Skip to Main Content

CDC Roadmap Outlines Agile Workstreams to Support Federal AI Development

Ongoing development and constant feedback are critical parts of the AI development process that requires culture change.

7m read
Written by:
CDC Chief Data Officer Alan Sim said Wednesday that many levels of CDC employees "wanted to be part of this enterprise process” of AI implementation.
CDC Chief Data Officer Alan Sim said Wednesday that many levels of CDC employees "wanted to be part of this enterprise process” of AI implementation. Photo Credit: GovAI Summit

The Centers for Disease Control and Prevention’s proposed roadmap for AI development incorporates collaborative, continuous workstreams that create a culture for innovation. Chief Data Officer Alan Sim sees the plan as one other agencies can adopt and one that can make impacts in environments where it often takes much longer to scale emerging technology.

“If we waited according to government time, by the time we publish something or put something in place, it would already be outdated,” Sim said Wednesday at the GovAI Summit in Arlington, Virginia.

“[Communication] is absolutely critical especially those individuals that want to utilize generative AI,” Sim added. “Part of that is the communication workstream, which is not only publishing it for data scientists to understand.”

The White House AI executive order directs agencies with developing standards for secure AI. The order tasks the Department of Health and Human Services, of which CDC is a subagency, to create programs analyzing harmful AI-related health care practices and developing resources to establish AI educational tools for health agencies.

Sim’s roadmap outlines a “Steering Committee Community of Practice,” which emphasizes bringing all workstreams to the community to define best practices and uphold standards for responsible AI, Sim said. He also noted that he envisions getting constant feedback from those implementing and using AI within the agency.

“Our programs really asked for engagement with various centers, institutes and offices that have data science experts. They wanted to be part of this enterprise process,” Sim said. “One way is through a community of practice, where you give the community an opportunity to share use cases, share approaches and — we’ve talked about this recently — how do we get scientists and data scientists to share failures?”

The roadmap would provide guidance for responsible AI development and use. Sim said that generative AI could particularly benefit from that guidance.

“We are in the process of publishing pre-decisional documents on generative AI guidance: about 15 pages of guidance, the do’s and don’ts,” Sim said Wednesday. “This is absolutely critical for those, especially, individuals that want to utilize generative AI.”

Sim also created the AI/ML Consultation Group (AMC) to serve as a sounding board for groups across the agency to get feedback on proposals about AI technology use.

“It’s difficult to make a decision or have ideas about what your options are [about processes]. So part of the goal of the AMC group is, on an as-needed basis, to provide feedback,” he said.

Additional responsibilities for the AMC include advising on developing and deploying new use cases leveraging ML and Al for better public health outcomes, assisting in the identification of promising use cases that align with the mission of different centers and supporting the governance process.

To minimize risk in AI development, Sim emphasized the need for guardrails in the process, including have “existing infrastructure to innovate broadly.”

“We had already established an enterprise data analytics and visualization platform where we had some machine-learning capability. And so during the COVID response, we leverage that platform for big data,” Sim said Wednesday. “When you can innovate and try things, and things break or fail and it doesn’t result in harm to the public, go ahead. But when you recognize it’s something that potentially could be dangerous, then you have to create that safe environment.”

“In some cases, you have to limit it. Test it out and learn from those activities,” Sim said. “Then you increase the aperture, the scope and implementation to do more. But it is a challenge.”

At GovCIO Media and Research’s Health IT Summit in September, Sim highlighted the nature of AI’s pace of evolution and how CDC is adapting to it cooperatively.

“In AI, it’s there, you really can’t future-proof,” Sim said. “It’s more of projecting and anticipating the change management.”

Related Content
Woman typing at computer

Stay in the know

Subscribe now to receive our curated newsletters

Subscribe