FAIR for AI: An interdisciplinary and international community building perspective
E. A. Huerta, Ben Blaiszik, L. Catherine Brinson, Kristofer E. Bouchard, D. Díaz, C. Doglioni, J. Duarte, Murali Emani, Ian Foster, Geoffrey Fox, Philip Harris, L. Heinrich, Shantenu Jha, Daniel S. Katz, Volodymyr Kindratenko, Christine R. Kirkpatrick, K. Lassila-Perini, Ravi Madduri, M. S. Neubauer, Fotis Psomopoulos, A. Roy, Oliver Rübel, Zhizhen Zhao, Ruike Zhu
Abstract
The production, collection, and curation of data require painstaking planning and the use of sophisticated experimental and computational facilities. In order to maximize the impact of these investments and create best practices that lead to scientific discovery and innovation, a diverse set of stakeholders defined a set of findable, accessible, interoperable, and reusable (FAIR) principles in 2016 1 , 2 . The original intent was that these principles would apply seamlessly to data and all scholarly digital objects, including research software 3 , workflows 4 , and even domain-specific custom digital objects 5 . However, because they were specifically written in the context of data, it became clear over time that the original set of FAIR principles would have to be translated or reinterpreted for digital assets beyond data 6 , 7 . This realization has led to initiatives that have proposed and/or developed practical FAIR definitions for research software and workflows, and more recently, for artificial intelligence (AI) models 8 , 9 .