US govt, tech sector team up to get academia making its own next-gen AI models
Launch National Artificial Intelligence Research Resource to help non-corporate boffins play catch-up
The US National Science Foundation has hooked up with tech companies to help academics secure computing power, data, and more to build their own AI models.
On Wednesday, the agency launched the National Artificial Intelligence Research Resource (NAIRR), a pilot project that pulls together resources from private and public sectors to drive research and development. NSF director Sethuraman Panchanathan said 10 federal agencies and more than 20 companies and organizations have pledged to support the initiative.
Giants including Google, Nvidia, OpenAI, and Amazon promised to share some of their tools, hardware, and datasets to researchers who are accepted for the program. Government agencies like the Department of Energy and DARPA also agreed to share their resources, giving academics access to supercomputers, like the Delta, Frontera, and Summit clusters.
Nvidia provided some specifics about its involvement, with the GPU maker committing $30 million in compute infrastructure to support the pilot, including $24 million of Nvidia DGX Cloud compute integrated with the requisite software tools and supported by technical experts to assist NAIRR pilot users.
Researchers can submit their research proposals to the NSF, with those selected gaining access to the pool of resources. Officials are looking to support transformative and novel AI projects that further the country's national interests.
Katie Antypas, director of the NSF's Office of Advanced Cyberinfrastructure, said NAIRR would help academics do things like obtaining datasets from the National Oceanic and Atmospheric Administration and NASA to study extreme weather events using AI.
"One example may be in an AI researcher that wants to investigate the validation and verification of large models. That researcher will be able to gain access to large scale computing resources that they may not otherwise have had access to," she added.
- Facial recognition tech has outpaced US law – and don't expect the Feds to catch up
- Pentagon using ChatGPT? Oh sure, for cyber-things and veterans, says OpenAI
- NIST: If someone's trying to sell you some secure AI, it's snake oil
- UK and US lead international efforts to raise AI security standards
Industry is at the forefront of AI, because of the high costs involved, academia often lags behind. NAIRR is designed to help level the playing field a little.
The pilot project is broadly split up across four different areas that focus on open AI research, security and privacy, software tools, and training and outreach. The NSF is currently interested in backing developing techniques to make models more trustworthy and responsible, and will later expand its interests to using AI to tackle climate change, improve healthcare, and education.
"[NAIRR] will transform how AI research is done in the United States, and who can participate, unleashing AI discovery to the benefit of us all," Panchanathan said in a press briefing. "I want to underscore the all part, the future of AI security, AI safety, AI job creation, all rely on ideas from every corner of our society."
The finer details of what models, datasets, and hardware will be made available by whom exactly are still being ironed out. Antypas told The Register that public agencies will contribute computing resources worth "tens of millions of dollars," but said the agency was still figuring out how much its industry partners are willing to donate.
NAIRR was created as part of US President Joe Biden's executive order signed last year in October. Biden urged the NSF to provide tools enabling researchers and students to access "key AI resources and data" in 90 days. ®