Google Public Sector Summit featured a packed lineup of AI leaders, panels on use cases and real-world government challenges.
The gist of the conference is that government generative AI customers can leverage commercial Google Cloud but be walled off. Google Public Sector is an independent entity that leverage Google Cloud technology, but takes it the last mile (with isolated instances in some cases). In an interview with analysts, Google Public Sector CEO Karen Dahut said the company's goal is to serve the public sector commercial cloud capabilities for government use.
"When we came into this market, what we found was traditional gov clouds. They're walled off and lack parity. It lacks the compute scale and doesn't have resiliency. What if we made our commercial cloud available to government by a software defined community cloud with all of the guardrails built in? OMB came to that same conclusion independent from us."
- Google Public Sector lands new clearances for Gemini, authorizations for Air Force Cloud One
- Google Public Sector: A look at the strategy
Here's a look at all the takeaways from the conference, lessons and best practices that emerged:
If you invested in data infrastructure, architecture and governance you're able to drive value from generative AI projects. Lakshmi Raman, Director of AI at the Central Intelligence Agency (CIA), said the agency was able to drive value quickly "due to investments made in AI, data and tooling over the last decade." "That investment enabled us to evaluate generative AI capabilities quickly," said Raman.
Improving data quality may be your most important genAI use case. Dr. Ted Kaouk, Chief Data & AI Officer and Director, Division of Data CFTC, said his agency is focused on the quality of data ingestion to focus on anomaly detection and "developing prototypes to detect bad actors."
Look at your data as a product. Zach Whitman, Chief Data Scientist & Chief AI Officer at GSA, said he's been focused on using generative AI to focus on "how to enable better data productization and groundwork that maximizes value in the future."
Ron Robinette, Deputy Secretary, Innovation and Technology & AIO at CA GovOps, seconded Whitman's take. "We have five proof of concepts in state of California and we need our data better prepared to take advantage of that opportunity," he said.
Mark Munsell, Director, Data and Digital Innovation, Founder of Moonshot Labs at National Geospatial-Intelligence Agency, said Munsell is improving its data by making sure everything possible can be entered into a database so it can have structure that can later help improve model training. This structure can then be combined with computer vision.
"We have 100s of petabytes of data from sensors and traditionally humans would look at the data and find signals, but now we need computer vision and model to cover places we can't," said Munsell.
Invest in metadata. Whitman said part of that data productization effort is to invest in metadata. "Overinvest in metadata so you can make the data explainable to the AI systems," he said. "Sometimes that's hard work and it's hard to get investment, but it's worth it."
"Metadata is critical," said Gulam Shakir CTO at National Archives & Records Administration (NARA). "We are leveraging several pilots."
Generative AI is breaking down silos. Whitman noted that conversations about generative AI use cases are going well beyond technology. Use case conversations are involving risk and safety, technology and the business. "We are seeing this cross pollination of great ideas," said Whitman. "It's a game changer that breaks down silos."
AI at the edge and hybrid use cases. At the Google Public Sector Summit, the company spent a lot of time talking to agency leaders about being the "best on-premises cloud" for workloads that are air-gapped, separated from networks and can still run models.
There's a reason for AI systems designed for the field: The public sector--especially the military--often has spotty connectivity. During a panel, Jane Overslaugh Rathbun,
CIO of the US Navy, said sailors are "disconnected continuously." She added that the Navy is looking for edge AI capabilities that can process the sensor data from ships in contested theaters and get sailors the data to make decisions.
Young J. Bang, Principal Deputy Assistant, Secretary of Army Acquisition, Logistics & Tech, noted that the Army is rarely connected at the edge. Bang said a hybrid approach to genAI will emerge where models are trained centrally, fine tuned and sent to the edge.
Smaller models are seen as key. Mark James, Director of Infrastructure and Support Services at the Department of Homeland Security, said AI at the edge is going to require smaller models. ""We're exploring smaller language models to support AI at the edge," said James. For the DHS, ports are a key edge location where smaller models can have impact augmenting officers' day-to-day activities by scanning documents.
Talent. Brig. Gen. Heather W. Blackwell at the US Air Force | JFHQ-DODIN said generative AI is critical to making sure your limited talent resources are used on high-value projects. "We need AI to find those things that my analysts can't see so we can use our limited analytics assets on things only humans can do," said Blackwell.
Maj Gen Anthony Genatempo, Program Executive Officer, Cyber and Networks Air Force Life Cycle Management Center, C3I&N, said you need the talent to also ensure use cases for generative AI work out. "I want to tackle one aspect of our business we do to see if AI can help us out. Right now, I want to cut my contracting timeline from 18 months to 14 days," said Genatempo. "There are aspects of the workforce who thinks AI is about getting rid of them. I'm not getting rid of one person. People that know how to use these tools will replace people who don't."
Generative AI is a cultural opportunity. Raman noted that "culture eats strategy for breakfast" so AI leaders need to make sure "the AI journey is aligned with organizational beliefs."
Culture was a theme echoed by General Chance Saltzman, Chief of Space Operations for US Space Force. He said government needs a different type of leader who knows how to innovate within government. Critical thinking will be critical.
Urs Hölzle, Google Fellow, said cultures need to evolve with an eye on longer term projects and a tolerance for failure. Takeaways from Hölzle on culture include:
- Cultural change is key to enabling transformative innovation within organizations.
- Embracing failure as part of the innovation process is crucial. Different projects should be categorized (e.g., core, experimental) to manage risk appropriately.
- It's tempting to rely on legacy methods in moments of pressure, but true progress requires focusing on new solutions and resisting this tendency.
- Structured prioritization helps ensure that resources are allocated effectively, avoiding the pitfall of focusing only on short-term wins.
- Effective leaders foster a culture that embraces learning from failures while being clear about project expectations.
