New research from General Dynamics Information Technology (GDIT) that tracks progress of Federal government agencies in their adoption of artificial intelligence technologies is finding that successful AI project completion timelines are narrowing, and that future demand for AI projects appears to be building around chat and coding-related functions.
Those are two of the takeaways from the AI in Full Bloom study released by GDIT on May 9, which surveyed 325 AI experts working across the Federal government in February 2024.
“In the wake of generative AI’s exponential rise beginning in 2022, we are witnessing the ‘full bloom’ of artificial intelligence across the federal government, marking a pivotal shift in how the technology is leveraged to serve the American public and our national security,” the report says.
The research reveals several primary findings, including the need for project scalability, the benefits and work entailed in incorporating data diversity, a 23 percent project failure rate due to misalignment between AI solutions and user expectations, and a strong belief that partnership efforts are required for successful AI projects.
GDIT officials involved with the research project said the finding of diminishing pilot-to-production timelines – to an average of about 14 months – was an “impressive” surprise based on some of their experience in the Federal government market.
“The reason we were asking this question is that we’ve heard a lot of folks in the industry, industry press, as well as sort of the common sense that getting from proof to production is really hard,” said Dave Vennergrund, GDIT’s VP of AI and data insights during a call with reporters this week.
“One of the first things that really surprised us was the average amount of time it took and that was 14 months on average for those that did make it to production,” he said.
“We’d expected to see a larger number and we have two thoughts on why it is so quick,” he explained. “The first is we think many AI projects are very specific to one kind of a task – one small task, not an enterprise activity, certainly not something at the scale of an ERP or payroll, but rather something like a fraud detection or an object identification, so very narrow scope … hopefully you can get something done quicker.”
“And that’s where we believe that our agencies are taking advantage of accelerators,” he continued. “They’re taking advantage of cloud-native services, tooling and other capabilities that they can build on. Rather than having to build the plumbing, they’re building on the plumbing.”
The research also shows strong hints of what may be coming from Federal agencies as they bring AI projects through the pipeline – 54 percent of those surveyed said projects currently under development have something to do with improving computer code, and 59 percent of projects involve the goals of “research and understanding,” which GDIT’s Vennergrund said may relate primarily to the creation of chat functions.
On the code project front, Colleen Kummet, program director and data science consultant at GDIT, said, “just from our previous experience with our government customers, they have a lot of production code in some very old software and so there is an interest in translating the older code in older software to newer – say Python or other cloud-based analytic tools code.”
“We know that they have these existing validated code bases that need translation, so we’re speculating that there is some activity in just code translation – generating code that that already exists but in a different language and translating that to a more modern language,” she said.
Regarding agencies’ project pipelines that show strong interest in AI projects geared toward research and understanding, Vennergrund said many of those projects involve creating domain-specific chat functionality.
“That’s the primary thing that folks are responding to in this bucket,” he said. “So building that chat tax return, or chat CMS policy, being able to interact with those and being able to ask questions like how do I fill out a form, how do I interact, what is covered, what is not covered,” Vennergrund said.
“Those things are going to be, in our opinion, those are going to roll out over the next two years everywhere,” he said. “You’ll see chat, domain-specific chats, almost across every agency.”
