The Department of Veterans Affairs is working to up its artificial intelligence capabilities, while also providing guidance to other agencies throughout the government from lessons learned.
Speaking Dec. 5 at the GovernmentCIO AI and RPA in Government event, the VA’s Director of Artificial Intelligence Gil Alterovitz said the agency recently finished an AI tech sprint that benefited both the department and industry.
“AI tech sprints are a way that we are looking [at] as an on-ramp by engaging with industry, with academia and others around some open data that [the VA] makes,” Alterovitz said.
The VA made data for the three-month sprint available to companies to see how companies can use that department data to develop AI tools. In the department’s recent tech sprint, the AI chief said it completed one project that matched more than 9 million veterans with clinical trials and experimental therapeutics, which also paired millions of Medicare beneficiaries with the trials. During that project, the VA learned that the natural language processing component couldn’t properly read large numbers in their numeric form, reading numbers like “1,000,000” down to the individual numbers.
“We were able to provide them input so that this will affect how machine learning and AI can be used for millions of people, not just this clinical trial application,” Alterovitz said.
David Maron, a statistician at the VA, said at the same event that the tech sprint undertaken at the VA is key to better the push-pull relationship between industry and federal agencies.
“There’s no commitment ... but it’s an opportunity for a relationship to be built where you work with your industry partner and they come to understand your problem in a better way because you’re closely working side by side,” said Maron.
In turn, the better relationship also affects the data. “An iterative sharing of data can help improve the quality of it,” said Maron.
The way the VA cleans and maintains its data now is a far cry from a few years ago, when he would manually going through databases to clean up data, Maron said. But AI, still in the early stages of adoption by the federal government, comes with its own challenges, such as the issue Alterovitz mentioned regarding natural language processing. Alterovitz said that the VA plans to publish the list of barriers his agency ran into during the tech sprint.
The department is also developing “about” three AI playbooks to serve as guidance for agencies, “as a way to see what are the steps toward having an agency work toward having pilots first" and moving toward other areas for AI development and scaling up from there, Alterovitz said.