Nvidia is contributing $30 million to a new National AI Research Resource pilot launched by the National Science Foundation and 10 other government agencies on Wednesday. Nearly 25 companies and nonprofits, including Nvidia, are involved in the project. Microsoft Azure is contributing $20 million in compute credits as well.
Nvidia’s $30 million includes $24 million worth of computing on Nvidia’s DGX platform with Nvidia AI tools. AI software platform licenses will go to the national supercomputing centers as well.
In all, the two-year pilot is expected to bring tens of millions of dollars in support from federal agencies and a similar amount from the private sector. Researchers will not be paid for their work directly through the program.
Officials described NAIRR as a first step towards a vision for shared research to strengthen and democratize access to critical resources need to promote responsible AI innovation. Part of NAIRR’s purpose is to ignite AI research investment.
NSF Director Sethuraman Panchanathan praised the number of partners involved, saying they underscore the urgency of the AI research mission. “To continue leading in AI research and development, we must create opportunities across the country to advance AI innovation and strengthen educational opportunities, empowering the nation to shape international standards and igniting economic growth,” Panchanathan said.
The launch of the pilot comes within 90 days after President Biden signed executive order 14110 last October to promote safe, secure and trustworthy development and use of AI. Some parts of NAIRR will be open data sets while other parts will require researchers to apply and be matched for access to government or private resources.
Initially, up to 50 research projects will be supported with the potential of up to 400 projects, said Katie Antypas, director of t he NSF office of advanced cyberinfrastructure. “NAIRR is providing an infrastructure, like a telescope is an infrastructure” for astronomers. “Researchers are going to request access to resources but NAIRR is not funding the researchers’ time” and those payments will come from grants from government agencies or industry partners.
In a background call with reporters, NSF officials said the pilot is just a first step toward sharing AI resources to democratize AI. Officials admitted there are many questions that need to be addressed but said NAIRR will begin to help researchers answer some of the more basic questions.
For example, researchers have only just begun to talk to computers with ChatGPT and large language models. “Researchers really need access to more resources about why [LLMs] behave the way they do,” said Michael Littman, a professor of computer science at Brown University and division director for information and intelligent systems at NSF. “They are tremendously powerful at language, but the language is ungrounded in some ways. The research community is itching to pick apart these models and see what makes them tick. NAIRR is an opportunity to ask deep questions about this is why they behave as they do and how to make them behave better.”
NAIRR will help develop trustworthy AI, which Littman said is a term of art still being hashed out. “Algorithms mimic data trained on errors” but the goal is to develop systems with advances in how they are trained and how the data is vetted so its “accurate and fair with safety and security and is privacy preserving.”
Tess DeBlanc-Knowles, special assistant to the director for AI, said NAIRR will inform the process of how algorithms are tested and verified and “how to navigate that moving forward.”
In addition to Nvidia and Azure, other companies involved in NAIRR include Intel, IBM, Meta (access to the Llama suite), OpenAI, Palantir and more. Vocareum is contributing $2.4 million in the form of access to Vocareum notebooks and cloud resources for 20,000 students for classroom AI education.