AI Accountability in Practice

🧠 What’s Covered

Foundations of Accountability

The workbook introduces accountability as a governing principle across AI design, development, and deployment. It focuses on two components:

Answerability: Ensuring human responsibility is clearly attached to all stages and decisions.
Auditability: Creating records and transparency that allow internal and external oversight.

These are translated into practices like traceability, justification of decisions in understandable language, and documentation.

Anticipatory vs. Remedial Accountability

The guide separates accountability into two key types:

Anticipatory (ex-ante): Planning and governance prior to AI deployment.
Remedial (ex-post): Addressing consequences, explanations, or harms after deployment.

This distinction is crucial for setting proactive governance rather than relying solely on retroactive fixes.

The Process-Based Governance (PBG) Framework

One of the workbook’s central innovations is the PBG Framework and Log, a structured, tabular method to track:

Governance actions
Roles and responsibilities
Timing and documentation
Mapping to project phases (design, development, deployment)

It uses tools like the Stakeholder Impact Assessment, Data Factsheet, Explainability Assurance Management (EAM), and Bias Risk Management, promoting holistic AI governance.

Hands-On Exercises and Case Study

The workbook includes a fictional case study—AI EduTech, a school district deploying an AI-powered educational platform. Participants are guided through structured activities:

Identifying accountability gaps
Mapping risks and governance actions
Assigning team roles
Completing accountability maps

The structured workshop approach is designed for public sector practitioners but can also support private orgs or civic tech actors.

💡 Why it matters?

Too many AI ethics frameworks stop at principles. This workbook provides something different: a scaffolded, concrete method to build accountability into day-to-day AI work. The Process-Based Governance model is particularly useful in sectors where decision-making must be transparent and traceable—such as education, health, policing, and public service delivery.

It also helps fill the “accountability gap” in complex multi-actor AI ecosystems by assigning and documenting roles at every lifecycle stage.

🔍 What’s Missing?

Private Sector Adaptation: While adaptable, the content is clearly targeted at UK public sector teams. More guidance on translating this to SMEs or multinationals would expand its utility.
Global Norms Integration: There’s limited explicit discussion of how this maps onto global standards (e.g., OECD AI Principles, ISO 42001).
Automation and Tooling: The workbook is manual and paper-heavy. While the digital companion is referenced, it would benefit from integrated templates or software toolkits (e.g., Notion dashboards, Git templates).
Cross-principle integration: The separation into different workbooks (e.g., fairness, safety, explainability) means teams must synthesize materials themselves when working across overlapping domains.