Readout of the Fourth National Artificial Intelligence Research Resource (NAIRR) Task Force Meeting

The NAIRR Task Force focused on data resources, access to testbeds, the user experience including educational offerings, as well as the imperative to design equity, privacy, and responsible research conduct into the policies and processes of the NAIRR

The National Artificial Intelligence (AI) Research Resource (NAIRR) Task Force convened its fourth virtual, public meeting on December 13 to further develop a vision and implementation plan for a national cyberinfrastructure that would expand participation in AI innovation to researchers and communities currently lacking resources to pursue cutting-edge research. Through provision of computational, data, and training resources and a strong governance model, a NAIRR could fuel breakthroughs in areas ranging from climate change to healthcare, and facilitate groundbreaking research in foundational areas of AI such as auditing, testing and evaluation, bias mitigation, and security. The Task Force is working toward consensus recommendations on a NAIRR implementation plan and roadmap that will be provided to Congress in 2022.

Co-Chairs Dr. Lynne Parker, Director of the National AI Initiative Office within the White House Office of Science and Technology Policy (OSTP), and Dr. Manish Parashar, Office Director for the Office of Advanced Cyberinfrastructure at the National Science Foundation (NSF), opened the meeting by kicking off a discussion on the Task Force’s shared vision for the NAIRR. The discussion focused on issues central to the use and governance of a NAIRR, with particular emphasis on the appropriate user base and resource allocation model. The Task Force discussed aligning the user base with the strategic objectives of the NAIRR, which include strengthening AI innovation and lowering barriers to entry to AI research, while ensuring the envisioned uses of the NAIRR represent appropriate investments of public funds. 

Building on ideas from external experts who spoke during the Task Force’s third public meeting, the Task Force discussed proposed recommendations on data resources, testbeds and other testing resources, and user tools and resources that could constitute key components of a NAIRR. The Task Force reached consensus that the NAIRR should institutionalize trust, curation, validation, and discoverability when facilitating user access to diverse and distributed data resources. Task Force members emphasized that the NAIRR’s user interface and training resources should be designed to support the full range of users and experience levels, from those seeking to engage in AI research for the first time to deep experts. In addition, the Task Force concluded that AI testbeds should be catalogued, made available, and nurtured through the NAIRR at a level that does not detract from investments in computational and data resources.

A panel of outside experts provided perspectives and insights on how to design the NAIRR to account for privacy, civil rights, and civil liberties, ensuring the NAIRR promotes fair, trustworthy, and responsible AI research and development. The speakers highlighted best practices in building inclusive datasets and governance approaches that treat equity as a core design principle. Technical design considerations related to privacy and fairness were raised, as was the potential for the NAIRR to facilitate impact assessments and promote accountability mechanisms for emerging AI systems.

The Task Force also considered security and access control, discussing the vulnerabilities and requirements unique to the NAIRR and its envisioned federated architecture. Particular emphasis was placed on the need to balance security with usability. Task Force members identified existing technical gaps in today’s cyberinfrastructure that will be critical to the success of the NAIRR, especially the technical integration of the anticipated heterogeneous computational and data resources that will comprise the NAIRR.

The meeting closed with the Task Force members answering questions posed by the public, describing how the NAIRR might interact with existing Federal enterprise architectures, provide computational capabilities such as cloud templates, and support existing businesses certified by Federal agencies.  

The Task Force will hold its fifth public meeting in February 2022. Details on how to participate in the Task Force meetings are available on AI.gov/nairrtf, along with materials from prior meetings.  

External Speakers at the December 13 Meeting:

Solon Barocas, Microsoft Research and Cornell University

Lujo Bauer, Carnegie Mellon University

danah boyd, Microsoft Research and Data & Society

Nicol Turner Lee, Brookings Institution

Hannah Quay-de la Vallee, Center for Democracy and Technology

Deborah Raji, Mozilla Foundation