New Report Details Costs and Structure of a National AI Research Resource
The roadmap includes $2.6 billion in investment over six years and sets the stage for ethical review and privacy protections.
Recently, the White House-led task force behind the National AI Research Resource (NAIRR) issued its final report, asking Congress for a six-year investment to build out AI resources.
The task force was mandated by Congress and organized by the National Science Foundation and White House Office of Science and Technology Policy to develop a roadmap for a shared research infrastructure that would provide AI researchers, scholars, and students with access to computational resources, data, educational tools, and support. The goal: to increase access to the computational power and data that is required to develop AI, currently too often limited to large technology companies.
Key among the task force’s suggestions: One federal agency should serve as the administrative home for NAIRR operations, with a steering committee of additional agencies driving strategy; the platform should support an easily accessible portal to broaden the reach and diversity of users; NAIRR should set standards around responsible AI research and train users on privacy and civil rights; and system safeguards should be built into the infrastructure to ensure security of legally protected data.
The report recommends rolling out the NAIRR in four phases, beginning with setting up the home agency and steering committee this year.
Here, Stanford Institute for Human-Centered AI Policy and Society Director Russell Wald and HAI Privacy and Data Policy Fellow Jennifer King share their insights into the final report, what it’s missing, and the next steps for Congress.
What are you pleased to see in this final report?
Wald: I’m surprisingly pleased there is a clear and direct recognition that academic researchers struggle to access this technology. While that may seem quite obvious, as it was a first order fulfillment of the task force, for me it was validating of the work HAI has done in this space to ensure AI infrastructure can be more universally accessed.
What I’m most pleased to see is the endorsement of an Ethics Advisory Board, which we proposed in our comprehensive report that served as a blueprint for the NAIRR. An Ethics Advisory Board has the means to incentivize standards and norms in AI research by offering the resources that researchers need in turn for adopting best practices. Essentially, you are offering the infrastructure, but the research must be conducted in a responsible manner.
King: In addition to the adoption of the Ethics Advisory Board, I’m also pleased to see both privacy and security concerns centered in this report, though I’m obviously looking forward to seeing more detail about implementation in future iterations. We especially need to understand how individual privacy will be guaranteed for any administrative data contributed to the NAIRR, as well as how any outputs of the NAIRR are reviewed to ensure that researchers do not build models or systems that create privacy risks.
What gives you pause or needs clarification?
Wald: I would have liked if the task force offered a bit more nuance in their report on the types of compute resources that would be available. Similar to our proposal, the task force calls for a hybrid model of relying on both government infrastructure and commercial cloud providers. However, our research noted, when under continuous use, there was a 7x cost savings to using federal computing over the subsidization of commercial cloud credits. This suggests that while you might need to initially over-rely on cloud providers, the eventual goal should be to increase reliance on federal high-performance computing clusters. The distinction is important and can determine how to best optimize the investment.
King: Again, while obviously more detail will be forthcoming as the NAIRR moves forward, one aspect that needs additional focus is that of recruiting and training the talent needed to manage the NAIRR as well as the government agencies that will provide it with data. The federal government needs to plan now for how to recruit and train for these highly technical, specialized roles, especially at agencies for which these types of skills are outside their normal focus.
The task force has recommended a $2.6 billion investment to get this up and running. How does that divvy up, and is that a reasonable investment?
Wald: Quantifying this type of investment is difficult without being able to access federal scoring methods and run different scenarios such as the adoption rate of the NAIRR.
But what that number tells me is the seriousness of this moment and how the U.S. has an opportunity to be the global steward for AI and ensure its development in a responsible manner that upholds human values. If we consider how one part of the CHIPS Act was to deny an authoritarian regime access to this technology, a logical additional step is to ensure broad, noncommercial access to that same technology to innovate, expand our talent pool, and compete globally.
One question that’s been floated is who gets to use this – just academics? Startups? Does this report answer that question?
Wald: While the NAIRR is intended for academic researchers and students, there is a carveout for startups and small businesses that have received federal support. The HAI white paper recognized this was a possibility and suggested a pilot program that was exclusive to academic use before expanding access to commercial research, in order to engender more scholarly research. However, the NAIRR Task Force was careful to note in their report that federally funded small businesses should be “allowed to access NAIRR resources, but only at limited levels and in support of research that is in the public interest.”
How likely is Congress to approve this report and allocate the funding for this resource?
Wald: While the politics of the moment are challenging, there are plenty of lawmakers on both sides of the aisle that are deeply passionate about the idea for a NAIRR. In fact, the original legislation that established the NAIRR Task Force had strong bipartisan support in both chambers of Congress, and that was during the 2020 presidential election year.
What is important to note is the NAIRR has a lot for numerous stakeholders. It will advance scientific research, help set standards in AI, add more diversity to the table, train the next generation, and help America globally compete in AI. These are universal concepts that most lawmakers can get behind.
Stanford HAI’s mission is to advance AI research, education, policy and practice to improve the human condition. Learn more.
Related News
AI Companions May Worsen Loneliness for Vulnerable Users, Stanford Study Finds
Stanford research finds that users with limited social networks who seek emotional support from AI companions experience lower well-being.
Stanford research finds that users with limited social networks who seek emotional support from AI companions experience lower well-being.
Open-Weight Models Aren’t Enough. We Need Truly Open Source AI Models for Science and Society.
As Chinese AI closes the capability gap, Washington and Silicon Valley debate open-weight models. Stanford HAI's James Landay says it's the right conversation framed the wrong way.
As Chinese AI closes the capability gap, Washington and Silicon Valley debate open-weight models. Stanford HAI's James Landay says it's the right conversation framed the wrong way.
As artificial intelligence moves beyond language into the physical world through "world models," Stanford researchers warn that policymakers face an even steeper governance challenge than with large language models—and the window to get ahead of the technology is closing fast.
As artificial intelligence moves beyond language into the physical world through "world models," Stanford researchers warn that policymakers face an even steeper governance challenge than with large language models—and the window to get ahead of the technology is closing fast.