According to the Center for Digital Government’s* Digital Counties, Digital States and Digital Cities 2022 surveys, chatbots have secured a foothold in all levels of local and state government as a majority of agencies either currently use them, or plan to in the next 12 to 18 months.
This new frontier can be difficult for government agencies of all sizes to navigate, so Government Technology called on senior IT analyst for Placer County, Calif., and public-sector chatbot expert Benjamin Palacio for insight. He was part of the team that developed Ask Placer, a government chatbot launched in 2019.
NEGLECTING WEBSITE CONTENT
Whether or not a government agency has implemented its own chatbot, information posted on its website is already part of the bot’s evolution. Palacio said that even before an agency considers launching their own bot, they need to be sure they have the systems in place to keep their web content up to date at all times.
“With ChatGPT, there was no announcement that it would be coming, they basically said, ‘We’re going to go start scanning websites and pulling content for our models,’” he said. “I have a fear that’s going to start causing issues, because the information that ends up in [bots] somewhere is not going to be the same information that might be on your website. A lot of times information goes stale on the website.”
FAILURE TO ASK “WHAT'S NEXT?”
One of the first decisions a government agency needs to make when developing a chatbot is whether they’ll create it in-house or purchase an off-the-shelf option.
He cautions agencies to avoid getting themselves in a situation where the chatbot isn’t able to connect with other agencies or be expanded upon. Constituents aren’t usually well-versed on how to complete tasks that involve multiple government offices, so if a user asks a question that applies to a different department and is given limited or incorrect information, the quality of the interaction will suffer, and residents are less likely to use the technology.
Palacio said the ability for a chatbot to consider context and pull from more than one source is essential.
“It gets tricky with some government agencies with different budgets and all these different things going on,” said Palacio. “Trying to design a global encompassing chatbot is difficult because not every department is going to want to fund connecting to their database to get information for something, so then it becomes a prioritization thing.”
The Ask Placer chatbot that Palacio works on for Placer County launched in 2019, and has since expanded to connect additional features and agencies. The county started with a third-party chatbot, and to execute an expansion they implemented another third-party vendor service inside the off-the-shelf chatbot.
“You can do these things in pieces, you can develop a chatbot and start with the content and the context, and then from there, you can start adding to it and see what can happen,” said Palacio. “That’s probably going to be more likely with a customized solution where you can add pieces to it, not to say it can’t happen with something off the shelf, but it’s going to be more limited.”
He added that future scope can often be gauged by the type of agency and how massive the set of information it encompasses is.
“I think with a county entity or federal agency that’s got a bigger set of information, the focus would be on developing APIs or connectors to other systems where they can pull out information,” he said.
The ability for a government chatbot to answer questions and direct users forward will be a key factor to its success. A Gartner survey of 497 B2B and B2C customers from December 2022 through February 2023 found the ability of a chatbot to move the customer’s issue forward was the top driver of adoption.
RELYING ON OR RULING OUT NATURAL LANGUAGE PROCESSING
When Palacio was first working on Ask Placer in 2018, the initial project focused heavily on virtual assistant technology to connect to common technology that was already being used inside homes like Amazon’s Alexa and Google Assistant. The idea was that residents could vocally ask questions and listen to the answers.
In reality, citizen interactions with the goverment agency were more complicated than “What temperature is it outside?” They wanted documents, forms and more specific information.
“I think we lost a little bit of that engagement there where there wasn’t a lot of traffic, and it just wasn’t as easy as ‘I’m on my laptop or my home computer and I can open it up and have it, print it out, or do something else with that documentation,’” said Palacio, adding that the ability for a chatbot to share links and documents and respond to a second, follow-up question quickly became a critical piece of the puzzle.
However, avoiding natural language processing altogether would also be a mistake. Some constituents are not able to type or read written responses, creating accessibility challenges.
“[Accessibility] is something that I think kind of got skipped at the beginning, just because of the coolness or the fanciness of it,” he said, adding that it’s also important for agencies to consider their demographics and the best practices to communicate with constituents which includes incorporating different languages into the system. “The value of having other languages is going to become a factor.”
SKIMPING ON SECURITY OR TRUST
A May 2023 Quinnipiac University Poll revealed that more Americans think AI poses a danger to humanity than those who believe it will benefit society.
Governments have to ensure constituents trust the chatbots they’re deploying to do the right thing. Chatbots that are able to access account and billing information require added cybersecurity features and data privacy practices.
“Analyzing the data before it even gets to the point where it would be accessed by a chatbot is essential,” said Palacio. “I think security would be more of an issue when you start talking about integrations to systems. You don’t want to have a system that’s going to be able to access everybody’s information. So you’re going to have to look at ‘How do I do a multifactor type of process?’ Where we’re not just going to give you the information because you put the account number in, there’s going to have to be some kind of a PIN code or some kind of an authorization process.”
Identifying a chatbot as a bot, rather than a live assistant, also is a critical transparency factor. There’s a wide range of ways government agencies develop their chatbots, some much more humanized than others (you can test your knowledge on some of the most uniquely named chatbots in Government Technology’s chatbot quiz.)
Ultimately, Palacio expressed that while it can be entertaining for a government chatbot to have a strong individual identity, it’s not necessary.
“I think the focus is trying to get it down right, with context and getting people to the right place,” he said. “It’s not ever going to be perfect, especially working in government. It’s difficult because there’s so many questions that overlap, and that’s where it gets really tricky.”
*The Center for Digital Government is part of e.Republic, Government Technology’s parent company.
