概述
A datasheet for datasets does the same for training data: why the data was collected, what it contains, how it was gathered and labeled, and how it should be maintained. These documents matter because they let buyers, users, auditors and regulators judge whether a system is fit for purpose, and laws such as the EU AI Act now require similar documentation.
深入探讨
Model cards were proposed by Margaret Mitchell and colleagues in the paper "Model Cards for Model Reporting", presented at the ACM FAT* conference in 2019. The suggested sections are model details, intended use, factors (such as demographic or environmental conditions), metrics, evaluation data, training data, quantitative analyses, ethical considerations, and caveats and recommendations. The key idea is disaggregated evaluation: reporting performance separately for relevant subgroups rather than as one average score that can hide failures. Datasheets for Datasets, by Timnit Gebru and colleagues, first circulated in 2018 and was later published in Communications of the ACM. It borrows from electronics datasheets. Dataset creators answer questions about motivation, composition, collection process, preprocessing and labeling, uses, distribution and maintenance. Related formats include data statements for language datasets and data cards. System cards, used by Meta and OpenAI since 2022 and prominently for GPT-4 in 2023, document a whole deployed system, including safety testing and mitigations, rather than only the model. Regulation has turned much of this practice into a requirement. The EU AI Act requires technical documentation for high-risk AI systems, set out in Annex IV. Providers of general-purpose AI models must keep technical documentation, give information to downstream providers, and publish a summary of training content using a Commission template. A common misconception is that a model card proves a model is safe. It is a disclosure, not a certification, and a card can be incomplete, outdated or written like marketing. Its value depends on honest limitations and evaluations that someone else can reproduce.
战略影响
成本与预算
多年来，架构决策决定着性能和运营成本。
更清晰的判决
技术教育帮助团队选择正确的堆栈，而不仅仅是最新的堆栈。
质量控制
更好的工程选择可以减少生产中的可靠性事故。
The Future of Model Cards and Datasheets for Datasets
Documentation is moving from voluntary best practice toward a compliance requirement. The EU AI Act, procurement rules and standards such as ISO/IEC 42001 all expect structured records of how a system was built and tested. Machine-readable formats and automated generation will likely become more common, so that cards can be checked and compared. The ongoing challenges are keeping documents accurate as models are updated, disclosing enough about training data without exposing trade secrets or personal information, and making sure readers actually use the documents when deciding.
现实世界的实施
A hospital evaluating a skin-lesion classifier reads its model card and finds accuracy was reported separately by skin tone. It then decides to run its own local validation before use.
A Hugging Face model repository shows a model card with YAML metadata listing its license, training datasets, languages and benchmark results, and people can filter search results by those fields.
A research team publishes a speech dataset with a datasheet that records speakers' consent, recording conditions and the accents that are under-represented.
An AI lab releases a system card for a chatbot describing red-teaming results, safety mitigations and the risks that remain, beyond what a basic model card covers.
风险与防护栏
优化一项基准测试可以隐藏更广泛的系统弱点。
基础设施和维护成本常常被低估。
随着系统变得更加复杂，安全性和可观察性差距可能会扩大。
实施路线图
在实施之前定义延迟、质量和成本目标。
在实际负载和数据条件下进行基准测试。
仪器监控错误、漂移和用户影响。
在扩展之前准备回滚和事件响应路径。
不断探索
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Model Cards and Datasheets for Datasets quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
常见问题
What is Model Cards and Datasheets for Datasets?
A model card is a short, structured document that explains what an AI model is for, how it was evaluated, how well it performs for different groups and conditions, and where it should not be used. A datasheet for datasets does the same for training data: why the data was collected, what it contains, how it was gathered and labeled, and how it should be maintained. These documents matter because they let buyers, users, auditors and regulators judge whether a system is fit for purpose, and laws such as the EU AI Act now require similar documentation.
What is the main purpose of a model card?
A model card gives readers what they need to judge whether a model fits a particular use, including where it performs poorly.
Who led the paper that proposed model cards?
Margaret Mitchell and colleagues presented "Model Cards for Model Reporting" at ACM FAT* in 2019.
What is disaggregated evaluation?
Splitting results by subgroup exposes failures that an overall average would hide.
Datasheets for Datasets borrows its idea from which field?
Electronic components come with datasheets that describe their characteristics. Gebru and colleagues applied that idea to datasets.
How does a system card differ from a basic model card?
System cards, such as OpenAI's for GPT-4, describe a deployed product's risks, red-teaming and safeguards.
继续学习
相关指南
为此主题精选的更多指南