The panelists took a close look at how to do that effectively, and where to focus.
“Operationalizing AI basically means that you’ve developed a real product that turned into a managed service, that has [service-level agreements] SLAs and performance metrics and security controls,” said Thomas Boon, director of the Division of Enterprise Information Services at the California Department of Corrections and Rehabilitation (CDCR).
Government’s predisposition to be deliberate may run counter to doing sprints and quick pivots. But Boon said there’s value to doing pilots and proofs of concept (POC), and to letting them fail.
“We did six POCs. And of those six POCs, I would say two of them were terrible because they didn’t work,” he said, recalling work the Office of Business and Economic Development did for the Office of the Small Business Advocate. “And for a wide variety of reasons, they didn’t work. And so … those POCs were effective because we knew what we didn’t want.”
If you’re thinking about starting somewhere, “start very small,” said Stephenson Loveson, CIO for the California Public Employees' Retirement System. Before you get to production, consider whether what you have is a genuinely repeatable process and whether it has virtual reality capability — but also think about “who you are as an organization, how you’re going to use [generative AI] GenAI and what your organization stands for.”
Define it, he said.
“So, for us, it’s squarely human-enabled AI. That means our policies reflect that. You, the user, are in charge of any output,” Loveson said. “GenAI means different things to different people. You have to get everybody on the same plane.”
Similar to beginning with a technology problem or need in mind, start your discussion about operationalizing AI with the outcomes you’d like to see, said Vidhu Shekhar, chief data and AI officer for the California Department of Transportation (Caltrans) — and remember, this is a team sport.
“AI, as much as about skill sets, at the end of the day, it’s about people taking an idea and pushing it uphill against all odds, funding notwithstanding, and making it a reality,” said Shekhar. He warned the capacity crowd not to be “enamored by the shiny object in the room.”
“The vision for Caltrans really is to be able to leverage this tool and this technology in a way that all 23,000 employees at Caltrans can use it in their own jobs,” he said. “And the first step to that is us being able to get — what do they do today and where are service challenges?”
Vision is key to sharing and securing data, he said. But so are governance and leadership, Boon said.
“If you’re looking for something that ensures that you maintain a certain level of security compliance, moves things forward, and it has a level of governance, you really need to find that leader or a team of leaders that can push it forward,” he said. “And to me, that’s key.”
