Nearly a year after President Donald Trump’s executive order on artificial intelligence, the White House has set some rules of the road for how agencies should regulate the private sector’s use of AI.
As for the federal government’s own use of AI, agency leaders continue to build a foundation for greater adoption on use cases that include enhanced cybersecurity, chatbots in call centers and predictive maintenance for military vehicles.
Insight by Tableau: Executives will discuss how data has driven the success behind their hiring and retention strategies in this exclusive executive briefing.
To scale up these efforts, the General Services Administration and Federal Chief Information Officer Suzette Kent stood up a governmentwide AI Community of Practice last November. Since then, membership has grown to more than 400 members from 26 agencies.
The AI community of practice, said Steven Babitch, the head of GSA’s AI portfolio, is looking at building a searchable “use case repository” that would give agencies a playbook of examples where agencies have successfully deployed AI for customer experience, human resources, advanced cybersecurity and business processes.
Members of the AI community of practice, he added, plan to hold quarterly meetings, but have also considered setting up webinars, seminars, workshops and other training to keep federal employees in the loop.
“What we’re hearing from federal agencies is there’s a need to become more aware of the kinds of problems that they are tackling, and if one agency is tackling something that might be similar to something in another agency, how can we share what they’ve learned?” Babitch said Wednesday at a GSA event on AI in procurement. “What are the challenges? What are the successes? What are the pitfalls they can now avoid when one agency experiences that?”
Officials who play a central role to this community see AI as one day having a transformative impact on the way the federal government does business. But for this moonshot effort to pay off, experts say the barriers to wider adoption fall along with three familiar lines of effort – technology, data and the workforce.
As for which of these barriers remains the thorniest issue, members of this community are mixed.
On the technology piece, Chris Hamm, director of the General Services Administration’s Federal Systems Integration and Management Center (FEDSIM), said the government’s lengthy Authority to Operate (ATO) process remains the “number-one problem” of interacting with industry on AI implementation so far.
“It would be far superior that each agency had an environment that was created that was in between, that we would we would allow everyone to use data that’s been sanitized in some way,” Hamm said “That allows us to test the algorithm that’s not necessarily the same as the production environment, but doesn’t require the entity to go through all of the security controls.”
As for data, Agriculture Department CIO Gary Washington said that to meet the goals of the Federal Data Strategy, which plays a supporting role in the fielding of AI, the agency now has assistant chief data officers in every mission area, led by the agency’s CDO Ted Kaouk.
“We work with them on how to get them access to more data, to answer questions that they’ve never been able to answer before,” Washington said.
But in order to develop a workforce with the right data-focused talent, Washington said USDA has gone through several reskilling pilots aimed at bringing current employees up to speed.
“You need people with analytical skills that can analyze data and look at data in different ways,” Washington said. “So this has forced our entire organization to look at the skillsets of the current employees and what we will recruit [for] in the future … This is a journey,” Washington said. “It’s not something that’s going to happen overnight. You have to be strategic about how you do this. You’re going to have current employees that don’t want to do things a different way, so you’re going to have to try to pull them along as well.”
Rebecca Hutchinson, a big-data leader in the Census Bureau’s Economic Indications Division, said the bureau rolled out data science and machine learning reskilling efforts that, less than two years into the program, about one-third of her office has completed.
While that reskilling project gave employees hard skills in learning programming languages including Python and R, as well as data visualization tools like ArcGIS and Tableau, Hutchinson said those reskilling efforts also help develop buy-in from the workforce on new projects.
“I think we’ve all been a position where we’re doing our job and we realize an inefficiency, and we can identify it but we don’t know what to do about it,” she said. “Once you start training your staff with the skills, they are coming up with the solutions.”
As a through-line to all of these lines of effort, Keith Nakasone, GSA’s deputy assistant commissioner for acquisition management, said greater AI implementation requires agencies taking on a “soft skills” perspective to change management.
“We want to make sure that they’re highly engaged at the lowest level, the people who are actually doing the work and performing the work, so that we would be able to inject this on a phased approach, rather than this whole ‘big bang’ theory,” Nakasone said.
To make that possible, GSA’s Deputy CIO Beth Killoran said automation requires a conversation between the IT workforce and front-line employees.
Of the half-dozen machine-learning pilots that GSA has in the works, Killoran said every one of them is driven by a focus on how to make the agency’s front offices more efficient.
“That kind of thought process is what has to happen from a business perspective. Where do they feel that they do not have enough resources and do and fulfill a particular mission or where they are having a subpar customer experience,” Killoran said. “And then we can look at what the data is around that particular capability, and where artificial intelligence could be potentially used to fill a gap or address a particular business need.”
At an upcoming meeting of the AI Community of Practice on Feb. 12, Babitch said officials will work on identifying “practice areas” where the agencies should focus most of their automation efforts.
Possible practice areas, Babitch said, include ethics, AI tools and techniques, and privacy and security. Once identified, he said the community of practice will form committees focused on addressing those break-out topics.
“It must be user-driven, so we want to collectively define what are the practice areas that we think are most relevant and important,” he said.
But for these concepts to make progress in agencies, Babitch said workforce development focused on AI remains “arguably one of the most important aspects” of this work.
“If they don’t understand these technologies at a deep enough level, it’s arguably far more difficult to actually implement these projects in an effective way that really delivers the value that they should,” Babitch said.