Insight by IBM

Yes, you can deploy AI responsibly and effectively

If the first round of federal IT modernization brought cloud computing and cybersecurity, the next phase brings artificial intelligence, automation and open source.

We recently sat down with Susan Wedge, managing partner for the U.S. public and federal market at IBM Consulting, and Mehul Sanghani, CEO of Octo (an IBM company), to discuss this next phase of the federal government’s IT modernization journey, the growing importance of AI as part of that journey and the work that IBM is doing to support federal agency’s efforts to develop responsible and trustworthy AI today and tomorrow.

For context: At the end of 2022, IBM acquired Octo, an IT modernization and digital transformation services provider exclusively serving the U.S. federal government. Since then, Octo’s 1,500-plus employees have joined IBM Consulting’s U.S. public and federal market organization, led by Wedge.

Today, federal agencies are pondering how to apply AI and expand on their initial efforts using the technology. For example, should it be used to improve citizen services under the customer experience movement or for making internal agency operations more efficient? Why can’t it be both? After all, the two are connected, Wedge pointed out.

“The most effective application of AI is that integrated workflow that goes from the end consumer of those missions or services all the way through, enabling the operations to be more effective and efficient,” she said.

Wedge believes that AI can make federal operations better, faster, smarter and cheaper. “When I think about the concerns with AI, it really comes down to the ethical use of AI and having transparency around AI,” she said. The big challenge for agencies is ensuring trustworthiness of the AI technology itself, the data, and also how the two are applied to decision-making and programs, Wedge said.

She noted that IBM practices what it preaches, having been among the first to establish an AI ethics board “to make sure that the application of the technology is not just going to drive impacts but drive them in the right way.”

What’s most critical when implementing AI? Getting it right

Elements of responsible development and deployment of AI include understanding how and with what data an algorithm or model has been trained, Wedge said. It’s also important to bring to every deployment the understanding “that AI is really to augment, not replace, human expertise and judgment — making sure that there’s a human in the loop where appropriate.”

Equally important, Sanghani added, is understanding not only how AI and automation applications perform at the outset of a deployment but also how they’ll perform over time. He pointed out that some recent generative AI applications have, in just a few months, already degraded.

For example, one platform has gone from 100% accuracy in solving math problems to just 4%. Sanghani highlighted a potential solution to this challenge: implementing technology that automatically retrains AI models, keeping them accurate through the continuous addition of new data.

He said the Department of Defense uses the technology to maintain accuracy of information sent to troops. It’s a key tactic, Sanghani said, because “almost all machine learning models have drift.”

Another example offered by Wedge and Sanghani is the need for AI to operate in edge environments, such as aboard naval ships where cloud connectivity may not be available.

To help with such challenges, Octo’s 14,000-square-foot innovation center, oLabs, has in-house computing capability that expands IBM’s ability to work collaboratively with agencies to develop, test and rapidly prototype emerging technologies, like those that may be used at the edge. It’s a secure environment that ensures data privacy, Sanghani said.

Whether at the military tactical edge, in health care or myriad other government environments, “it’s important to make sure that work can get done in the optimal location,” Wedge said. “Being able to get the capability to the edge has been a big focus for us.”

AI in government? It’s happening now

Agencies are already tapping into the power of AI to improve operations, processes and procedures; reduce costs; increase efficiency and mission effectiveness; and support decision-making — among other benefits.

Wedge cited AI-driven work that IBM is doing with the Veterans Benefits Administration to help speed up the agency’s claims processing. She said the Department of Veterans Affairs’ agency receives many benefits claims through the mail as letters and forms that are not natively machine-readable.

“We’ve implemented a platform with AI and automation to essentially ‘read’ their mail” after it’s been digitized, Wedge explained. “We read the mail, make sense of the mail and take action on the mail.”

The actions can range from asking for more information to scheduling a medical appointment. The AI-powered solution has reduced a three- to four-month processing time down to days or hours, Wedge said.

“At the same time, VA is seeing productivity benefits where they’ve been able to redeploy north of 500 claims specialists to higher-value work,” she said. “They’re spending their time on the most impactful tasks as it relates to veterans’ benefits.”

IBM has a demonstrated commitment to supporting the federal government’s digital transformation, both Wedge and Sanghai said. The company has strived for more than a century to bring innovative technologies like AI into the world responsibly to augment, not replace, human expertise and judgment, they said.

Listen to the full show:

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

Featured speakers

  • Susan Wedge

    Managing Partner, U.S Public and Federal Market, IBM Consulting

  • Mehul Sanghani

    CEO and Founder, Octo, an IBM Company

  • Tom Temin

    Host, Federal Drive, Federal News Network