Generative AI’s fleet-footed evolution is causing quandaries for federal acquisition
Government leaders and federal contractors said the rapid development of generative AI makes it difficult for agencies to determine what types of tools it needs.
As the federal government tries to grasp on to the burgeoning utility of generative artificial intelligence, the technology’s fast-paced development is likely to remain at odds with the plodding pace of acquisition regulations.
“I know from my experience that for the government to perform better, they've got to do a better job of taking advantage of what contractors bring. And that's a number of attributes. It's access to the latest technologies, as the government itself takes a long time to incorporate. You can get it faster,” said David Berteau — president and CEO of the Professional Services Council, a trade association for federal contractors — in an interview with Government Executive.
At a conference hosted by PSC on Thursday, speakers discussed ongoing issues with government acquisition of generative AI tools and programs, such as matching company products with government needs, possible reforms to acquisition rules and the meta problem of using AI in contractor proposals. But attendees offered few clear-cut solutions.
What does the government need?
Udaya Patnaik, a chief innovation strategist at the General Services Administration, said during a panel at the conference that the unknown potential of generative AI makes it difficult to plan for how to use it.
“There's a lot of cases where we're trying to wrangle a constantly evolving technology. And the challenge with that is that in many cases, we don't always know what it can do, what it can't do, what we think it can do and what it might not be able to do,” he said.
Patnaik urged potential contractors to be transparent about the opportunities and limitations of their products, which Geoff Sage — a procurement director at NASA — said has not been his experience.
“What I’m seeing as a buyer of this type of technology is I’m being sold the world and when I go to look at it, it’s not the world. It’s just a little dirt path on the corner,” he said during the same panel.
On the other hand, Berteau argued that it’s difficult for contractors to sell to agencies when government officials are unsure about what specific technology they’re looking for.
“What is it that the government needs and wants and how will they use it?...I don't know that companies know how to answer that question without the government being able to be part of that answer,” he said. “And right now, I don't see constancy and consistency coming out of the government [with respect to] ‘This is what I'm going to need.’”
Acquisition reform
Berteau also believes that the normal timelines for purchasing a technological tool under a contract can’t keep pace with the rapid development of generative AI.
“Those timelines are so out of sync with the evolution of the technology itself that, at a minimum, there’s a disconnect, and it may be an unsolvable disconnect,” he said.
He expects federal leaders will need to figure out how to update contract requirements for generative AI.
Along those lines, Dorice Kenely — a policy manager at Amazon Web Services who moderated the panel — compared using some longstanding acquisition principles to putting a “round peg in a square hole” with respect to AI.
Sage emphasized that federal contracting officials are looking more at issues like privacy rights and copyright infringement.
“All these things…existed before but now the proliferation of — or at least the illusion of — use of AI in a lot of different spaces is causing our lawyers to start freaking out a little bit,” he said.
Patnaik excitedly told attendees about a GSA website with a resource guide intended to assist federal officials with acquiring such technology.
“We put it up as a digital site, not as a PDF, because we know it’s going to have to evolve,” he said.
AI-written proposals
Officials also have to confront contract proposals that are themselves created by generative AI, which Patnaik said has been happening for “a little while.”
“Somebody had submitted a bid, and then somewhere in the middle of the proposal, it just went off on this random paragraph having to do with zoo animals,” he said, encouraging humans to still proofread.
Perhaps an example of the government being slower than the private sector to adopt new technology, the panelists said contracting officials will not be using AI to evaluate bid proposals anytime soon.
“I’ve not seen anything that I’m going to go into a negotiation with a company and just say ‘AI versus your AI. Go at it,’” Sage said.
But Berteau thinks the government will need to address the problem sometime soon.
“What are you gonna do when you get 3,500 proposals all written by some generative AI system? How are you gonna evaluate them? How are you gonna know what's real and what's not?” he asked. “And [the government doesn’t] have a good answer for that other than maybe they'll sort of facetiously say ‘well, we'll use AI to evaluate the AI.’ I'm not quite sure that's the world we want to live in.”