Responsible Data Stewardship in Practice

📘 What’s Covered

The workbook is part of a broader series on AI ethics and governance, developed for UK public sector bodies. It serves both as a facilitator guide and a participant manual, blending theoretical groundwork with practical exercises. The focus is on helping civil servants and AI ethics champions understand and apply responsible data stewardship in real-world contexts.

Key Concepts Section:

Introduces the data lifecycle, showing its iterative and interconnected stages—from planning and collection to reuse and decommissioning (visualized on page 11).
Breaks down the three pillars of responsible data stewardship:
- Data Integrity: Ensures traceability, contemporaneity, and auditability.
- Data Quality: Covers accuracy, representativeness, fitness for purpose, and relevance.
- Data Protection & Privacy: Anchored in legal frameworks like the GDPR, highlighting consent, transparency, security, and proportionality.
Emphasizes data equity (pp. 21–22), linking stewardship to social justice by acknowledging and mitigating bias in data sources and uses—particularly in policing contexts.

Practice Section:

Offers a detailed Data Factsheet template (pp. 27–43) to be completed at various AI lifecycle stages. This template encourages reflection on project-specific data issues, ensuring ongoing scrutiny of risks like bias, security, and consent.
Encourages continuous use of the Factsheet as a live document, supporting iterative improvement over time.

Workshop Activities (pp. 45–57):

Designed around real-world public sector scenarios, including a case study on facial recognition in policing.
Activities help teams critically assess data risks, link stewardship to public interest, and collaboratively complete parts of the Data Factsheet.

Appendix A (pp. 58–61):

Tackles current challenges posed by generative AI and large-scale uncurated datasets.
Outlines serious risks like data poisoning, privacy breaches, and representational harm in foundation model training, connecting these back to the need for robust data stewardship.

💡 Why it matters? (500–600 characters)

Too often, data governance in AI projects is reactive or fragmented. This workbook flips the script by offering a structured yet flexible way to proactively build stewardship into every stage of an AI project. It stands out for its accessibility (no jargon overload), focus on real-life case studies, and clear links to legal, ethical, and social accountability. It’s especially timely given rising public concern over AI deployments in sensitive areas like law enforcement and healthcare.

🔍 What’s Missing

While comprehensive, the workbook could benefit from deeper discussion on:

The tensions between open data practices and privacy (especially in contexts like public surveillance).
Interoperability challenges when stewarding data across different government systems.
Concrete examples of failed stewardship—learning from what went wrong could make the workbook even stronger.It also assumes some familiarity with AI systems, which may be a hurdle for non-technical participants.

🧑‍💼 Best For

Ideal for civil servants, policy advisors, and public sector technologists working on AI-enabled services. Also useful for ethics officers and project leads who want a practical toolkit to ensure data handling aligns with legal duties and public expectations. Equally relevant for NGOs focused on government accountability.