GSA AI chief says USAi has become like ‘code development’ between agencies
Three months after its launch, the General Services Administration’s USAi tool is not only being used for artificial intelligence testing in the federal government, but also for collaboration and development between agencies.
Zach Whitman, GSA’s chief AI and data officer, provided an update on the evaluation suite Friday at the ACT-IAC Imagine Nation ELC event in National Harbor, Md., telling attendees that USAi has given the agency “practical, hands-on data” on how the federal government is using and adopting AI into its workflows.
“I would argue that the USAi play has turned much more into a code development across multiple agencies,” Whitman said during a panel discussion. “It really isn’t a pure GSA play, because we’re all learning from each other. That’s been the beauty of being able to take GSAi and open it up to other agencies in terms of, ‘what do they need it to do? How do they like to look at the data? How do they like to control the organizational usage of these tools?’”
The USAi tool builds on GSA’s internal chatbot, GSAi, which the agency rolled out to employees earlier this year to provide access to various enterprise AI models. By expanding to the entire government, agencies can test numerous AI models to determine which is best to procure from the marketplace.
Whitman explained that the tool is essentially being plugged into business processes that have various “different forms.”
“Whenever you do introduce a new technology to your workflows, you have to start to build in place ways in which you can control that new technology, ways in which you can harness its true value,” Whitman said. “And there’s ways you could go into blind corners and ultimately waste a lot of time and energy.”
“For so long, we’ve been reliant on the market to tell us exactly how these things should be used,” Whitman continued. “With a tool like USAi, you can really get practical, hands-on data to suggest these are the ways in which we’re seeing this adoption.”
At its launch in August, USAi offered access to AI models from Anthropic, OpenAI, Google and Meta, but has since expanded to Amazon and Microsoft.
A GSA spokesperson told FedScoop that the agency is “actively partnering” with more than 15 agencies to deploy USAi.
During the same panel discussion Friday, David Shive, GSA’s chief information officer, laid out other examples of what USAi data has shown the agency in recent months.
“One of the things that we ask in the user community is to evaluate the quality of responses that comes back and then we analyze that over time,” he said. “One, because the projection of information in there, but it also allows us to rank … the industry providers of large language models that are being used.”
The feature allows agencies to evaluate the types of questions — from acquisition to facilities to technology — that one or another model might be best suited for, Shive explained.
“What that’s allowing us to do is reverse engineer back into the tools from intelligence so that we can have the problems upon engineering,” he added.
