“The NAIRR should support research irrespective of potential commercial implications; support research into the safety, security, privacy, and equity of AI systems; and provide resources to underserved communities of developers,” CSET’s Dakota Cary writes. He submitted this comment in response to a public request by the Office of Science and Technology Policy and the National Science Foundation to support the work of the NAIRR Task Force. The NAIRR will make the most impact by focusing on data, computational capacity, and talent to improve the quality and access to AI development for actors that the market does not adequately support.
To implement these changes, Cary recommends the following:
- The NAIRR should facilitate access to compute through NSF’s new CloudBank program.
- The NAIRR should provide computational resources for undergraduate and graduate students in the U.S. through cloud access beyond what is freely available through Google’s Colab or similar industry resources.
- The NAIRR should provide computational capacity comparable to the compute used to train the private sector’s largest models for specific research by PhD researchers.
- The Task Force should engage with the Department of Commerce and state commerce departments to determine an appropriate mechanism that allows entrepreneurs to access NAIRR compute resources.
- The NAIRR should create a data consortium, potentially through a public-private partnership (PPP), that cultivates, enriches, and maintains datasets for public use.
- The NAIRR should facilitate the sharing of data brought by researchers to CloudBank and enable sharing by default.