Government websites generally aren’t the easiest to navigate, but the City of Amarillo plans to change that with the help of artificial intelligence.
With the help of Dell Technologies, Amarillo officials built a virtual assistant called “Emma,” powered by AI, specifically for the city. Emma is one of the first of its kind in the country, but likely not the last.
Rich Gagnon, assistant manager for the city of Amarillo, spoke to Texas Standard about how Emma was built and its capabilities. Listen to the interview above or read the transcript below.
This transcript has been edited lightly for clarity:
Texas Standard: So what’s so special about this technology that you guys are going to be deploying called “Emma”?
Rich Gagnon: So Emma is technically a digital human embedded in our website. Instead of going to a website and searching for an answer, you can actually just ask a question.
So something simple like, “how do I pay my water bill?” That allows a resident not to have to be technically capable or even speak English. Then Emma will not only answer the question, but navigate them to the page they need and give them the resources or information that they’re looking for in a way that’s easy and, frankly, the way that we as humans prefer to communicate.
You said it doesn’t have to be in English. What languages are supported by this service?
Right now we have the capability of up to 46 different languages – rolling it out in English first, of course, to test the functionality. And then we’ve taken our top six languages and we’ll run through those and test her response in that language by native speakers to make sure that the information translates well, because that’s one of our discoveries. By doing this project, it’s forced us to look at, even on the website, how we communicate with our residents.
So it’s not only changing the way we communicate, but it’s making our website better because we’re having to make things easy to understand in order for that to translate well into multiple languages.
Well, why did the city make this investment? You know, people need to know what their trash day is. Why is this the the best way to get that done, moving forward?
For a couple of reasons. One, we’re a little different maybe than many cities in Texas. We have more refugees per capita than any city in Texas and most of the country. Actually, I have one middle school speaking 62 languages and dialects. So in order to provide that kind of service to our residents, it would cost us a fortune in translation.
And then we have residents that are not really technically capable. So it’s really difficult for them to find things using government websites. And typically how they’re designed, it’s not always easy.
So this makes it very easy to get quick answers without the staffing costs it would take to manage that many incoming calls.
What is the process like of developing Emma? Is that something that you were closely involved in, or is it something that the folks at Dell kind of did independently, and what was it like?
No, we’re doing that internally. So, you know, we’re the first doing this in the local government space. So we didn’t know what we didn’t know. And it’s actually been a fun process to go through.
You know we had to learn things. And our first hurdle was to get over hallucination. You know, these language models are designed to give an answer, period, even if it’s the wrong answer.
So we worked with a partner called Pryon that actually does the knowledge fabric. And developing that capability of creating the city knowledge and being able to say, “well, I don’t have an answer for that; here’s the number to call to get that answer” instead of just creating something that was probably our biggest hurdle, which we’ve gone over.
But I think the most fascinating part for me is it’s really forced us to design the way we communicate with our residents in a very human way. Because if you think about how it’s almost like having a really smart intern with no context, and we’re teaching her about our city, and in doing that, we’ve gone department by department to test her. It’s made us relook at how we communicate.
I’ll give you a simple example: our finances. We took our budget, which right now is a 400-page PDF that we publish every year, and just did that and then just asked simple questions that a resident would want to know. And we couldn’t answer it because it was all in government accounting language. So we had to take a step back. And if Emma can’t answer it, then our residents probably aren’t getting the answers they need.
So do you think that this could have a positive effect on, say, government transparency?
Oh, I think so, for sure. As I said, we’re co-developing this with our residents and it’s actually driving tremendous transparency because the questions they’re asking, we see on a daily basis. Those go to our directors of the department.
And now we’ve set up an SLA so that when she goes live every day, we at leadership, at the council level, at the director level, at the staff level can see, “okay, here’s the questions that are getting asked.” And we need to make sure that we’re closing the gap as we go forward and answering those questions efficiently.
How much better do you think Emma will get over time? This is version 1.0, I guess. Do you see even more ambitious possibilities in the future?
Oh, absolutely. So full transparency: the team that’s developing this, I’m already working on 2.0 and 3.0.
If you think about a conversation, a conversation is two-way, right? Right now, Emma 1.0, a resident asks a question and Emma gives an answer. But we’ve found a way to actually turn that the other direction.
So imagine if you asked Emma to help you book a table at one of our parks. What we’re working on now is the capability of Emma not just taking you to the page and helping you book that table, but then she could ask a simple question like, “hey, did you know that the city has this park program that we’re rolling out? What do you think about that?” And now Emma becomes a two-way conversation where we’re gathering feedback from our public and creating almost a sentiment engine.
