It won’t be in the Olympics anytime soon but Oki Mek considers artificial intelligence “a team sport.” As the chief artificial intelligence officer for the Department of Health and Human Services, Mek may be a little biased, but as his agency works through its AI strategy — released in January — collaboration and knowledge exchange will be paramount.
The strategy aims to promote AI adoption, and to ensure that algorithms are fair, legal and ethical. Three core pieces of the strategy are adoption and bringing the entire department up to speed on the language of AI; scaling best practices, and accelerated adoption. As for the first piece, Mek said culture change plays a pivotal role.
“The main risks here is not AI itself, it’s not the technology itself, it’s more of a culture shift. It’s getting away from that industrial culture, more into a data-centric digital culture. Because the two obstacles in AI that we’re going to face [are] data acquisition, because health data is quite heavily regulated, but also data processing — formatting the data and tagging the data,” he said during a web event hosted by ACT-IAC on Tuesday.
It’s important to educate and train all of HHS on what AI and machine learning are — their respective principles, what are bots, what is robotic process automation — to ensure everyone is on the same page before the organization moves forward. Then, HHS can see what is valuable and risky with regards to AI, he said.
HHS has 107 AI use cases in flight right now, and Mek’s office wants to see where commonalities exist to bridge some of those projects and avoid duplicative efforts. These can be things such as grants, supervised and unsupervised learning, reinforcement training or even genomic research. He noted that HHS is working with a much smaller budget than, say, the Joint Artificial Intelligence Center, the Defense Department’s AI Center of Excellence. But budget constraints can help plan better and he said the JAIC has more hard authority than his own office in implementing AI strategy.
“Get people to come together and learn from each other and share, but also there may be opportunity to leverage and say, ‘we don’t have this, we don’t have that much funding on our side. But you’re already doing this and you have a lot of funding — maybe we could jump on your project?’ Because maybe the data sets of the use cases are very familiar?” he said.
With health care comes the question of privacy, and particularly what is or is not prohibited under Health Insurance Portability and Accountability Act of 1996 (HIPAA). Having a “data culture,” Mek said, means that requests for someone’s health data include clear justifications for that request; being able to demonstrate the benefits of properly categorizing data, and what exactly it will be used for, will help build that culture. From there, it gets easier to “negotiate for data,” tagging and formatting it, and processing it for AI technology. Mek admitted it’s a process but it accounts for 80%-90% of the work and cannot be skipped.
“We can’t just jump to the new toys. I think we need to focus on culture and shifting the culture in the right direction,” he said.
Aside from personal health information, HHS also deals with population-level and trend data. As the department builds out AI use, Mek said it helps to identify where that data resides. He doesn’t want to overstep boundaries with the chief data officer, as well.
“Let’s say I have this AI use case and I’m pulling data from 20 systems. It’d be good to have that understanding where that data, what data is in that system and [where it] resides,” he said. “But also you could tag it to say, OK, this is [Personal Identifiable Information]-related, this is HIPAA-related — that would be helpful for somebody that wants to start working on AI. But right now, it’s hard to understand where the data resides, though. Who has the data?”
The Office of Management and Budget is requesting another data call around what HHS is doing with AI. The community of practice, and giving some guidelines for general direction could be helpful to agencies which lack strong authority or funding to expand AI implementation, Mek said.