Software

When It Comes to Integrating AI Into Agencies’ Work, the NAIRR Is a Start

The National Artificial Intelligence Research Resource democratizes researchers’ access to computing resources, but it’s on agencies to take the next steps.

Dave Nyczepir

Dave Nyczepir is a Senior Editor for Manifest.

The National Artificial Intelligence Research Resource pilot is an important but limited step in getting agencies to start planning to integrate AI technology into their work, according to an NVIDIA spokesperson.

Led by the U.S. National Science Foundation in partnership with 10 other agencies and 25 companies, the two-year pilot devotes a shared research infrastructure for responsible AI discovery and innovation.

The National Artificial Intelligence Initiative Act of 2020 instructed NSF and the White House Office of Science and Technology Policy to form a task force that published a roadmap for the NAIRR in January 2023. President Biden’s AI executive order, issued in October, called for that roadmap to be implemented as an initial pilot. Launched this January, the pilot is primarily concerned with democratizing the NSF academic research community’s access to AI computing resources, not that of other agencies.

“They each need to consider how AI is going to impact their future work, and collaboration with an academic community can be part of that,” the NVIDIA spokesperson says. “But this is really trying to make sure that academic researchers have access to a research resource, so it's hard to plan for the specific outcomes that will emerge because it will depend on the proposals that are written and accepted in terms of specific, strategic outcomes at any of these agencies.”

Click on the banner to learn how platform engineering works.

NIVIDIA Brings AI Resources to the NAIRR

Other agencies involved in the pilot include the Department of Energy and the Department of Defense, as well as the National Institutes of Health, the National Institute of Standards and Technology, and the U.S. Patent and Trademark Office. They stand to benefit from pathfinding projects undertaken using the NAIRR, but the extent will depend on what’s proposed by researchers.

The initial call for allocations to high-performance computing resources ended March 1, but a second is expected as soon as this month. That will include access to in-kind contributions — computational, data, software, model, training and user support resources — from industry partners such as NVIDIA.

NVIDIA is offering about $30 million worth of resources to the NAIRR, the spokesperson says.

The first offering is access to the NVIDIA DGX Cloud product, effectively 1,000 node months of computing time on clusters stood up for the NAIRR, worth $24 million. The second offering is 1,000 seats of software licenses for NVIDIA AI Enterprise, a bundle of AI frameworks for training, inferencing and data analytics that sells for $4.5 million, the spokesperson says.

Lastly, NVIDIA is offering an AI and data analytics training curriculum called the NVIDIA Deep Learning Institute, as well as a series of hackathons, boot camps and workshops around AI and high-performance computing worth about $1.5 million, the spokesperson says.

WATCH: The U.S. Army Cyber School trains soldiers remotely and securely.

Understanding How AI Can Help Agencies

NSF researchers could begin accessing NVIDIA’s resources as early as April to study AI and its applications in areas of science and engineering including physics, chemistry, biology, environmental science and climate science, the spokesperson says.

For NVIDIA’s part, it will learn about the effectiveness of its AI products in applications they weren’t necessarily designed for.

“We expect to get stretched by these use cases, some of the brightest minds working on some of the hardest problems,” the spokesperson says. “From that, we’ll gain experience and be able to improve our products in the future.”

NVIDIA further expects to build confidence in the NAIRR model itself while gathering more information on current trends in academic research around AI and how the technology can further agencies’ missions.

“I think that there’s a lot of interest in the data sets that federal agencies will contribute to the research community, since broader access to these data sets would be a rare, relatively new thing,” the spokesperson says.

Boyuan Chen/Columbia Engineering