Model Cards for Model Reporting
By Margaret Mitchell, Simone Wu, Andrew Zaldivar, Parker Barnes, Lucy Vasserman, Ben Hutchinson, Elena Spitzer, Inioluwa Deborah Raji, Timnit Gebru. 2018
2,935 citations recorded by the source corpus.
Trained machine learning models are increasingly used to perform high-impact tasks in areas such as law enforcement, medicine, education, and employment. In order to clarify the intended use cases of machine learning models and minimize their usage in contexts for which they are not well suited, we recommend that released models be accompanied by documentation detailing their performance characteristics. In this paper, we propose a framework that we call model cards, to encourage such transparent model reporting. Model cards are short documents accompanying trained machine learning models that provide benchmarked evaluation in a variety of conditions, such as across different cultural, demographic, or phenotypic groups (e.g., race, geographic location, sex, Fitzpatrick skin type [15]) and intersectional groups (e.g., age and race, or sex and Fitzpatrick skin type) that are relevant to the intended application domains. Model cards also disclose the context in which models are intended to be used, details of the performance evaluation procedures, and other relevant information. While we focus primarily on human-centered machine learning models in the application fields of computer vision and natural language processing, this framework can be used to document any trained machine learning model. To solidify the concept, we provide cards for two supervised models: One trained to detect smiling faces in images, and one trained to detect toxic comments in text. 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. We hope this work encourages those releasing trained machine learning models to accompany model releases with similar detailed evaluation numbers and other relevant documentation.
Subfields: Artificial Intelligence, Political Science and International Relations, Safety Research, Safety, Robustness & Interpretability
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Mohit Bansal
- Luke S. Zettlemoyer
- Percy Liang
- Jianfeng Gao
- Yejin Choi
- Quoc V. Le
- Noah A. Smith
- Yoshua Bengio
In Political Science and International Relations
- Maarten Sap
- Yejin Choi
- Noah A. Smith
- Irene Solaiman
- Daniel S. Schiff
- Eric Michael Smith
- Adina Williams
- Christopher D. Manning
In Safety Research
- Emilio Ferrara
- Emily L. Denton
- Daniel S. Schiff
- A. Hanna
- Q. Liao
- Ben Hutchinson
- Irene Solaiman
- Inioluwa Deborah Raji
In Safety, Robustness & Interpretability
- Adina Williams
- Yulia Tsvetkov
- Eric Michael Smith
- Yejin Choi
- Kai-Wei Chang
- Dirk Hovy
- Noah A. Smith
- Percy Liang
Citations in the corpus
- Training Compute-Optimal Large Language Models (2022, positive): “We include a Chinchilla model card (Mitchell et al., 2019) in Table A8.”
- PaLM: Scaling Language Modeling with Pathways (2022, methodology): “Since entirely eliminating all such societal biases from data and models may not be feasible, it is essential to analyze and document such potential undesirable associations and risks through transparency artifacts such as model cards (Mitchell et al., 2019) and datasheets (Gebru et al., 2021).”
- Llama 2: Open Foundation and Fine-Tuned Chat Models (2023, background): “Table 52 presents a model card (Mitchell et al., 2018; Anil et al., 2023) that summarizes details of the models.”
- CTRL: A Conditional Transformer Language Model for Controllable Generation (2019, background): “• Finally, the README asks users to develop appropriate documentation (PAI, 2019; Arnold et al., 2018; Mitchell et al., 2019) when building on CTRL and to tell the research team how they are using CTRL by emailing ctrl-monitoring@salesforce.com .”
- RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models (2020, background): “Therefore, echoing calls for transparency in NLP research (Bender and Friedman, 2018; Mitchell et al., 2019; Dodge et al., 2019), we recommend researchers publicly release all relevant information during data collection (e.”
- Repairing the Cracked Foundation: A Survey of Obstacles in Evaluation Practices for Generated Text (2022, methodology): “formal evaluation processes (Mitchell et al., 2019; Ribeiro et al., 2020), we propose releasing evalu-”
- AI and the Everything in the Whole Wide World Benchmark (2021, methodology): “Raji and Buolamwini, 2019] and integrated into frameworks of standardized model reporting [Mitchell et al., 2019].”
- Designing Disaggregated Evaluations of AI Systems: Choices, Considerations, and Tradeoffs (2021, background): “2018) and model cards (Mitchell et al. 2019), may be of value here.”
- The GEM Benchmark: Natural Language Generation, its Evaluation and Metrics (2021, methodology): “In addition, we require every submitter to answer a series of questions that we will use to construct a model card (Mitchell et al., 2019) and externalize”
- Social Biases in NLP Models as Barriers for Persons with Disabilities (2020, background): “Due to the complex situational nature of these issues, release of NLP models should be accompanied by information about intended and non-intended uses, about training data, and about known model biases (Mitchell et al., 2019).”
- Datasheets for datasets (2018, background): “Researchers at Google published follow-up work on model cards that document machine learning models [20] and released a data card (a lightweight version of a datasheet) along with the Open Images dataset [17].”
- BLOOM: A 176B-Parameter Open-Access Multilingual Language Model (2022, methodology): “The license contains 13 behavioral-use restrictions that have been identified based on the intended uses and limitations described in the BLOOM Model Card, as well as the BigScience ethical charter.”
- OPT: Open Pre-trained Transformer Language Models (2022, methodology): “Following Mitchell et al. (2018), we provide a model card for OPT-175B.”
- Sustainable AI: Environmental Implications, Challenges and Opportunities (2021, background): “In addition, new models must be associated with a model card that, among other aspects of data sets and models [107], describes the model’s overall carbon footprint to train and conduct inference.”
- The RefinedWeb Dataset for Falcon LLM: Outperforming Curated Corpora with Web Data, and Web Data Only (2023, positive): “Table 7: Model card for Falcon-RW , following the framework introduced by Mitchell et al. (2019).”
- Scaling Instruction-Finetuned Language Models (2022, methodology): “A model card (Mitchell et al., 2019) is included in the appendix.”
- Anticipating Safety Issues in E2E Conversational AI: Framework and Tooling (2021, methodology): “Consider the methodology laid down for Model Cards in Mitchell et al. (2019) to clarify the intended use cases of machine learning models and minimize their usages that fall outside of these parameters.”
- Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design (2022, methodology): “Consider the methodology laid down for Model Cards by Mitchell et al. (2019) to clarify the intended use cases of machine learning models and minimise their usages that fall outside of these parameters.”