AI pilots ‘deceptively easy’ to start, but ‘fiendishly hard’ to scale up

Federal Chief Information Officer said the next generation of the federal workforce should not only have the capability to work with emerging tech tools, but al...

Best listening experience is on Chrome, Firefox or Safari. Subscribe to Federal Drive’s daily audio interviews on Apple Podcasts or PodcastOne.

The Office of Management and Budget, as well as current and former Defense Department leadership, have acknowledged that improving data management, enhancing agency IT architecture and preparing the workforce all serve as the key conditions to ensuring the success of artificial intelligence in government.

Federal Chief Information Officer Suzette Kent said OMB will release a draft of the administration’s federal data strategy “very soon,” and highlighted some of the upcoming strategy’s top goals for the rest of this year.

Those priorities include safeguarding privacy and building public trust in the government’s use of data, as well as “prioritizing certain data sets” in government that can stimulate economic growth and improve on existing research and development projects.

Drilling down within those goals, Kent said the administration plans to expand its focus on geospatial data, calling it “one of the most successful areas of open data,” as well as its focus on financial transparency.

“Quite frankly, some of the investments that we have to make in data are some of the toughest work that we’re going to have to do in the next few years,” Kent said Wednesday at an AFCEA DC summit on AI.

Under the Foundations for Evidence-Based Policymaking Act, all 24 agencies subject to the Chief Financial Officers Act must appoint chief data and evaluation officers by July.

In addition, Kent said the administration is working on building “reusable” data tools for the public.

“We’ve made a big push to make data available publicly. When you go to, there’s a whole lot of data sets, but it may be very difficult to use them,” Kent said. “We have to make those more usable, and leverage modern technology.”

Actions from a handful of agencies have already helped propel the administration toward meeting those goals.

In the lead up to the 2020 decennial count, the Census Bureau has leveraged satellite imagery data to dramatically reduce the amount of manpower needed on its address canvassing efforts.

Meanwhile, the Treasury Department’s Bureau of the Fiscal Service last month released its annual progress statement, which serves as its roadmap for the bureau to achieve the cross-agency priority goals in the President’s Management Agenda focused on getting payments right.

How AI ‘fundamentally changes the nature of work’

The American AI Initiative launched by President Donald Trump’s executive order last month aims to train “the next generation” of AI researchers.

Kent said preparing for the future of AI, both in and out of government, requires building a pipeline for future talent. The next generation of the workforce, she added, should not only have the capability to work with emerging tech tools, but also be prepared for how AI “fundamentally changes the nature of work.”

“We have to make the investments in our educational system that are going to support the types of individuals and capabilities that we need to have in a long-term marathon,” Kent said. “This is not just something we’re going to be talking about today, tomorrow, a fad for a couple of years. This the direction that we’re going in long-term.”

Next month, 25 federal employees across the government will begin coursework at the Federal Cyber Reskilling Academy, which will retrain non-IT workers into cybersecurity-focused roles.

“We’re reducing manual work, but we’re creating new opportunities in roles that didn’t exist before in the federal government,” Kent said.

To that point, Kent said her office is working “very closely” with the Office of Personnel Management to ensure the government gets the right talent it needs to oversee emerging technology.

Kent said she would meet with the governmentwide CIO Council later Wednesday to discuss how agencies can build a “foundation” for AI in government.

The Centers for Medicare & Medicaid Services on Wednesday launched an AI competition where participants will develop algorithms that “predict health outcomes” based on Medicare fee-for-service data that will be used for front-line clinicians and physicians to provide better data-driven care.

AI pilots ‘deceptively easy’ to start, but ‘fiendishly hard’ to scale up

Lt. Gen. Jack Shanahan, the director of DoD’s Joint Artificial Intelligence Center (JAIC), said one of the major stumbling blocks around AI is this paradox: That it is “deceptively easy” to launch AI pilots with initially powerful results, but at the same time, it is “fiendishly hard” to scale up AI to enterprise-wide applications.

While DoD has launched several AI-focused pilots, Shanahan said the JAIC serves to coordinate AI projects across the entire military enterprise. Prior to the JAIC, Shanahan said there were too many “bespoke efforts” within DoD.

“You weren’t cross-pollinating from program to program,” he said. “It was staying in research and development, and maybe getting out as a pilot or a prototype. But we didn’t get it across to operational capability.”

After overcoming challenges over data management, Shanahan said the team behind DoD’s Project Maven AI program spent most of its time bolting cutting-edge AI tools onto legacy IT systems, but with limited success.

“You cannot get to a future of success in the Department of Defense in AI-enabled capabilities without enterprise cloud,” he said. “Enterprise cloud is part of this solution.”

While the Defense Advanced Research Projects Agency continues its work on rolling out the “third wave” of AI, former deputy defense secretary Bob Work, now a senior fellow at the Center for a New American Security, said the current generation of technology has “democratized AI” by reducing the level of formal education needed to work on the subject.

“If you were taught in the fundamentals of computational thinking in the eighth grade, in the 12th grade, you will be as good at developing [machine learning] programs as a Ph.D.,” Work said.

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

Related Stories