Getting AI to enhance government work will take serious R&D funding, officials tell Congress

As a House subcommittee continues examining the government's relationship to artificial intelligence, members hear a simple message: Agencies can't make the most of it without the right amount of funding.

Research and development funding is necessary to spur innovations in artificial intelligence that could eventually help make the federal government more efficient, officials told lawmakers at a hearing Wednesday.

The hearing was the second in a series of three on AI by the House Oversight Subcommittee on IT. Subcommittee Chairman Will Hurd, R-Texas, said the purpose is to gather insight on how AI applications can make the government more efficient.

“First, [AI] should make every interaction an individual has with the federal government take less time, cost less money, and be more secure,” Hurd said in his opening statement. “Second, AI should produce efficiencies and cost savings that will help us do more for less money and help to provide better, more transparent citizen facing services.”

In her opening remarks, ranking member Robin Kelly, D-Ill., expressed concern about proposed cuts to R&D across several technology agencies under President Donald Trump’s fiscal 2019 budget. Kelly added that Trump administration’s immigration policies are driving away talent and brain power. She also cited statistics suggesting that China could outpace the U.S. in R&D spending as early as next year.


“The future of U.S. innovation is at stake. This should be a cause of concern for everyone,” Kelly said. “Unfortunately this administration’s science, immigration and education policies are all working together to reduce the U.S.’s lead in AI technologies.”

There was a general consensus on the panel that more funding for R&D is always welcome. For his part, John Everett, deputy director of the Information Innovation Office at the Defense Advanced Research Projects Agency, said that DARPA is satisfied with Trump’s proposed budget.

“There is capacity to do more,” said James Kurose, an assistant director at the National Science Foundation.

In a string of rhetorical questions, subcommittee member Gerry Connolly, D-Va., pointed out that many technological advancements in modern history either originated from or have seen significant development through federal agencies like the ones represented on the panel. He expressed concern that cuts to such research might eliminate potential advancements in the future.

“You four represent the face of the government that has transformed the world in R&D investment,” Connolly said. “When we say we’re going to cut a couple of billion dollars out of federal R&D, and I look at this record, I tremble. What are we cutting? Is it the next GPS? Is it the next drones? Is it the next human genome project? Is it the next internet? We don’t know, but the opportunity cost, I fear, is enormous.”


Hurd sought input from the panel on how to make it easier to get managers in government agencies to consider AI when looking for solutions.

“When we talk to operators in the field and they’re looking for solutions, they don’t necessarily say ‘I need an AI solution to my problem.’ They come to us and say ‘I need a new widget’ or a new ‘this,’” said Doug Maughn, who heads the cybersecurity division of the Homeland Security Department’s Advanced Research Projects Agency. “The operations community don’t know they need an AI-based solution. But if you give them a solution that solves their problem, they don’t care if it’s AI-based or not. They’ll use it. They’ll deploy it.”

Hurd also gauged the panelists’ ideas on what an AI “moonshot” would be. “What is the equivalent of going to the moon with artificial intelligence?” he asked.

Panelists responded by saying they look forward to a movement from narrowly focused AI applications to more general ones. Kurose said that applications such as image classification and speech recognition are seeing significant advancements, but that their applications are limited.

“If you look at, for instance what an 18-month-old child and how the child can transfer learning from one environment to another, how a child can understand intent and meaning—that’s really the grand challenge. General AI still remains a grand challenge,” he said.


Everett echoed that sentiment in saying that AI could be more efficient and have broader applications.

“Right now we know that people learn by having as few as one examples. Yet we need terabytes of data to get our systems to learn. We may look back at this time as the era of inefficient machine learning,” Everett said. “So a moonshot might take us to the point where computers actually do understand us in ways that our tools today don’t. But what we have today are tools.”

At a hearing last month, the subcommittee heard from the private sector about how to spur innovation in AI. A hearing in April will focus on to best regulate the use of AI.

Latest Podcasts