Agencies Are Scaling AI by Turning to Experimentation

Federal agency leaders are enabling broader experimentation in artificial intelligence amid recent White House executive orders and guidance for agencies to ramp up development and use of the technology.

“We’re all going through the same things, thinking about governance,” said State Department Acting Chief Data and AI Officer Amy Ritualo at GovCIO Media & Research’s AI Summit in Tysons, Virginia last week. “What are you doing and what tools are you developing? What infrastructure are you creating? What vendors are you using? How did you create that chatbot? … That collaboration has been incredible.”

Environmental Protection Agency (EPA) CAIO and CTO Niki Maslin noted that implementing AI at her agency requires a balance between technical specifications and the specialized needs of scientists and program offices. She noted the agency is focusing on mission and how AI enables it, such as using AI to expedite the chemical review process, for example.

“We focused an approach that starts with good governance. We want to have good, responsible AI that’s been able to deliver for the mission across a number of different technologies,” Maslin said.

The EPA is also pioneering an “ambassador” program to ensure AI expertise is available at every level of the organization.

“We have folks at the ground level, the grassroots level, who are using it. We start to talk about some of our senior folks across the agency. We really wanted to make sure that they could act as ambassadors for technology,” said Maslin.

AI Strategies for the Workforce

State is in the midst of a plan to adopt AI across 200 global missions and 80,000 employees. Ritualo said that State has already issued its second data and AI strategy and is currently maintaining a comprehensive AI inventory.

One of its most successful tools is “StateChat,” which currently has over 50,000 users.

The department created decentralized spaces to explore and troubleshoot tools like this, including “office hours” at 10:00 p.m. or 5:00 a.m., accommodating embassy staff across every time zone globally.

The agency also maintains an enterprise “use case library” and a repository of three-minute videos where employees demonstrate how they specifically apply AI to their workflows.

Ritualo noted the effort is to focus training on AI being an enhancement to existing workflows rather than a mandatory burden.

“This isn’t the case of, ‘we’re building a tool that you have to use,” she said. “This is a tool that you should be using to improve your workflow processes.”

She said that learning from peers has been more effective than top-down mandates.

“The econ officer in Niger that is using AI and the econ officer in Nigeria who is using AI — they want to hear from them,” Ritualo said. “I direct a team of data scientists, engineers, product owners and project managers,” she said. “The econ officer doesn’t want to hear from me or my staff how they should be using AI.”

The Case for a Unified AI Framework

Enterprise-wide architectures benefit agencies in being able to add flexibility to their tool stacks. Lumen Senior Vice President of Public Sector Josh Finke noted that without an enterprise-wide architecture, efforts like scaling AI can become siloed and break down.

“At the end of the day, AI is one of the most classic examples of what goes in, what comes out,” he said. “The quality of the policy, the quality of the decision and, most critically, enterprise-wide decisions are what steer success.”

Finke noted a “consortium model” would enable agencies and private industry to design a secure, national AI infrastructure. While industry has deployed millions of miles of fiber for hyperscalers, federal agencies often struggle to deploy simple circuits because of the lack of a unified framework.

“There is no one company that any agency is leveraging for all of its AI needs or all of its computer infrastructure needs. There is no one company that’s going to be a solve across any mission,” Finke said. “The reality today is that there needs to be a consortium model where it’s the right players in industry and the right agencies coming together and then evaluating what tools can be leveraged at once.”

Where the AI Action Plan Fits

A major component of the federal AI Action Plan in particular directs the Labor Department to build sustainable talent pipelines and close the gap between industry innovation and public services.

“There’s so much innovation going on in the private sector, and so there’s such an opportunity to partner there,” said Chief Innovation Officer Taylor Stockton at the event.

Stockton said that internal literacy and external hiring must work in tandem to create a resilient workforce.

“This is a question of talent pipeline building,” said Stockton. “How do we not only think about the transformation with our existing workforce [that encourages] AI skills development that’s going to allow us to be most effective? … How can we complement that with talent coming in externally?”

Ritualo noted the need for more agile hiring mechanisms to address the need for specialized, short-term talent.

“How can you bring in quickly technical expertise in certain areas that we may only need for two weeks, and I don’t want to pay a lot of money for it,” she said.

Despite these challenges, Ritualo said the executive orders have increased interagency cooperation.

“I have never seen more coordination and collaboration across the U.S. government on any other issue than AI,” Ritualo said.