How to Build a National AI Research Resource
For the thousands of academics around the nation longing to tackle some of society’s most pressing problems, research doesn’t come cheap.
Nowhere is that more true than in the field of advanced artificial intelligence, one of the most consequential technologies of the 21st century. Massive — and expensive — amounts of compute power and data are needed to run the huge machine learning models that power this critical research, limiting their use to those with the deepest pockets: the Googles, the Microsofts, the Stanfords.
In early 2020, in an attempt to level the playing field and increase opportunity nationwide, the Stanford Institute for Human-Centered Artificial Intelligence (HAI) asked the federal government to create a task force to study the creation of a national research cloud (NRC). The proposal called for a close partnership between academia, government, and industry to provide researchers across the nation with affordable access to otherwise unattainable resources. The initiative was supported by 22 of the top 30 computer science universities in the country and was successful: Congress passed the National AI Research Resource Task Force Act of 2020 in December.
Envisioning a Cloud — Rigorously
To assist this future task force, HAI has brought together a diverse group of student researchers in law, engineering, computer science, economics, and business to figure out how best to design and operate a national research cloud. Offered as part of Stanford Law School’s growing “policy practicum” program, the course “Creating a National Research Cloud” tasks those students with researching similar existing models, examining the myriad of issues involved, and producing a report to inform federal task force members and the public.
“After we saw our success with this initiative being accepted by Congress, we decided to not just call it a day, but instead to put intellectual rigor and design behind the cloud itself,” says Russell Wald, director of policy for HAI and co-instructor. “This report will be available to hand over to the task force and say ‘We just made your job significantly easier, because we did the investigation with the technologists, the computer scientists, the lawyers, and the users of this technology to be able to effectively make this happen.’ ”
Students will draw heavily on the resident expertise at Stanford and also interview a range of national and global experts to investigate questions ranging from cybersecurity to interagency cooperation, says Daniel E. Ho, the William Benjamin Scott and Luna M. Scott Professor of Law and a Stanford HAI associate director and co-instructor.
“It’s impossible to think about something as ambitious as the design of the national research cloud without having a wide range of disciplinary perspectives,” Ho says. “That includes hardware engineers who will consider the computing infrastructure, computer scientists who will draw on their AI training to conceive of the best innovation environment, lawyers who can navigate the privacy, security, and IP thicket of liberating data to be offered as part of the cloud, and policy analysts and business school students who will consider the economic and business model of such an initiative.”
Practicum students will work on small investigative teams studying, among other things, the market forces that led to the need for an NRC; the privacy and data concerns that must be addressed in such a project; and already existing models, such as Compute Canada, which provides state-of-the-art computing resources to academic and industrial researchers across that country.
“This kind of rigorous research is our bread and butter here at Stanford; it’s what we do in our own research, and what we’re bringing to bear on these very important questions,” says practicum student Shushman Choudhury, an international student from India who expects to receive his PhD in computer science this year. “It’s easy to say, ‘Of course everyone should have access to all this, why not?’ But we need to make a well-crafted case for why a national research cloud is worth the cost, what broader implications it has, and what actual research we can draw upon to support it.”
Opening the AI door to all
Democratizing access to compute power and government databases would allow students and researchers anywhere — not just at elite institutions — the ability to train and innovate in AI and eliminate the need for them to abandon academia for better-resourced corporate settings. It would also open the door for innovation not focused on commercial gain, says Jen King, HAI’s Privacy and Data Policy Fellow and co-instructor.
“This approach will support publicly funded, publicly focused research projects, and that’s important,” she says. “So much of what we’re seeing in the AI space now is led by private industry, and there’s a real risk when all of that competitiveness and expertise is tied up only on the private side of things. It’s important to also have this expertise in the public sector.”
The practicum offers Stanford students a chance to influence a project that could unleash tremendous potential in AI, Wald adds.
“They have the means to make an impact on the future of U.S. AI infrastructure,” he says. “Very few people knew how to code or create a web page 30 years ago, and now that’s all ubiquitous. A national research cloud opens the door for researchers at every level to become proficient — or expert — in AI in their hometowns without having to pick up and go to an MIT or a Stanford, or go work at Google or Facebook. That’s beneficial across the board for the United States.”
Choudhury says he and his fellow practicum students recognize their research could affect not only the course of AI equity in the U.S., but around the globe.
“Doing this is an honor and a privilege and not one that any of us take lightly because ultimately we’re dealing with issues that could impact the government’s investment in this,” he says. “As Stanford students, we’re among the privileged ones who have our own funding and a large endowment and research grants, but we’ll hopefully have an impact in helping our colleagues at other institutes around the country gain that access and be able to better pursue new technologies in applications across the board.
Stanford HAI's mission is to advance AI research, education, policy and practice to improve the human condition. Learn more.