Last week the National AI Research Resource (NAIRR) Task Force released its final report and roadmap for building a national AI infrastructure to include computational resources, high-quality data, educational tools, and user support. The report was directed by the National AI Initiative Act of 2020 (passed in 2021). The roadmap calls for spending $2.6 billion over six years to create NAIRR.

It’s an ambitious plan that requires further congressional action to provide funding and guidance. Here’s an excerpt from report describing key deliverables:

“A widely accessible AI research cyberinfrastructure that brings together computational resources, data, testbeds, algorithms, software, services, networks, and expertise, as described in this report, would help to democratize the AI research and development (R&D) landscape in the United States for the benefit of all. It would help create pathways to broaden the range of researchers involved in AI, and to grow and diversify approaches to, and applications of, AI. This cyberinfrastructure can also help to open up new opportunities for progress across all scientific fields and disciplines, including in critical areas such as AI auditing, testing and evaluation, trustworthy AI, bias mitigation, and AI safety.”

The report task force was headed by NSF director Sethuraman Panchanathan and Arati Prabhakar, assistant to the President, in OSTP, and includes familiar HPC leaders such as Fred Streitz at Lawrence Livermore National Laboratory and Dan Stanzione at the Texas Advanced Computing Center. It spells out four key goal – spur innovation, increase diversity of talent, improve capacity, and advance trustworthy AI.

Other key implementation points include:

• The NAIRR administration and governance should follow a cooperative stewardship model, whereby a single Federal agency serves as the administrative home for NAIRR operations and a Steering Committee comprising principals from Federal agencies with equities in AI research drives the strategic direction of the NAIRR.
• The NAIRR should provide access to a federated mix of computational and data resources, testbeds, software and testing tools, and user support services via an integrated portal.
• The NAIRR must be broadly accessible to a range of users and provide a platform that can be used for educational and community-building activities in order to lower the barriers to participation in the AI research ecosystem and increase the diversity of AI researchers.
• The NAIRR should set the standard for responsible AI research through the design and implementation of its governance
• The NAIRR should implement system safeguards in accordance with established guidelines.

Worry over harnessing the rapid advance of AI technology along with the desire to remain a global leader were among the motivating elements behind the National AI Initiative Act.

The report notes, “Breakthroughs in AI capabilities have been achieved through the creation of large, computationally-intensive deep learning models. In the pursuit of more generalizable capabilities, such models have been growing in size: OpenAI’s GPT-3 in 2020 broke barriers at 175 billion parameters. Google followed suit in 2021 with a 1.6 trillion- parameter model, and the Beijing Academy of Artificial Intelligence with a 1.75 trillion-parameter model soon after. Published cost estimates ballpark that training a 110 million-parameter language model costs about $50,000, a 340 million-parameter model costs about $200,000, and a 1.5 billion-parameter model costs about $1.6 million.”

No one wants to be left behind for a wide range of economic, security, science, and geopolitical reasons.

As always, the devil is in the details. The report recommends a roughly four-year phased implementation plan with initial “operational capability” ramping up in year three. More from the report on developing the budget:

“Specifically, using one agency’s current advanced computing investments during the period FY 2016–FY 2021 as a proxy and considering known oversubscription of about 125 percent, the Task Force identified that an investment of over $1 billion would have been necessary during this period. These investments would have provided advanced computing resources to a community of about 19,000 users spanning about 2,300 active projects totaling about $6 billion in Federal R&D investment. Put another way, the average advanced computing investment needed per 1,000 users is about $53 million, and the average advanced computing investment needed per $1 billion of Federal R&D funding is about $169 million.”

“In arriving at the final budget estimate, the Task Force took the above investments and estimates into account, along with an assumption that the scale of investment and size of the AI community will continue to grow rapidly in the years ahead. The bulk of the estimated budget of $2.6 billion (i.e., $2.25 billion) funds the NAIRR resource providers. Resource providers should be brought online every two years with a six-year lifetime, requiring a new $750 million investment to be made every two years to ensure that NAIRR resources remain at the state of the art.”

“The Operating Entity budget is estimated at approximately $60 million per year to support the coordination and management of NAIRR activities (see Table 2). An additional $5 million per year is needed to support the Operating Entity’s external evaluation process. The budget for the Operating Entity is based upon historical experience that the annual cost of operations for complex cyberinfrastructure is approximately 20 percent of the cost of the cyberinfrastructure resources themselves. Funding for the Operating Entity and external evaluation should be appropriated by Congress to the administrative home of the NAIRR, with suitable language to permit funds to be used to initiate and staff the Program Management Office. Funding for the NAIRR resource providers should be appropriated by Congress to the agencies that will fund them.”

Bringing the NAIRR into existence will require further action and this report not only details next steps, but also offers draft legislation. Here’s another excerpt:

“For the President and Executive Branch Departments and Agencies

“The development and sustainment of the NAIRR will require active involvement by many Federal agencies, which will need to participate in the Steering Committee and the Program Management Office, allocate funds for the resource providers, and oversee the NAIRR’s execution. The agency serving as the administrative home will need to establish a Program Management Office and allocate funds for the Operating Entity.

“For the NAIRR to be successful, it will need to reach all major AI-using research communities—and for that to occur, all of the Federal research agencies that invest in AI R&D will need to participate in the management and funding of the NAIRR.

“For Congress

“Congressional legislation has continually reaffirmed the Federal Government’s commitment to funding cutting-edge information technology R&D. The success of the NAIRR initiative will depend on similar commitments from the Federal Government using similar legislative tools and authorities. The long-term continuation of U.S. strategic advancement and leadership in AI depends on guidance and commitment from Congress. (See Appendix I for proposed NAIRR authorizing legislation drafted by the Task Force.)”

It will be interesting to see how all of this unfolds given current economic worries, a divided congress, and the forthcoming electioneering cycle for U.S. presidential election in 2024. Stay tuned.

Link to report: https://www.ai.gov/wp-content/uploads/2023/01/NAIRR-TF-Final-Report-2023.pdf