Making AI Work for Government

Artificial intelligence (AI) holds both promise and risk for government users, according to Paul Smith, Co-Founder and Chief Technology Officer of Ad Hoc, a digital services company.

“Government is a special environment when it comes to technology,” Smith said. “Things like security and privacy come to the fore. But there’s an opportunity here for a big transformative moment, to use AI and ML to make government services much, much more accessible to people, and improve the user experience as well.”

You have probably heard by now about large language model (LLM) AI chatbots that scrape the internet for data and often produce untrustworthy results. Smith recommends “extreme caution” when using an AI to act on its own — updating a database or sending an email. “There are too many unknowns and risks,” he said.

But an AI trained exclusively on a specific set of agency data is much more reliable, Smith explained. “Using a system trained on your enterprise data to make predictions about it and then having a decision-maker in the loop to act on those predictions or recommendations is promising for government.”

Translating the Complexity

Ad Hoc’s business concentrates on helping government agencies better serve the public by building web and mobile applications that provide good digital experiences. “We’re looking at AI and ML to improve how people interact with services,” Smith said. “Government programs can be very complex. They can have lots of different rules and processes and procedures, and it can be hard for a member of the public to understand.”

But recent developments in AI have made the technology very good at translating, Smith observed. “An AI trained specifically on the documentation of a service, with all its complex jargon and technical and legal issues, that could deliver answers in plain language would make it much easier for constituents to get what they need.”

Bringing Structure to Unstructured Data

One agency is using AI to turn unstructured data into something more usable for evaluating the progress of its grant program . “Grantees have to update regularly on how they’re doing,” Smith said. “That’s essentially a block of text typed into the database. That field has a lot of important information, but it’s unstructured text. It’s hard to do comparisons to see how they’re doing across all the grantees.”

So they are using a language model AI. “The model is very good at understanding the underlying meaning and the intent of the written word,” Smith said. “They can take two pieces of text that may have been written differently but have similar underlying concepts, and say, ‘these are actually the same thing.’”

That helps the agency make comparisons across those fields. “It would have been prohibitively expensive from a manual labor perspective to go over each of those fields. With our approach, they were able to clean up tens of thousands of records in essentially one day,” Smith said.

This article appears in our new guide, “AI: A Crash Course.” To read more about how AI can (and will) change your work, download it here:

Photo by Tara Winstead at pexels.com