Supporting leading universities in a plan to give domestic research communities more compute

America’s largest tech companies and research institutions have thrown their weight behind the legislation aiming to build a national cloud platform for artificial intelligence R&D.

Organizations backing the National AI Research Resource Task Force Act include Amazon, OpenAI, Google, Nvidia, Stanford, Carnegie Mellon, and Upenn, among others.

"With China focused on toppling the United States' leadership in AI, we need to redouble our efforts with a sustained commitment to the best and brightest by developing a national research cloud to ensure our technical researchers get the tools they need to succeed," Republican Senator Rob Portman said.

It’s about China

Bills were introduced by a bipartisan group in both the House and the Senate, seeking the establishment of a ‘National AI Research Resource Task Force.’

The task force, featuring scientists, industry figures, and senior members of the Departments of Defense and Energy, would then help create the national cloud platform – defining its scale, scope, and budget.

Rather than building an entirely new cloud system with dozens of huge data centers across the country, it is thought that the initiative will rely on existing commercial cloud providers, who will be paid to offer discounted or free computing resources to artificial intelligence researchers.

It is therefore not surprising that this week, the major cloud vendors – Amazon, Google, Microsoft, and IBM – all came out in support of the bill, which was originally introduced in early June.

“The widespread support for the National AI Research Resource Task Force Act from our country’s preeminent research universities and leading technology firms demonstrates how critical the legislation is for our country to retain our global lead in AI research,” the group of tech firms and research institutes said in a joint statement.

"A National AI Research Resource will help accelerate US progress in artificial intelligence and advanced technologies by providing academic researchers access to the cloud computing resources necessary for experiments at scale," Jeff Dean, SVP of Google Research, said in a statement.

Despite advances in artificial intelligence sometimes allowing for more to be done with less computing power – as seen in autonomous drone systems – cutting edge AI research has increasingly required more and more compute.

Last July, a study by the Allen Institute for AI came to a startling realization: in just seven years, the computational training cost of the most complex deep learning models had increased by more than 300,000 times – and was doubling every few months.

To compete, those doing cutting-edge research in AI need access to ever more powerful systems, which are increasingly available in the cloud. This is why OpenAI, which provided the data for the Allen paper, this year signed up for one of the world’s fastest cloud supercomputers, built using Microsoft Azure.

Of course, researchers can access the US government’s own collection of supercomputers, which will soon include exascale systems – but a significant proportion is reserved for classified nuclear workloads, or other Department of Energy projects. Researchers have to compete for limited time on limited systems.

This, backers of the cloud platform argue, is not enough to satisfy the needs of the AI community, particularly those without easy access to systems of their own. It is certainly not enough to compete with China, they say, with the geopolitical rival targeting 2030 as the year it hopes to lead the world in AI.

"This legislation takes the first steps towards a national research cloud. By democratizing access to computing power we ensure that any American with computer science talent can pursue their good ideas,” Senator Portman said.

Smaller AI companies also came out in support of the bills, which still have to pass both chambers of Congress.

CalypsoAI, a company that shares few details about its national security work, called the legislation "critical to our nation’s ability to lead the world in building secure and operational AI.” The company claims its AI system – which is used by the intelligence community – is able to benchmark other AI systems, and model risks they pose.

Orbital Insight was another company to support the legislation, a developer of AI-based geospatial tools that analyze unmanned aerial vehicle images and cell phone geolocation data to study human activity.

It is unclear if such companies would gain access to the research cloud, or if they would simply benefit from advances uncovered by the academic community.

How the compute would be allocated, and who would control the innovation produced, are questions that are yet to be answered. First, the task force needs to be formed – and to achieve that, the bill has to progress beyond referral to the Senate Subcommittee on Commerce, Science, and Transportation.