Rima Seiilova-Olson wasn’t sure why she was the only startup founder to be on a panel of academics.
“I feel a little puzzled,” said Seiilova-Olson, co-founder and chief scientist of machine learning at mental health AI startup Kintsugi, speaking to Protocol about her attendance at a meeting of the group of February 16 federal work on how she could use a federally funded AI research cloud.
The National AI Research Resource, or NAIRR, would be a repository of data and tools for AI research combined with access to the computing power needed to develop machine learning and other AI systems. The proposed project is currently being evaluated by a task force overseen by the White House Office of Science and Technology Policy and the National Science Foundation.
Who will be able to use it remains a question. The National Artificial Intelligence Initiative, created by Congress in 2020, envisions NAIRR as a research hub “for AI researchers and students from all fields and scientific disciplines,” including “communities, institutions and traditionally underserved areas”.
This legislation, which established the task force to plan the NAIRR, does not exclude researchers from the private sector, but some believe they should not belong to this community. Researchers from the Stanford Institute for Human-Centered Artificial Intelligence, an influential proponent of the project, said the government-funded research center should focus on the needs of academic and nonprofit researchers.
“Public investment in AI research for non-commercial purposes can help address some of the social harm issues we currently see in commercial contexts, as well as help shift the broader focus of the field towards technology developed in the public interest by the public sector and civil society, including academia,” the institute wrote in an October 2021 white paper.
The Boot Access Case
Among representatives from five colleges and universities, Seiilova-Olson was the only speaker representing the private sector during the virtual roundtable discussing the needs of various potential NAIRR users.
She discussed her goals as a startup founder trying to compete with Big Tech. Kintsugi is developing machine learning models to help detect clinical depression and anxiety based on voice data, and she said this means the company needs access to expensive computing power to treat large-scale unstructured data.
But in addition to the cogs of building AI, Seiilova-Olson said NAIRR should also provide training or community resources for people without access to traditional computer science education.
“An open resource that is freely available to people, I think is extremely important. It has played a huge role in my journey from regular software engineer to machine learning scientist,” she said during the panel, explaining his experience using online training to advance his AI skills.
Rima Seilova-Olson.Courtesy of Rima Seilova-Olson
“There is a great need for small players like me to benefit from these resources, and I am not talking about computing power and data,” she told Protocol last week, adding that she hopes that NAIRR will be able to provide access to STEM-related educational resources for underrepresented people.
After Seiilova-Olson spoke at the NAIRR meeting, Lynne Parker, co-chair of the working group and founding director of the National Office of the AI Initiative at OSTP, told her that she had “a very strong argument for how availability of resources can really make a huge impact, including especially for startups.
Jen King, chief privacy and data policy officer at the Stanford Institute, said use by private sector researchers would pose legal and logistical challenges, while distracting from the cloud’s core mission. of research. “Overloading this resource to cater to very different user groups — small businesses and academic researchers — can undermine its development and, ultimately, its effectiveness,” she told Protocol.
Seiilova-Olson said she was personally invited by Parker to join the roundtable, but didn’t know why. His company has ties to the National Science Foundation. She said she gave Kintsugi about $1 million in NSF grants, some of which the company has yet to receive.
“In NAIRR Task Force discussions to date, a number of Task Force members have suggested that small businesses and startups, particularly those funded by the federal government [small business innovation] grants, have important insights that could help inform the work of the task force,” said Andres Anzola, press officer at OSTP. He said the task force co-chairs worked with federal subject matter experts to identify a grant recipient to participate in the roundtable.
Seiilova-Olson, who co-founded her company after a personal experience of postpartum depression in hopes of helping other parents, said startups are important contributors to AI research, in part because that they are developing an AI intended to be commercialized, and therefore perhaps the needs of people.
However, noting her respect for academic research, she also said that “academic researchers have their own ideas that can be a bit detached from the needs of ordinary citizens.”
King said NAIRR may not be the right place for early stage AI research. “Small business AI may express legitimate needs and constraints in the face of competition from large AI, but the NAIRR may not be the right solution to address them,” King said.
Presence of the private sector in the working group
The NAIRR task force already has a presence in the private sector, which is intentional in accordance with the legislation that established it.
Its 12 members include Andrew Moore, Google Cloud’s vice president and general manager for AI and industry solutions; Daniela Braga, founder and CEO of AI startup Defined.ai, formerly DefinedCrowd; and Oren Etzioni, a venture capital partner at investment firm Madrona Venture Group who also works with new AI startups through an incubator fund affiliated with AI2, a nonprofit that he runs and was founded by Microsoft co-founder Paul Allen.
Google said it wants researchers who don’t need the computing power of the search cloud to be able to access NAIRR data. On the one hand, it would ensure that researchers from commercial cloud providers like Google, Amazon and Microsoft would be able to take advantage of the data flowing through the system.
Healthcare data giant Cerner, recently acquired by Oracle, has also expressed interest in NAIRR. The company, which helps healthcare customers manage patient data and increases its use of AI for hospital administration and patient care, has emphasized public-private partnerships in concerns how the resource’s data is processed.
“It is essential that the governance structure of NAIRR involves[s] representation of public-private entities competent and competent for the prevailing types of data and collection purposes,” the company wrote in an October 2021 response to a task force request for information. “Public policy and laws are core competencies of government. However, the administration of the data asset and associated technical aspects may be more suited to a public-private partner.
Meanwhile, opponents of the project, including some AI policy advisers at the Federal Trade Commission, have raised concerns that the NAIRR will be designed primarily to enable AI projects. scale that, by default, would require the help of large private sector cloud providers. .
Yet there is no shortage of academics who see value in the private cloud providers that create and maintain the NAIRR. “I would strongly recommend that NAIRR be based on existing cloud providers in the commercial space,” said Tom Dietterich, Distinguished Professor Emeritus at Oregon State University’s Collaborative Robotics and Intelligent Systems Institute, during the panel discussion.
A working group of the NAIRR compute-focused task force is also evaluating the possibility of working with commercial cloud providers.
In a separate session from the February task force meeting, research staff at the Science and Technology Policy Institute, a federally funded research and development center, said that partnering with established entities such as commercial cloud providers could provide NAIRR with a range of data sources, a workforce pipeline, and timely updates to data architectures and data security technologies.
However, said Policy Institute research staff member Emily Grumbling, “drawing on these resources could ultimately increase reliance on private sector and for-profit resources for AI.” [research and development] ecosystem “.