The Department of Veterans Affairs is leveraging artificial intelligence to address a large backlog of disability claims, but watchdog agencies and Congress raised concerns about maintaining governance and staffing during a House Veterans Affairs subcommittee hearing on Monday.
With 80% of the Veterans Benefits Administration’s 600,000 pending claims stuck in the evidence-gathering stage, Robert Orifici, VA’s acting deputy chief information officer, said AI could help the agency make “faster and better decisions” with the help of humans.
“All disability claims will be determined by trained VA personnel, not AI or automation,” he said. “These tools support human decision-making, not replace it.”
But Democratic Nikki Budzinski of Illinois, ranking member of the House Veterans Affairs Subcommittee on Technology Modernization, said the reduction to 2,700 claims examiners starting in January 2025 has made it increasingly difficult to keep humans on top of information.
“There is no world in which there is no impact on the speed with which veterans receive disability claims or the effectiveness of the VA’s ability to implement modern cloud-based IT systems for VBA. To stay ahead of this topic, letting computers do the work will not solve the talent shortage,” she said. “AI and automated systems often make life harder for claims processors.”
Additionally, Budzinski said, “so-called fixes pushed by the VA, such as automation and piloting of artificial intelligence tools, often create inaccurate information, complicate matters, and slow production,” leading examiners to view themselves as “unpaid software testers rather than claims examiners.”
“VA routinely punishes employees who fail to meet production standards, yet fails to give them reliable tools to do their jobs. This is unacceptable and will only lead to further employee exodus,” she said.
The Comptroller’s Office was also concerned about how the VA was using AI, citing the agency’s past technology “deficits” and concerns that its use was outpacing governance. Witnesses before the committee said the rollout of electronic medical record modernization would also expedite the insurance claims process by making it easier to transfer medical record data.
“The VA has responded to some of these challenges and made real improvements,” GAO Chief Scientist Sterling Thomas said of technical issues in the VA’s disability compensation program. “However, our previous research shows that despite these efforts, VA is not consistently meeting improvement goals.”
The Veterans Affairs Office of the Inspector General has also been in trouble for the VBA recently, with an April report showing that 8,000 automated pension and fiduciary service decisions and letters omitted favorable findings and provided incomplete summaries of evidence.
Thomas also said he was “optimistic” about some of the VA’s other uses of AI, such as a payment redirection fraud model to detect fraudulent direct deposit changes, but said the agency “needs to be thoughtful about the implementation and oversight” of AI. Thomas recommended that the Department of Veterans Affairs use GAO’s AI Accountability Framework for guidance.
“Broadly speaking, AI is still in its early stages of development and implementation, and rapid deployment without an intentional focus on governance is already having unintended consequences,” he said. “Before we pour data science into a problem, we need hard, reliable, truthful data and humans in the loop to ensure the trustworthiness of the data and the application of the technology.”
On the other side of the aisle, lawmakers and officials agreed that the current claims system is overburdening veterans and that something needs to change instead of leaving everything to robots.
“Instead of a shift, the men and women who serve in the service face incredible burdens to earn the benefits they earn. They are forced to become their own private investigators, couriers, and managers,” said Subcommittee Chairman Tom Barrett (R-Mich.). “The use of AI and automation requires quality assurance policies that ensure that human decisions are always made in the process.”
