Artificial intelligence (AI) has come into its own for federal agencies. Federal IT professionals have acquired practical understanding of AI technology, and now they can concentrate on identifying use cases and employing  the appropriate AI technology for the job.

Those are among the observations of two of IBM’s top U.S. federal market executives.

“We’ve gotten past the ‘help us understand the technology’ to agencies really beginning to get hands-on with the technology to understand and imagine what the future looks like,” said Susan Wedge, managing partner for the U.S. public and federal market at IBM Consulting. Now, she said, agencies are thinking about “how can they reimagine delivery of their mission, the outcomes that they can achieve.”

“The AI executive order certainly put into focus how agencies need to be thinking about AI and the adoption of AI,” Wedge said. “And we’re really seeing agencies make a shift.”

Generative AI operates differently than what you might call traditional AI. Therefore, said Mark Johnson, vice president of technology for the U.S. federal market at IBM, agencies should take a step-by-step approach to generative AI. The process involves “finding those use cases, applying generative AI [or other AI technologies] and seeing what comes out,” Johnson said. “Then iterating back again, as we discover some interesting things, and we realize we want to know more [about new] questions.”

For example, Johnson cited human resources and its sometimes-convoluted processes. Generative AI, he said, can reveal ways to simplify or re-engineer HR processes and make operations more efficient for HR practitioners. IBM has had success with AI in its own HR function, to the point that 94% of employee questions are successfully answered by the technology.

“That doesn’t mean there’s not a human in the loop,” Wedge said. “It means that a human is there to handle the more complex, more strategic issues.”

In all use cases, success in AI requires careful curation and handling of training data. Moreover, Johnson said, the algorithm or large language model you train must itself have guard rails to protect data.

“You don’t want to go just throwing [your data] out there onto the Internet, into some large language model that you don’t know the provenance of,” Johnson said.

More than software development

AI projects have some characteristics in common with software development, Wedge suggested. As with software development, it’s “important to curate the stakeholders that participate within those pilots or proofs of technology.” More than simply a technology and data exercise, AI projects must pull in a cross section of program managers and anyone else with an interest in performance, safety and efficiency of mission delivery, Wedge said.

Johnson said that, to a greater extent than in pure coding, you must involve users throughout the process. AI touches “the mission of the agency,” he said. “And that’s where you must get it in the hands of those folks who know what they want the outcome to be. And then let them play with it.”

A crucial best practice, Johnson said, establishes oversight of the ethics and fairness of AI as deployed. He noted that IBM has its own internal AI ethics board.

Equally important: a governance setup to ensure AI outcomes stay within acceptable ranges, and avoiding the kind of drift that can affect generative AI such that at some point, one plus one fails to equal two, Wedge and Johnson said.

The most promising use cases “are not about the technology doing the work of a human, but about making the human more productive,” Wedge said. Case management provides another rich possibility, aside from HR.

“Multiple federal agencies are responsible for managing, responding to, engaging on various cases,” Wedge said. “Imagine if you could use generative AI to generate a summary of the case, and then enable that caseworker to drill down in specific areas.”

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