summit?

I am really excited about the ReFRESH aspect. Karen Frost is looking at everything and wants to make a strong connection for the water sector, the vendors and the utilities: How do we all connect?

One of the things I am excited about is sharing some of the learnings from this report. With this new era of AI, or generative AI, utilities now have to rethink their data governance. I am going to talk about how important and critical data governance is.

The second most important thing is security. We want to make sure that if utilities are using large language models such as Gemini, ChatGPT or Claude, they get an enterprise version for security compliance.

The other big area we are focusing on is workforce training. We had numerous calls across this one year of research in which utilities said that different people were talking about AI completely differently. Some people say, “I am not ready for AI because I do not have a computer science degree. I do not have a data science degree, so I cannot really engage with AI.” But generative AI makes it easy for our grandparents to start engaging with AI. We have to rethink the way we talk about it.

We also realized that we needed some type of training for the majority of employees across water utilities. I worked with The Water Tower and Lisa Haney to put together a five-hour, self-paced course for utilities to learn the foundation. By the end, you will know how to write a good prompt, build a chatbot, build an agent and understand what agentic AI is all about.