This interview is part of a series, Artificial to Advantage: Using AI to Advance Government Missions.

Back in September, there were more than 700 uses cases in the federal artificial intelligence inventory. By now, there are probably hundreds more. In the last eight-plus months, agencies have launched AI sandboxes and other test beds, have mostly relaxed policies in how they try out certain types of AI tools and overall have grown more comfortable with the capabilities.

Some agencies like the departments of Energy and Homeland Security are more forward leaning, while others are taking a more measured and cautious approach.

But before these agencies can jump into the deep end of the AI pool, agencies need to take some specific steps to continue to prepare for the rise of AI.

Ed Van Buren, the strategic growth leader for AI for Deloitte Consulting, said while there is a lot of enthusiasm for applying these capabilities to mission areas, agencies need to feel more comfortable with what the tools can do.

“We’re seeing a lot of prototypes getting developed; some of them are getting more matured and are getting put into production,” Van Buren said on the discussion Scaling AI in Government. “In the last three years, we’ve seen a shift of the understanding of AI. Three years ago, the typical AI conversation was hard for folks to wrap their head around and hard to understand exactly how to apply it to various use cases. Once they did, it was a very complicated activity to select the algorithms and get your data cleansed, and then train an algorithm and try and apply it to the use case. Once generative AI came out and received such fanfare, I think that government leaders have been able to automatically understand what the power of this is.”

Focus on AI use cases

Now many agency leaders are asking questions around how they use the AI tools in the boundaries of their mission and how they upskill their workforce to use the power of AI.

Van Buren said this is especially true around generative AI. Users are quickly understanding that GenAI could give them polished and well thought-out answers, complete hallucinations and everything in between.

“Organizations are having to look at their use cases and go on a journey, starting with getting some basic insight into the topic, and then using it to get more finished products,” he said. “If you view these tools as giving you a good draft, it’s easy to understand. But if your standard is ‘I want a finished product that’s ready to ship straight out of the tool,’ that’s a much higher bar to clear.”

This is part of the reason agencies are creating AI sandboxes and other areas for employees to safely test out tools in a low risk environment.

Van Buren said a lot of public and private sector organizations are setting up guardrails for deployment, timelines and approaches that limit the tool’s functionality, including taking advantage of secure cloud environments, to prevent potential data leaks or other problems.

“There’s definitely an interest in making sure that these tools are used in an ethical, appropriate and responsible way, which is a great thing for society,” he said. “We’re also seeing inside organizations, the Office of General Counsel, the chief data officers and the program leaders all working together to align their policies and draft policies that help to make sure the organization’s using these tools in a safe way.”

Getting comfortable with GenAI

For those agencies who are farther along in their AI journeys, Van Buren said there are some common use cases that are emerging.

The first is around knowledge management using GenAI.

“Generative AI specifically allows organizations to quickly access data from previous documents, from policy manuals, from procedures, and bring that information much more rapidly to the worker, or to the external stakeholder in a much easier to consume fashion than the hundreds of pages that you might get out of a typical search,” he said. “We’ve been helping our government clients deploy policy engines and things of that nature because it’s such a powerful use case.”

A second common use case is around case management. Van Buren said everything from grants management to healthcare to many others have caseworkers who interact with information all the time and the AI tools can help them make decisions faster.

“The generative AI tools allow that process to be more efficient,” he said. “Another big one we’re seeing is all of the interactions with citizens and external stakeholders. So that could be through a web portal, it could be through a call center, it could be some sort of an intake scenario, and in all of those places people are trying to get information. They’re trying to start some sort of government transaction where they’re trying to share their information. And these generative AI tools really make that process smoother, and do it in a way that’s both more efficient, but easier and more natural for the citizen to engage with.”

At the same time, Van Buren said agencies need to understand the risks of GenAI. He said agencies should apply these new capabilities to areas they already are trying to improve.

“It’s a journey from where you are today to that future end state. And on that journey, the organizations are going to get more comfortable using these tools, data quality is going to get better and better and better. The tools are going to get better and better,” he said. “We’re going to start with more measured use cases, things where we can control the risk and we can get comfortable. We can get our workforce brought along to understand how to use this safely. We can let society start to grapple with the power of these tools, and their expectations are going to change as well. Then, we can incrementally advance the use cases toward those ultimate use cases that I think many leaders are imagining. We just need to make sure that we get there safely and responsibly.”

For more in the series, Artificial to Advantage: Using AI to Advance Government Missions, click here.

Copyright © 2024 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.