Keeping a human in the AI loop is ‘untenable’ due to cuts, USDA IT official says

An Agriculture Department IT official is pushing back on the popular adage of always keeping a human in the loop for artificial intelligence tasks due to government staffing and resource cuts. 

Rudolf Rojas, USDA’s IT program manager and security lead, said at FedInsider’s Innovation Summit in Washington, D.C., on Wednesday that due to cutbacks, the agency lacks resources and contract support for a human in the loop.

“I think that’s untenable in our industry or in government — we just don’t have the resources anymore,” he said. “So we’re going to have to get smarter and utilize these tools so we can have them do the work, and then report back to us.”

That doesn’t mean agents get to run wild; Rojas said he wants agents to have “minimal privileges, minimal role” and would only let them run without a person if the agency has “a  really good risk management framework, you’ve done analysis of the networks, and you’ve got a great demilitarized zone within your agency, like we do at USDA.”

By using zero-trust principles, Rojas said, “We can do continuous monitoring of those networks and the tools.” 

“So it strikes a balance,” he said. “It’s going to take some out-of-the-box type of thinking that we’re still trying to develop today, and I think it’s going to take vendors to really up their game in terms of implementing AI in their tools that we buy.”

Rojas said America needs to “start coming around and coalescing around development of some rules and regulations,” however, and said he “kind of” disagrees with what he described as the administration’s position that AI companies should have “free rein” to develop unhindered.

“I think we’ve had good regulation in the past — railroads, nuclear energy and so forth,” he said. “We gotta come together as a nation and think about regulating so that we can use these tools in a safe way and protect against these bad actors,” referring to China, Russia, “and all these nefarious nation-state actors.”

Ajay Phogat, senior advisor to the chief information officer at U.S. Customs and Border Protection, is for humans in the loop and said the threat AI poses to security “as it stands today … is real.”

“All the pillars that we have, things that we have thought about as being extremely secure, become meaningless, right?” he said. “If the barbarians are through the gate, you’re done for, right? That’s what it is. They are at the gate right now.”

He continued: “I don’t know if AI is going to destroy us all, but if you’re going to give a gun to a child, you are still responsible for the consequences. And that’s where we are at with this.”

Other officials from other agencies who spoke at the summit said their departments are growing and accelerating with AI, with governance.

Joe Croghan, associate director of informatics at the National Institutes of Health’s Office of Cyber Infrastructure and Computational Biology, said right now the institute is building an “AI deployment infrastructure where you can deploy policies, data governance, financial governance, and so forth. Because I think if you don’t do that now … it’ll be like herding cats.”

Croghan said NIH naturally has a pretty high tolerance for risk by its nature of being a research body, and AI has helped it get to the point of failure within days rather than months. He said a year ago they were using AI for personal productivity, and have pivoted to focus on “purpose-built applications.”

Vendors and contractors having flexibility to the changing needs of government agencies and the rapidly changing technology is key for the Library of Congress’ director of digital strategy, Natalie Buda Smith.

“AI is really more like a utility than a capital investment,” she said. “We found that a lot of tools, as many have said here, have improved dramatically over the last year.”

Buda Smith said participation in LOC’s AI sandbox experiments has grown so much that “we often have too many proposals, but we give them a lot of support, so we’re limited in what we can provide,” while stressing that governance is essential and needs to be “active.”

“Sometimes when we are deploying or rolling out a new tool, it’s really exposing the ways that we need to communicate better and work better together,” she said.

Integrating AI experimentation into employees’ workflows instead of adding it as extra duties is helpful to increase participation, she said, and making sure that a team with a wide set of skills beyond just data scientists also helps. She said they have a role called “AI ops because we are supporting people building use cases in our platforms, and they need someone to do the operational work,” for example.

Some agencies like the State Department are integrating AI into training, said Gray Barrett, acting director of super intelligence workforce enablement at State’s Bureau of Diplomatic Technology and Center for Analytics.

Barrett said new foreign service officers now have an “SI module” giving them the rules of the road, tools to use and ways they should not be used. 

“Literally from day dot, they’re able to know, ‘OK, this is how I should frame my thinking about foreign policy as I go out into the world,’” he said.

Barrett estimates the department’s AI chatbot StateChat saves people about two hours a week on average, “which really adds up in a big federal agency,” he said. Their media monitoring tool, North Star, also saves time, he said.

Some agencies like the National Nuclear Security Administration are only using specific models, which Bill Watkins, the director of federal IT management, said is “purposeful” so they can “trust exactly what these models are going to produce,” with “continuous monitoring.” 

However, the NNSA is having some “bottlenecks” with finding data center space, power, chips and hardware, some with “long lead times, six months in some cases, for us to get some of this technology,” he said.

A self-described auditor and double-checker, Kevin Walsh, director of the IT and cybersecurity team at the Government Accountability Office, said that because AI tech is moving very fast, it is forcing the government to move quickly on cybersecurity. 

“The attackers are using it, and you really need to use it on defense if they’re using it on offense,” he said. “We always used to say in cybersecurity: it’s not a matter of if, it’s when you’re going to get breached. That’s absolutely still true, but now you’re going to get breached, and you’re not even going to know it until months later if you’re not using AI to be on that spectrum.”

Walsh is concerned the government can’t keep up. 

“I think the government is over its skis,” he said. “We don’t have enough oversight or ability to truly understand the technology and what it’s doing, and not just what it’s doing today or a month from now.”

While AI speeds onward, the lack of definition for “legacy systems” comes to the forefront, which matters quite a bit to a government that doesn’t “get rid of old things,” he said. Proactivity to keep the update cadence rolling is paramount now, he said.

“These tools are great, they’re awesome,” he said. “But we also need to be going in with our eyes open as to how we make sure that they’re doing the right thing over and over and over again, and that we do have the expertise and the capability to say, ‘that’s not right.’”