The Department of Veterans Affairs (VA) is planning to conduct a series of artificial intelligence (AI) technology “sprints” following the Biden administration’s Oct. 30 release of its AI Executive Order that features a broad mandate for Federal agencies to pursue responsible development and deployment of AI tech while mitigating risks that the technology may pose.
Dr. John Scott, physician informaticist in the Clinical Informatics and Data Management Office at the Veterans Health Administration (VHA), talked about the planned sprints today – and the vast potential of AI development at the agency – during a keynote presentation at the Red Hat Government Symposium 2023 event in Washington, D.C.
Speaking to the direction issued to agencies in the new executive order, Scott said those are “90 percent about Federal agencies must establish trustworthy processes … but it’s ten percent ‘start actually doing something.’”
“So, there is an incentive there, a direction to Federal agencies to start doing things,” he said.
“We already are doing things [with AI],” but in response to the executive order, VA will undertake a “couple of tech sprints a year,” Scott said.
Shortly after the order was released, VA announced the first of those sprints involving the development of AI-enabled tools that will allow clinicians to spend less time on non-clinical work – such as writing up notes following appointments with patients – with the aim of reducing healthcare worker burnout rates. Winning teams in the sprint effort will receive $1 million in total prizes, VA said earlier this month.
Scott said during today’s keynote presentation that current sprint will involve two use cases “about reducing provider burden where AI large language models can be applied” by employing AI tech to help the provider write a first draft of a note after meeting with a patient.
Writing up notes to capture the details of patient appointments, Scott said, can take as much time – if not more – as the time spent with the patient during the appointment, and is something that many clinicians “have complained about since they first started being required to record what they did … back before the Civil War when they were writing on paper.”
Ben Cushing, chief architect, Health and Life Sciences at Red Hat, who moderated the keynote discussion, suggested that VA tech sprint can be a good example for other Federal agencies that will be going down the same road of AI exploration.
“I suggest for every Federal agency represented here, look at the VA and really consider what’s going on there because it’s actually laying a bit of a roadmap for how to incorporate AI practice, start to embrace trust, build the foundations that are required to be successful, and also take lots of baby steps towards this very bright future that we have,” Cushing said.
Beyond the tech sprints, Scott explained that VA’s path forward on development and deployment of AI technology presents a unique opportunity to figure out how the technology can work with vast amounts of legally protected data.
VA, he said, “is certainly not a world leader in technologies.” But what makes “an almost unique opportunity at VA is the volume that we can do this with and apply to healthcare as a HIPAA (Health Insurance Portability and Accountability Act)-covered entity.”
“It is our mission to look at the information that has existed in our delivery of care to veterans, and do better – do performance improvement,” he said.
“We can do big data in the VA as a HIPAA-covered use case for performance improvement – that’s what makes it such an exciting opportunity for the VA,” he said.
He explained that VA has more experience than many agencies on the AI discovery path, having launched an agency AI working group following the Trump administration’s release in December 2020 of its own executive order to promote the use of trustworthy AI in the Federal government.
Scott explained that VA had a “huge lead” over other agencies in moving to comply with the 2020 order because by virtue of its mission the agency had in place “some of the fundamental underpinnings necessary” to explore development of responsible AI tech.
“VA has decades of use of electronic medical records, for example, and it has a budget to provide care to millions of veterans so it has an enormous volume of data that it that it must manage,” Scott said, adding, “for that reason, [the agency] has been struggling with the fundamentals of enterprise data management for a long time.”
Also as a result of the 2020 White House order, VA created an interoperability leadership group that “has been working on how do we improve standardization and data quality so that we can share data between components in the VA,” Scott said.
He said that leadership group, along with the National AI Institute, first created a tiger team to tackle AI ethics, and that team later evolved into VA’s AI work group. And as a parallel effort, he explained that VA has been hard at work in recent years developing a data governance structure with the help of the addition of the chief data officer position in 2019.
As a result of those efforts on data management, Scott said, the agency’s AI work group “didn’t have to start from scratch.”
“The lesson from that is you don’t start from scratch to start trying to govern AI,” Scott said. “You start thinking about what data governance experts, what data governance capabilities, who is working on that already, [and] incorporate them into creating infrastructure that then AI is a special case for.”
“We’ve been wanting to do big data analytics as long as I’ve been involved at the headquarters level in DoD [Department of Defense] and VA,” he said. “We’ve been trying to create the technology and governance structures for that for a long time,” he said, explaining that the VA AI workgroup “lays on top of that, for the special governance that’s necessary” for development of AI use cases.
