AI Research Infrastructure Task Force Needs Input on ‘Democratizing’ Resources

The National Science Foundation and the White House’s Office of Science and Technology Policy are developing a plan to make high-performance computing, machine-learning datasets and other resources more widely available to artificial intelligence researchers at every level.

Developing artificial intelligence tools and technologies requires lots of data and even more computing resources. Gaining a national advantage in this area will require a significant concentration of work that is currently limited to agencies and organizations that have those resources.

But the best, groundbreaking ideas aren’t always centered in places with the most resources.

To address this issue, the 2021 National Defense Authorization Act charged NSF and OSTP with developing a plan to build up those resources and make sure they are available to people throughout the U.S. who can make good use of them.

The NDAA included creation of the National AI Research Resource Task Force, which “has been directed by Congress to develop an implementation roadmap for a shared research infrastructure that would provide artificial intelligence researchers and students across scientific disciplines with access to computational resources, high-quality data, educational tools, and user support,” according to a notice set to publish Wednesday in the Federal Register.

As the team develops the plan for making this a reality, officials put out a request for information and opened a comment period to hear about the “options, models and priorities” that should be considered, “as well as how the NAIRR can reinforce principles and practices of ethical and responsible research and development of AI,” the notice states.

All of this will go toward building “essentially a shared computing and data infrastructure,” Erwin Gianchandani, NSF senior adviser to the director for translation, innovation and partnerships, told Nextgov. “Trying to stitch together computational assets, as well as data assets and thinking through aspects of security and privacy preservation.”

The resources will also likely include education tools to help new users get up to speed on the tools and offerings, said Gianchandani, who also co-chairs the NAIRR Task Force with his OSTP counterpart Lynne Parker, director of the National Artificial Intelligence Initiative.

The task force is charged with sorting the specifics for all of this—or at least a plan to do so. But the upshot will be a technological infrastructure “that enables AI research, as well as education activities for artificial intelligence,” he said.

The goal will be to “democratize access” to these resources to improve equity and availability of things like the high-performance computing needed for AI research.

“I may be at a well-resourced institution or I might be at a less-resourced institution. But I, nonetheless, can have access to these assets that are critical for me to be able to conduct research and make seminal breakthroughs, make seminal advances in the field of AI,” Gianchandani said. “And in the many fields and sectors of our economy that AI touches.”

The plan also needs to ensure the security of the data and systems being accessed, the privacy of those involved—including the people on which the data is based. At the same time, the task force wants to focus on civil liberties and civil rights issues around AI, including systemic bias.

The comment period for public input was originally set to close Sept. 1 but NSF and OSTP extended the deadline to Oct. 1 “in response to requests by prospective commenters that they would benefit from additional time to adequately consider and respond.”

Multiple stakeholders reached out to the task force to express their interest in responding to the RFI but needed more time to get through the large set of complex but important questions being asked. That, and many were on vacation this summer and said they would appreciate the extension, Gianchandani said.

He said he hopes the extension shows that the task force is sincere in its request for feedback and sees the comment period as more than a compliance exercise.

“We are really, earnestly interested in input from the community,” he said.

The task force was established in June, from which time Congress gave them one year to deliver a draft implementation plan. Gianchandani said the team is on track to deliver that by spring 2022, with the goal of delivering the final plan to Congress and the White House by next fall.