⚡ Quick Summary
Published by The Alan Turing Institute, this is the facilitator-annotated Part Two of the AI Ethics and Governance in Practice Programme's workbooks on sustainability, subtitled Sustainability Throughout the AI Workflow. It shows public sector teams how to put the SUM Values and the principle of Sustainability into practice across the Design, Development and Deployment phases of the AI lifecycle, with the Stakeholder Impact Assessment (SIA) as the central instrument. The workbook argues that an initial SIA is only a first step: production, implementation and environmental factors change, so teams must re-assess, monitor, update or deprovision systems as conditions shift. It supplies a three-part SIA template, guidance on weighing conflicting values through consequences-based and principles-based reasoning, the preconditions of meaningful deliberation, and four timed workshop activities built on an urban planning case study in which a local authority proposes a Random Forest classifier to identify housing development sites. Participants work with eleven stakeholder profiles, a project proposal, model performance metrics and deployment-phase feedback samples.
🧩 What’s Covered
An introductory section frames the workbook before the Key Concepts and Activities that follow.
- Programme background: the series builds on the UK Government's official Public Sector Guidance on AI Ethics and Safety (2019) and a 2021 National AI Strategy recommendation; eight workbooks each cover one component of the Process-Based Governance (PBG) Framework — Sustainability, Technical Safety, Accountability, Fairness, Explainability and Data Stewardship — paired with a domain.
- Introduction to sustainability: SIAs are distinguished from Data Protection Impact Assessments required by Data Protection Law and from Equality Impact Assessments; they complement rather than replace these obligatory assessments and document the collaborative evaluation of potential harms and benefits.
- The SIA template: three parts with dated completion fields covering Design Phase project planning, problem formulation and revisitation of the Project Summary Report and engagement objectives; Development Phase model reporting; and Deployment Phase system use and monitoring, each scheduling public consultation and re-assessment.
- Skills for conducting SIAs: weighing values and trade-offs, consequences-based versus principles-based approaches, Grice's maxims of quantity, quality, relation and manner, the preconditions of meaningful deliberation, and power-aware mitigation of obstructive power dynamics.
- Lifecycle responsiveness: production and implementation factors, environmental factors, a children's social care example on predictive risk models, proportional governance of engagement goals and methods, and monitoring, updating and deprovisioning.
- Activities: the urban planning case study with stakeholder profiles and four timed activities — SIA (Design Phase) 40 minutes, Balancing Values 45, Revisiting Engagement Method 65, SIA (Deployment Phase) 35 — with facilitator considerations and answers.
- Model detail: a Random Forest Classifier trained on a pre-labelled database of 1,300 local sites split 70/30, listed site features, and reported precision of 97%, accuracy of 97% and recall of 95% at a 95% confidence interval.
- Reference apparatus: endnotes and a bibliography with further readings on stakeholder impact assessment and AI in urban planning.
💡 Why it matters?
Public sector teams must be able to show why an AI project is justified, who may be harmed and how decisions about value conflicts were reached. This workbook provides a repeatable procedure for that: a template that sits alongside DPIAs and EIAs and prompts on autonomy, wellbeing, integrity, privacy, non-discrimination, the information ecosystem and future generations. Its treatment of value conflict — consequences-based and principles-based reasoning, the preconditions of meaningful deliberation, and power-aware facilitation — moves teams beyond listing impacts. The lifecycle framing makes re-assessment, monitoring, updating and deprovisioning explicit governance activities with timeframes proportional to potential impact, linking project practice to the PBG Framework covered elsewhere in the programme.
❓ What’s Missing
The workbook assumes access to Part One, the linked Miro board and outputs from the previous workshop, so it does not function entirely on its own. Sustainability is treated in social and organisational terms; environmental or computational footprint is not discussed. The SIA template offers no thresholds for judging when an impact is too severe, and decisions about proportional timeframes for re-assessment and monitoring are left to team judgement. Model figures (97% precision, 97% accuracy, 95% recall) are presented without methodology, and the case study is explicitly fictional. Legal duties are referred to rather than set out, and the framing and examples are drawn from UK public sector practice.
👥 Best For
Facilitators and AI Ethics Champions delivering the programme's workshops; public sector project teams and analysts conducting Design, Development and Deployment phase stakeholder impact assessments; governance, policy and compliance staff who need a documented method for weighing impacts and values; and data scientists or product managers supplying model reporting, performance metrics and monitoring evidence.
📄 Source Details
AI Sustainability in Practice Part Two: Sustainability Throughout the AI Workflow, by David Leslie, Cami Rincón, Morgan Briggs, Antonella Perini, Smera Jayadeva, Ann Borda, SJ Bennett, Christopher Burr, Mhairi Aitken, Michael Katell, Claudia Fischer, Janis Wong and Ismael Kherroubi Garcia, The Alan Turing Institute, 2023, Version 1.2, 79 pages, English. Annotated facilitator edition of the AI Ethics and Governance in Practice Programme workbook series. The text was extracted in full from all 79 pages. The document prints turing.ac.uk/ai-ethics-governance and a Creative Commons licence URL; no URL for the workbook itself is given.
