model card

Described in a 2019 paper, see quotations.

model card (plural model cards)

1. (artificial intelligence) A document describing a machine learning model, often including information about training datasets, biases, benchmarks, and ethical considerations.

   Synonym: system card

   • 2019 January 29, Margaret Mitchell, Simone Wu, Andrew Zaldivar, Parker Barnes, Lucy Vasserman, “Model Cards for Model Reporting”, in Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT* '19), New York, NY, USA: Association for Computing Machinery, →DOI, →ISBN, pages 220–229:

     We propose model cards as a step towards the responsible democratization of machine learning and related artificial intelligence technology, increasing transparency into how well artificial intelligence technology works.

   • 2020 July 29, Huanming Fang, Hui Miao, “Introducing the Model Card Toolkit for Easier Model Transparency Reporting”, in Google Research, Google:

     To guide the Model Card creator to organize model information

   • 2020, Abhishek Wadhwani, Priyank Jain, “Machine Learning Model Cards Transparency Review : Using model card toolkit”, in 2020 IEEE Pune Section International Conference (PuneCon), →DOI, pages 133-137:

     Model cards are a very recent and hot topic of research. In Machine Learning (ML), transparency with model cards is relevant as it affects a wide range of domains, from health care to finance and jobs, etc.

   • 2025, Fabian Ferrari, “The Governanceof Generative AI: Three Conditions for Research and Policy”, in José Dijck, Karin Es, Anne Helmond, Fernando Vlist, editors, Governing the Digital Society: Platforms, Artificial Intelligence, and Public Values‎^[1], Taylor & Francis, →ISBN:

     So-called “model cards” have emerged as a standardization tool for AI developers to comprehensively document all key aspects of generative AI systems, including domainspecific training datasets, biases, and ethical considerations (Mitchell et al. 2019).