Machine learning (ML) and other big data synthesis and prediction techniques make it possible to use artificial intelligence (AI) for decision making. With these pivotal technology advances and the emerging field of data-centric AI, robust research needs to occur alongside these innovations in several other dimensions. For example, an understanding of the relationship of FAIR data to ML accuracy and performance; the reproducibility of ML output and how this varies between software stacks, hardware processor types, and environmental effects; and how researchers leading data facilities can make it easier to use ML on their collections. The researchers studying these areas need sufficient collective organizing capacity to keep pace alongside ML adoption.
The term FAIR originates from the 2014 Lorentz Workshop resulting in 15 guiding principles published in 2016 to make research Findable, Accessible, Interoperable, and Reusable. (See more on the 15 guiding principles.)
The RCN will concentrate on three themes: FAIR in ML, AI readiness, and AI reproducibility, chosen for the urgent needs of researchers and alignment with NSF CISE and GEO priorities.
Better Practices for AI
Published 9 Jan 2023 | doi: https://doi.org/10.1038/d41586-023-00023-2
This Nature article asks the question: Health-care systems are rolling out artificial-intelligence tools for diagnosis and monitoring. But how reliable are the models?