As federal agencies seek to further adopt artificial intelligence (AI) technologies, government and industry experts stress the importance of ensuring that AI is embedded in the mission and that agencies s are moving towards a more distributed AI approach.

Pamela Isom, Director of the AI ​​and Technology Office at the Department of Energy; Curt Smith, NVIDIA Vice President, GPU Datacenter Architect; Jay Boisseau, Artificial Intelligence and High Performance Computing Technology Strategist at Dell; and Judson Graves, Director of Analytics and AI at ViON, each shared where they’ve seen progress in federal AI adoption during the MeriTalk webinar “The Edge of AI: Federal AI Adoption” January 12.

“The federal government as a whole, from my perspective, is making pretty good progress when it comes to understanding how to apply AI to the mission,” Isom said during the roundtable. “And that’s the most important thing because if we respond to the mission, we respond to the needs of the citizens.”

It’s an approach shared by Smith, who said it’s crucial for organizations to have a customer experience and a mission-driven view of how AI can help.

“One of the very first parts of the work an organization needs to undertake in its AI journey is to start with a vision and how that vision can transform the customer experience and how it relates to the mission of the organization. “Smith said.

Isom said he’s noticed a shift from federal adoption of defense-focused AI to a more distributed approach within the federal government. At DOE, Isom said the agency is looking at how AI can be used to help the agency in its fight to mitigate the impacts of climate change.

Boisseau highlighted the importance of making data accessible in building AI models.

“First and foremost, make sure you always have access to all the right data that can help train” AI models, Boisseau said. “This is one of the challenges of the current era.”

“We created all these data silos – and we tend to think of silos as a negative word now – but there were many good reasons to separate the databases of different party organizations, to reduce risk and to allow you to create the right schemas and tables and such that are optimized only for the people who were using it,” he said.

“There were therefore many good reasons to specialize data access and secure subsets of data,” Boisseau added. “But as we adopt these more powerful data analysis techniques – even including AI techniques – we often want more access to more data across departmental boundaries, and even across borders. agencies, while at the same time recognizing that this data has increased value to external threats as well, so we need to balance that and make it more accessible to the people who need it, while continuing to strengthen the secure protection of this data against external threats.

Graves warned agencies working to structure and design pilot AI programs not to skip any key steps.

“What I stumbled over a lot was transitioning from that demo to proof of concept to pilot transition, and honestly we’ve seen a lot of people shorten that,” Graves said. “We see a lot of them going into a demo situation and being blown away with some kind of out-of-the-box capability, maybe even integrating a bit with some of their data. And then kind of jumping the pilot phase, then go straight to trying to make it in production, and then fail miserably.

“The truth is that AI over the past few years ‘represents’ an intersection of massive new data storage capabilities, incredibly fast transfer speeds over networks, and the vast amount of computing power that we are suddenly able to release on these datasets,” he added. “That three-way combination is really what enabled this kind of new capability in AI.”

Isom said his office is currently exploring how to use AI for projects related to decarbonization, grid and energy resilience, and hydrogen and solar efficiency. She said the DOE is also working on the use of AI for autonomous vehicles and drones and sees a distributed AI approach as the future of integration at the federal level.

“My office is very focused on not only helping us ensure that we stay focused on the mission, which includes research and development, but also on protection – protecting AI when we develop or acquire solutions,” said she declared. “[DOE plans for] more work and more ideas and more education and training on distributed AI. And that will complete the whole concept of AI and some of that capability. So we need to put that more into action and into operation.

To hear all of Isom and DOE’s plans, and more from the rest of the panel, register here for MeriTalk’s “The Edge of AI: Federal AI Adoption” webinar and watch it all on demand.