Datasheets for Datasets
This paper proposes datasheets for datasets as a standardized documentation process for machine learning datasets, analogous to component datasheets in the electronics industry. The framework aims to increase transparency and accountability, mitigate unwanted societal biases, and help researchers select appropriate datasets by documenting motivation, composition, collection process, preprocessing, uses, distribution, and maintenance. The authors present a comprehensive set of questions and a workflow designed to be answered by dataset creators across the dataset lifecycle.
Summary and content fields are drawn from the document text and are pending editorial review.
Who issued it
Who it governs
Dataset creators and consumers in machine learning should read this to understand how to systematically document and evaluate datasets for responsible use.
What the document contains
The document covers a comprehensive set of questions and a workflow for documenting machine learning datasets, organized around key stages of the dataset lifecycle including motivation, composition, collection, preprocessing, uses, distribution, and maintenance.
The document does not provide a complete solution for eliminating societal biases or fully anticipating every possible dataset use, nor does it prescribe a single automated documentation method or enforce legal compliance judgments.
- Every machine learning dataset should be accompanied by a datasheet documenting its motivation, composition, collection process, and recommended uses.
- Datasheets for datasets promote transparency and accountability while helping consumers choose appropriate datasets and avoid unintended misuse.
- The datasheet creation process encourages dataset creators to reflect carefully on assumptions, risks, and potential harms throughout the dataset lifecycle.
- The framework includes specific guidance for datasets that relate to people, addressing compliance with regulations like the GDPR.
- Adopting datasheets requires modifying existing organizational workflows and incentives, but the community benefits outweigh the costs.
· 1 Introduction - 1.1 Objectives · 2 Development Process · 3 Questions and Workflow - 3.1 Motivation - 3.2 Composition - 3.3 Collection Process - 3.4 Preprocessing/cleaning/labeling - 3.5 Uses - 3.6 Distribution - 3.7 Maintenance · 4 Impact and Challenges · Appendix
How it scores
Each score is backed by a quote verified against the document text. Editorial review has not yet confirmed them.
Accountability
4/5Clear responsibility for AI outcomes
Datasheets for datasets have the potential to increase transparency and accountability within the machine learning community
1 Introduction
screening score · 2026-10-24
Fairness
3/5Non-discrimination and equitable treatment
a model is unlikely to perform well in the wild if its deployment context does not match its training or evaluation datasets, or if these datasets reflect unwanted societal biases.
1 Introduction
screening score · 2026-10-24
Human Dignity
0/5Respect for the inherent worth of persons
screening score · 2026-10-24
Human Oversight
0/5Humans remain in control of consequential decisions
screening score · 2026-10-24
Privacy
3/5Protection of personal data and private life
Some of the questions are designed to elicit information about compliance with the EU’s General Data Protection Regulation (GDPR) or comparable regulations in other jurisdictions.
3.2 Composition
screening score · 2026-10-24
Enforcement
0/5Consequences and compliance mechanisms with real force
screening score · 2026-10-24
Impact Assessment
2/5Required assessment of an AI system's risks/impacts before or during use
For dataset creators, the primary objective is to encourage careful reflection on the process of creating, distributing, and maintaining a dataset, including any underlying assumptions, potential risks or harms, and implications of use.
1.1 Objectives
screening score · 2026-10-24