Advertisement

NSF’s new AI Institutes will leverage industry resources to expand network

The agency has flagged eight new thematic areas where AI research is needed.
(Getty Images)

The National Science Foundation’s program solicitation for eight more Artificial Intelligence Institutes in 2021 leverages industry funding, datasets, software and expertise for the first time to accelerate research in opportune disciplines.

Accenture, Amazon, Google and Intel committed to help put more than $160 million toward the second round of institutes, making growing the network a multi-sector endeavor.

Industry’s involvement comes as NSF looks to identify new areas where AI research is needed, its original solicitation having spanned more than half of all states in terms of the organizations involved.

“We feel like we’re sort of at a point where the long history of NSF investment is positioning us to be able to make these center-scale investments that allow us to really accelerate advances in AI across foundations and use-inspired or more translational areas of AI,” Erwin Gianchandani, deputy assistant director for Computer and Information Science and Engineering (CISE) at NSF, told FedScoop.

Advertisement

NSF’s latest solicitation seeks institute proposals across eight thematic areas:

  • human-AI interaction and collaboration,
  • optimization,
  • cyberinfrastructure,
  • computer and network systems,
  • dynamic systems,
  • learning,
  • biology, and
  • agriculture and food systems.

Those areas are not set in stone; rather the research community is expected to weigh in via its proposals on the most pressing opportunities

“We look for areas of national importance, places where AI research is ripe for progress at larger scale and longer timeframes,” said James Donlon, program director at NSF. “And we look for an assessment on the part of the research community at these areas — whether it be an area of AI or the intersection of AI with other sectors, disciplines, etc. — that those communities are ready to do research activities at that greater scale.”

Program officers in the funding agencies will hold formal and informal discussions with research communities to refine the themes, beginning with a Sept. 21 webinar where NSF will go over the solicitation and answer questions.

Advertisement

While there are sure to be thematic gaps, the academic-led institutes are a multiyear endeavor, Gianchandani said.

Google plans to provide $5 million to support the National AI Research Institute for Human-AI Interaction and Collaboration, as well as expertise and cloud support.

“The institute we are announcing with NSF will support interdisciplinary research on a variety of modes of interaction between people and AI—like speech, written language, visuals and gestures—and how to make these interactions more effective,” wrote Charina Chou, global policy lead for emerging technologies at Google.

Institutes generally spend the first quarter forming coalitions and creating strategy and implementation plans before becoming operational. Full functionality comes a year in, though institutes may vary, Donlon said.

The original institutes are addressing AI topics like trustworthiness, foundations of machine learning and robotics and intelligent agents.

Advertisement

Prior to NSF establishing the institutes, funding for AI came almost entirely from small-scale, project-oriented research advancing particular methods or applications. Industry techniques like reinforcement learning, which forms the basis of Netflix’s recommendations engine, have their roots in NSF-funded research.

Institutes take NSF’s AI research to the next level.

“Each one of these is a significant activity in its own right to catalyze education, to create a future AI-ready workforce and also to be nexus points for growing new partnerships,” Donlon said.

Latest Podcasts