The CIA is taking a ‘crawl, walk, run’ approach to GenAI

The normally secretive Central Intelligence Agency isn’t shying away from its use of generative artificial intelligence tools across some of its enterprise.

Speaking at last week’s ServiceNow Federal Forum, one of the agency’s technologists told a standing-room only crowd outside Washington, D.C. that the CIA has turned to generative AI in part to make intelligence analysts’ jobs easier.

“It’s not going to put our [intelligence] officers, it’s not going to put our employees out of business, it’s just going to get them to focus on what they need to be focusing on,” said Gary Novotny, chief of the ServiceNow Program Management Office at CIA.

Novotny said senior leaders at Langley have been emphatic that generative AI will help, not threaten its existing workforce, even as surveys indicate many Americans harbor fear the technology may take their jobs.

Novotny said generative AI can free up an analyst’s time by providing a “unification of a knowledge base.” For example, querying a generative AI bot could save time by providing answers without forcing a manual query against various databases or websites for open-source intelligence. Novotny said that leaves internal generative AI users with more time to spend “on the really impactful mission-centric work.”

Novotny’s remarks come months after CIA officials confirmed the agency’s development of a ChatGPT-style AI for use across the intelligence community. Per the report, the large language model will provide summations of open-source information — with citations — and chat with users. Novotny said the CIA’s recent generative AI efforts go beyond hyper-automation.

“We’re not using that data in the background to really start generating answers to [user] questions,” he said. “It’s not just identifying what you need to automate, it’s actually querying those databases.”

While the CIA is moving quickly with the nascent technology, Novotny offered advice to organizations considering generative AI, advocating a “crawl, walk, run” approach.

In the “crawl,” or earliest stage, he said organizations and agencies should examine business practices, solidify back-end processes and start small.

“You want to start small, you want to start with proof-of-concepts, but you want to take the entire organization — you want to take a look at the entire organization before you really start implementing because you’re going to lose that customer experience that you want if the person searching for something and [the generative AI] just gives [them] crickets,” Novotny said. “Don’t try to create the whole thing at once.”

Communication is key, he added, suggesting two-way communication up to senior leaders and down to users helps foster buy-in and make improvements if necessary.

“Communicate with your senior leaders and show them the automation and show them the time and money that they can save,” Novotny said. “But then also communicate to the workforce – that we’re not taking your jobs away from you, but that you’re giving back to what they need to be doing.”

Novotny said organizations must also be mindful of the infrastructure required on the back end to deploy generative AI. The CIA uses an on-premise platform to handle generative AI queries, but operating that platform isn’t free.

“So you need to understand that infrastructure,” he said. “Talk to your budget folks and make sure that you have that infrastructure set.”

Lastly, Novotny recommended organizations take stock of metrics and performance. That can be as simple as speeding up a query or saving time.

“How long did it take you to do this before? How long does it take you now?” he said.