The AI journey at State

The State Department is one of the early adopters of workplace artificial intelligence, with use cases in place covering translation, media summarization and other aspects of the production of written materials for use in diplomacy. We spoke to Matthew Graviss, who leads data and AI at State, in late August to discuss progress on delivering AI-powered tools to the agency's 80,000+ workforce. This article has been edited for length and clarity.

Nextgov/FCW: State really seems to have gone all-in on AI. You're executing on an enterprise AI strategy that is involving people from every corner of the agency. Can you give us an overview of the strategy and an update on where you are right now?

Graviss: It's really an exciting time at the State Department when it comes to using technology to enable diplomacy. We've had an exciting year since the AI strategy got signed in October of last year. We started moving quickly, as you suggested at the top, we established an AI conference governance board. I co-chair the governance board with the deputy secretary. We have an AI steering committee that is really focused on driving the implementation of the AI strategy for the department. 

There are a lot of folks across the department that are really energized and really engaged in bringing these kinds of capability to our workforce. When you look at the strategy, it can come across as pretty straightforward, and that's by design. We have a goal on infrastructure. We have goals on culture, communication and training. We have a goal on policy and governance and ensuring that we have the guardrails around the technology that we employ. And then we have an innovation goal that's focused on really that last mile delivery.

Nextgov/FCW: How are you onboarding AI tools?

Graviss: The partnership with the chief information officer, Dr. Kelly Fletcher, has been game-changing when it comes to delivering secure and safe technology for the department. We went through an experience of open source, generative AI models and large language models. We went through an experience of onboarding in a secure way, through the FedRAMP process and and through our own security processes, to bring models into the department. 

Nextgov/FCW: You're the chief data officer and the chief AI officer. Why does it make sense to have those roles intertwined?

Graviss: I'm excited to have both hats. Good AI rides on good data. You know, other parts of the State Department will come to us and say, "Hey, we need this AI model to do X, Y and Z." We say, "Sure, send us the data," and the data is not ready. And so data readiness, in our view, has been key. We have the Data.State platform that we've been maturing over the last several years. It's designed to create data access across the department to enable more data informed diplomacy. We've done extensive work on data sharing and data governance to ensure access, to ensure accountability, to ensure transparency. 

The other thing I'll point out is .. the CIO and I focused on AI use cases where we were using publicly available tools on publicly available data, like media summarization and things like that. That started to increase the culture and the literacy across the department, while at the same time we were executing those other goals of the strategy, getting the infrastructure ready, and preparing our data for use in AI. 

Nextgov/FCW: Secretary of State Antony Blinken spoke at an event in June, which you moderated, to introduce AI to agency employees. What does it say about State's commitment to AI that the agency head is prioritizing employee engagement? 

Graviss: It is a really big deal when you can get your cabinet-level agency head to spend time, not just during that event but leading up to the event, and getting to know the progress that we've made, and getting to see some of the tools in action. What leadership spends their time on is a strong indicator and mover of progress. 

Nextgov/FCW: You mentioned in June at the event that you were getting a lot of feedback from power users of AI products and services. How is that sort of workforce response kind of driving your efforts and supporting your efforts?

Graviss: We've combined our delivery of technology with a very robust communications program and training program to really understand and educate people on how to use it, but also to create that feedback. So I'll give you an example: We really started doing a lot of robust cybersecurity testing, and ethics testing. As we grew, we started asking for feedback. So they started to not only spotlight where they're seeing it work well, but they're "grooming the backlog." to use some software terms. They are grooming the backlog of what should come next based on what they're seeing as the need. And so we've aligned our roadmap to what the employees need across the board.

Nextgov/FCW: As we close in on the 1-year anniversary of the public release of the enterprise AI strategy, what updates can you share?

Graviss: I'm really excited about one development on the workforce front. We have this bureau chief data officer program that we started a couple of years ago trying to get data leaders and AI leaders in each of the bureaus and headquarters to be as close to policy decision makers as possible. We have over the last couple years onboarded 15 bureau chief data officers, and we're putting out an announcement in the next couple of weeks for more bureau chief data officers. So we're really excited about that, because I think that's the next wave. 

We need to create both the ecosystem, from an infrastructure standpoint, but also create these leadership roles that can take advantage of the technology opportunities that are placed in front of us So keep an eye out for playbooks that the State Department develops that can that can hopefully elevate the adoption of this kind of technology across the government.