AI Sustainability in Practice

AI Ethics and Governance in Practice Programme
AI Sustainability
in Practice
Part Two: Sustainability Throughout
the AI Workflow
ParticipantWorkbook
Facilitator Workbook
Annotatedfor
Intended to participants
support facilitators
to engage
in delivering
with in the
preparation for,activities.
accompanying and during, workshops.
AI Sustainability in Practice
1
Part Two: Sustainability Throughout the AI Workflow
Acknowledgements
This workbook was written 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 creation of this workbook would not have been possible without the support and
efforts of various partners and collaborators. As ever, all members of our brilliant team
of researchers in the Ethics Theme of the Public Policy Programme at The Alan Turing
Institute have been crucial and inimitable supports of this project from its inception several
years ago, as have our Public Policy Programme Co-Directors, Helen Margetts and Cosmina
Dorobantu. We are deeply thankful to Conor Rigby, who led the design of this workbook
and provided extraordinary feedback across its iterations. We also want to acknowledge
Johnny Lighthands, who created various illustrations for this document, and Alex Krook and
John Gilbert, whose input and insights helped get the workbook over the finish line. Special
thanks must be given to the Ministry of Justice for helping us test the activities and review
the content included in this workbook. Lastly, we want to thank Youmna Hashem (The Alan
Turing Institute) and Sabeehah Mahomed (The Alan Turing Institute) for their meticulous
peer review and timely feedback, which greatly enriched this document.
This work was supported by Wave 1 of The UKRI Strategic Priorities Fund under the EPSRC
Grant EP/W006022/1, particularly the Public Policy Programme theme within that grant &
The Alan Turing Institute; Towards Turing 2.0 under the EPSRC Grant EP/W037211/1 & The
Alan Turing Institute; and the Ecosystem Leadership Award under the EPSRC Grant EP/
X03870X/1 & The Alan Turing Institute.
Cite this work as: Leslie, D., Rincón, C., Briggs, M., Perini, A., Jayadeva, S., Borda, A.,
Bennett, SJ. Burr, C., Aitken, M., Katell, M., Fischer, C., Wong, J., and Kherroubi Garcia, I.
(2023). AI Sustainability in Practice Part Two: Sustainability Throughout the AI Workflow.
The Alan Turing Institute.
2
Contents
31 Proportional Governance of Engagement Goals
About the Workbook Series and Methods
31 Deployment Phase Re-Assessment and

4 Who We Are Other Necessary Monitoring, Updating, and
Deprovisioning Activities
4 Origins of the Workbook Series
5 About the Workbooks
6 Intended Audience Activities
7 Introduction to This Workbook

34 Activities Overview
36 Interactive Case Study: AI in Urban

Key Concepts Planning
40 Stakeholder Profiles
10 Introduction to Sustainability: Stakeholder
Impact Assessments 45 Stakeholder Impact Assessment
(Design Phase)
14 A Closer Look at Stakeholder Impact
Assessments 46 Project Proposal
15 Stakeholder Impact Assessment Template 48 A Closer Look at the Model
21 Skills for Conducting Stakeholder Impact

54 Balancing Values
Assessments
58 Revisiting Engagement Method
21 Weighing the Values and Considering Trade-
Offs
67 Stakeholder Impact Assessment
22 Consequences-Based and Principles-Based (Deployment Phase)
Approaches to Balancing Values
21 Ensuring Meaningful and Inclusive Deliberation
27 Addressing and Mitigating Power Dynamics Further Readings

that May Obstruct Meaningful and Inclusive
Deliberation
74 Endnotes
28 Sustainability Throughout the AI Lifecycle
77 Bibliography and Further Readings
28 The Need for Responsiveness Across the AI
Lifecycle
77 AI in Urban Planning
30 Example in Focus: Challenges to AI
Sustainability in Children’s Social Care 78 Resources Informing Activities
3
About the AI Ethics and
Governance in Practice
Workbook Series
Who We Are
The Public Policy Programme at The Alan Turing Institute was set up in May 2018 with the
aim of developing research, tools, and techniques that help governments innovate with
data-intensive technologies and improve the quality of people’s lives. We work alongside
policymakers to explore how data science and artificial intelligence can inform public policy
and improve the provision of public services. We believe that governments can reap the
benefits of these technologies only if they make considerations of ethics and safety a first
priority.
Origins of the Workbook Series

In 2019, The Alan Turing Institute’s Public Policy Programme, in collaboration with the
UK’s Office for Artificial Intelligence and the Government Digital Service, published the
UK Government’s official Public Sector Guidance on AI Ethics and Safety.
Safety This document
provides end-to-end guidance on how to apply principles of AI ethics and safety to the
design, development, and implementation of algorithmic systems in the public sector. It
provides a governance framework designed to assist AI project teams in ensuring that the
AI technologies they build, procure, or use are ethical, safe, and responsible.
In 2021, the UK’s National AI Strategy recommended as a ‘key action’ the update and
expansion of this original guidance. From 2021 to 2023, with the support of funding from
the Office for AI and the Engineering and Physical Sciences Research Council as well
as with the assistance of several public sector bodies, we undertook this updating and
expansion. The result is the AI Ethics and Governance in Practice Programme, a bespoke
series of eight workbooks and a forthcoming digital platform designed to equip the
public sector with tools, training, and support for adopting what we call a Process-Based
Governance (PBG) Framework to carry out projects in line with state-of-the-art practices in
responsible and trustworthy AI innovation.
4
About the Workbooks
The AI Ethics and Governance in Practice Programme curriculum is composed of a series
of eight workbooks. Each of the workbooks in the series covers how to implement a
key component of the PBG Framework. These include Sustainability, Technical Safety,
Accountability, Fairness, Explainability, and Data Stewardship. Each of the workbooks also
focuses on a specific domain, so that case studies can be used to promote ethical reflection
and animate the Key Concepts.
Programme Curriculum: AI Ethics and Governance in Practice

Workbook Series
1 AI Ethics and Governance in 5 Responsible Data Stewardship

Practice: An Introduction in Practice
Multiple Domains AI in Policing and Criminal Justice
2 AI Sustainability in Practice 6
AI Safety in Practice
Part One
AI in Transport
AI in Urban Planning
3 AI Sustainability in Practice 7 AI Transparency and

Part Two Explainability in Practice
AI in Urban Planning AI in Social Care
4 8
AI Fairness in Practice AI Accountability in Practice
AI in Healthcare AI in Education
Explore the full curriculum and additional resources on the AI Ethics and Governance in
Practice Platform at aiethics.turing.ac.uk
aiethics.turing.ac.uk..
Taken together,, the workbooks are intended to provide public sector bodies with the skills
required for putting AI ethics and governance principles into practice through the full
implementation of the guidance. To this end, they contain activities with instructions for
either facilitating or participating in capacity-building workshops.
Please note, these workbooks are living documents that will evolve and improve with input
from users, affected stakeholders, and interested parties. We need your participation.
Please share feedback with us at [email protected]
[email protected].
5
Programme Roadmap
The graphic below visualises this workbook in context alongside key frameworks, values
and principles discussed within this programme. For more information on how these
elements build upon one another, refer to AI Ethics and Governance in Practice: An
Introduction.
Introduction

1 2 3 4 5 6 7 8

S F D S E A

C A R E
ACT
1 2 3 4 5 6 7 8

Intended Audience
This workbook series is primarily aimed at civil servants engaging in the AI Ethics and
Governance in Practice Programme - either AI Ethics Champions delivering the curriculum
within their organisations by facilitating peer-learning workshops, or participants
completing the programme by attending workshops. Anyone interested in learning about
AI ethics, however, can make use of the programme curriculum, the workbooks, and
resources provided. These have been designed to serve as stand-alone, open access
resources. Find out more at aiethics.turing.ac.uk
aiethics.turing.ac.uk.
There are two versions of each workbook:
• Facilitator Workbooks (such as this document) are annotated with additional guidance
and resources for preparing and facilitating training workshops.
• Participant Workbooks are intended for workshop participants to engage with in

preparation for, and during, workshops.
6
Introduction to This Workbook
This workbook is part two of two workbooks:
• AI Sustainaiblity in Practice Part One: Foundations for Sustainable AI Projects.

Projects
• AI Sustainaiblity in Practice Part Two: Sustainability Throughout the AI Workflow

(this workbook).
Both workbooks are intended to help facilitate the delivery of a two-part workshop on the
concepts of SUM Values and Sustainability.
AI Sustainability in Practice Part Two: Sustainability Throughout the AI

Workflow
This workbook explores how to put the SUM Values and the principle of Sustainability into
practice throughout the Design, Development, and Deployment Phases of the AI lifecycle.
It discusses Stakeholder Impact Assessments in depth, providing tools and training
resources to help AI project teams to conduct these. This workbook is divided into two
sections, Key Concepts and Activities:
Key Concepts Section
This section discusses frameworks for establishing the foundations for sustainable AI
projects:
1 2
Introduction to Sustainability: A Closer Look at Stakeholder Impact

Stakeholder Impact Assessments Assessments
3 4
Skills for Conducting Stakeholder Impact Sustainability Throughout the AI

Assessments: Consequences-Based and Lifecycle
Values-Based Approaches to Balancing Values
7
Activities Section
This section contains instructions for group-based activities (each corresponding to a

section in the Key Concepts). These activities are intended to increase understanding of
Key Concepts by using them.
Case studies within the AI Ethics and Governance in Practice workbook series are grounded
in public sector use cases, but do not reference specific AI projects.
Stakeholder Impact Assessment (Design Phase)
Practise answering key questions within Stakeholder Impact Assessments (SIAs).
Balancing Values
Practise weighing tensions between values when assessing the ethical permissibility
of AI projects by considering consequence-based and values-based approaches and
engaging in deliberation.
Revisiting Engagement Method
Practise undertaking practical considerations of resources, capacities, timeframes,

and logistics as well as stakeholder needs to establish an engagement method for the
following SIA.
Stakeholder Impact Assessment (Deployment Phase)
Practise using SIAs to formulate proportional monitoring activities for the development
and deployment of AI models.
Note for Facilitators
Additionally, you will find facilitator instructions (and where appropriate, considerations)
required for facilitating activities and delivering capacity-building workshops.
8
AI Sustainability in Practice Part Two:
Sustainability Throughout the AI Workflow
Key
Concepts
10 Introduction to Sustainability: Stakeholder 28 Sustainability Throughout the AI Lifecycle

Impact Assessments
28 The Need for Responsiveness Across the AI
14 A Closer Look at Stakeholder Impact Lifecycle
Assessments
30 Example in Focus: Challenges to AI
15 Stakeholder Impact Assessment Template Sustainability in Children’s Social Care
21 Skills for Conducting Stakeholder Impact 31 Proportional Governance of Engagement Goals

Assessments and Methods
21 Weighing the Values and Considering Trade- 31 Deployment Phase Re-Assessment and
Offs Other Necessary Monitoring, Updating, and
22 Consequences-Based and Principles-Based
21 Ensuring Meaningful and Inclusive Deliberation
27 Addressing and Mitigating Power Dynamics

Deliberation
Key Concepts 9
Introduction to Sustainability:
Stakeholder Impact Assessments
AI systems may have transformative and long-term effects on individuals and society.
Designers and users of AI systems should remain aware of this. To ensure that the
deployment of your AI system remains sustainable and supports the sustainability of the
communities it will affect, you and your team should proceed with a continuous sensitivity
to its real-world effects. You and your project team should come together to evaluate
the social impact and sustainability of your AI project through a Stakeholder Impact
Assessment (SIA).
The SUM Values introduced in the AI Sustainability in Practice Part One workbook form
the basis of the SIA. They are not intended to provide a comprehensive inventory of
moral concerns and solutions. Instead, they are a launching point for open and inclusive
conversations about the individual and societal impacts of data science research and AI
innovation projects. When starting a project, the SUM Values should provide the normative
point of departure for collaborative and anticipatory reflection. They should also allow for
the respectful and interculturally sensitive inclusion of other points of view.
Key Concepts Introduction to Sustainability 10

KEY CONCEPT
Stakeholder Impact Assessment (SIA)

Over the past few years, several different types of “impact assessment” have become
relevant for public sector AI innovation projects. Data Protection Law requires Data
Protection Impact Assessments (DPIAs) to be carried out in cases where the
processing of personal data is likely to result in a high risk to individuals.[1] DPIAs assess
the necessity and proportionality of the processing of personal data, identify risks that may
emerge in that processing, and present measures taken to mitigate those risks. Equality
Impact Assessments (EIAs) assist public authorities in fulfilling the requirements of the
equality duties, specifically regarding race, gender, and disability equality. They identify the
ways government can proactively promote equality.
DPIAs and EIAs provide relevant insights into the ethical stakes of AI innovation projects.
However, they go only part of the way in identifying and assessing the full range of
potential individual and societal impacts of the design, development, and deployment
of AI and data-intensive technologies. Reaching a comprehensive assessment of these
impacts is the purpose of SIAs. SIAs are tools that create a procedure for, and a means
of, documenting the collaborative evaluation and reflective anticipation of the possible
harms and benefits of AI innovation projects. SIAs are not intended to replace DPIAs or
EIAs, which are obligatory. Rather, SIAs are meant to be integrated into the wider impact
assessment regime. This demonstrates that sufficient attention has been paid to the ethical
permissibility, transparency, accountability, and equity of AI innovation projects.
The purpose of carrying out an SIA is multidimensional. SIAs can serve several purposes,
some of which include:
Help to build public confidence Underwrite well-informed

that the design and deployment decision-making and transparent
of your AI system has been done innovation practices.
responsibly.
Demonstrate forethought and

Facilitate and strengthen your due diligence not only within
accountability framework. your organisation, but also to the
wider public.
Shed light on unseen risks that

threaten to affect individuals and
the public good.

Your team should convene to evaluate the social impact and sustainability of your AI
project through the SIA at three critical points in the project delivery lifecycle:
SECTION 1: Design Phase

Problem Formulation
Carry out an initial SIA to determine the ethical

permissibility of the project. Refer to the SUM Values as a
starting point for the considerations of the possible effects of
your project on individual wellbeing and public welfare. You
should include stakeholder engagement objectives and methods
for the Development Phase SIA established in your initial
Project Summary Report (PS Report), described in the previous
workbook. This will help to consider public views in ways that
are proportional to potential project impacts, and appropriate
to team positionality. The participation of a more representative
range of stakeholders will bolster the inclusion of a diversity
of voices and opinions into the design and development
processes.[2] [3] [4] [5] The Design Phase SIA includes a revisitation
of the PS Report. Revisions of the engagement objectives and
methods, as well as other relevant revisions, should be reflected
in an update to the PS Report.
SECTION 2: Development Phase

Model Reporting
Once your model has been trained, tested, and validated,

you and your team should revisit your initial SIA to confirm
that the AI system to be implemented is still in line with the
evaluations and conclusions of your original assessment.
This check-in should be logged in the Development Phase
section of the SIA with any applicable changes added
and discussed. The method of stakeholder engagement
that accompanies the SIA process will have been initially
established in the PS Report and revisited in the Design
Phase SIA. This report should be revisited again during the
Development Phase SIA and updated where needed. At
this point you must also set a timeframe for re-assessment
once the system is in operation. The timeframes for these
re-assessments should be decided by your team on a case-
by-case basis, but should be proportional to the scale of
potential impact that the system might have on individuals
and communities it will affect.

SECTION 3: Deployment Phase
System Use and Monitoring
After your AI system has gone live, your team should

iteratively revisit and re-evaluate your SIA. These
check-ins should be logged in the Deployment Phase section
of the SIA with any applicable changes added and discussed.
Deployment Phase SIAs should focus on evaluating the
existing SIA against real-world impacts. They should also
focus on considering how to mitigate the unforeseen or
unintended consequences that may have ensued in the
wake of the deployment of the system. As with each SIA
iteration, the PS Report should be revisited at this point,
when objectives, methods, and timeframes for the next
Deployment Phase SIA are established.

A Closer Look at Stakeholder
Impact Assessments
Stakeholder Impact Assessments (SIAs) provide you with the opportunity to draw on the
learning and insights you have gained in your Stakeholder Engagement Processes (SEPs),
and on the lived experience of engaged people, in order to delve more deeply into the
potential impacts of your project. Your SIAs should enable you:
• To re-examine and re-evaluate the potential impacts you have already identified in your
PS Report.
• To contextualise and corroborate these potential impacts in dialogue with stakeholders,

when appropriate.
• To identify and analyse further potential impacts. By engaging in extended reflection

and giving stakeholders the opportunity (where appropriate) to uncover previously
unexplored harms, gaps in the completeness and comprehensiveness of the previously
enumerated harms can be identified.
To illustrate how to implement an SIA, we have provided a we have provided a three-

part template, with each part corresponding to a stage of the project development - from
Design through to Deployment. Section 1 guides the Design Phase, addressing Project
Planning, Problem Formulation, as well as revisitation of Stakeholder Analysis, Positionally
Reflection, and Engagement Objectives and Methods. Section 2 provides a touchpoint for
evaluation and reflection during Development Phase of models and outputs, and facilitates
ongoing model reporting. Section 3 supports ongoing ethical deliberation and reflection
during the Deployment Phase of resultant project outputs, recording relevant changes from
earlier iterations of the SIA.
You might find it helpful to refer back to the Project Summary Report found
in AI Sustainability In Practice Part One,
One while answering these questions.
Key Concepts A Closer Look at Stakeholder Impact Assessments 14

Download this template here on the AI Ethics and Governance in practice Platform.
Platform
Stakeholder Impact Assessment for:

SIA
Project Name
SECTION 1A: Design Phase (Project Planning)

General questions
Date completed: Team members involved: External stakeholders consulted:
1. Horizon-Scanning and the Decision to Design
Have you assessed whether building an AI d. the resources (material and human)
model or tool is the right solution to help available to your project;
you deliver the desired services given:
e. the nature of the policy problem you
a. the existing technologies and are trying to solve; and
processes already in place to solve the
problem; f. whether an AI-based solution is
appropriate for the complexity of its
b. current user needs; potential use contexts?
c. the current state of available data;
Do these initial assessments support the justifiability and reasonableness of choosing to build
an AI system or tool to help you deliver the desired services?
For more details on “Assessing if artificial intelligence is the right solution” see guidance by
the Office for AI and Central Digital and Data Office. For further details about understanding
user needs, see Section 1 of the Data Ethics Framework and the user research section of the
Gov.UK Service Manual.
Manual
• Has a thorough assessment of the human rights compliant business practices of all
businesses, parties, and entities involved in the value chain of the AI product or service
been undertaken? This would include all businesses, parties, and entities directly
linked to your business lifecycle through supply chains, operations, contracting, sales,
consulting, and partnering. If not, do you have plans to do this?

2. Goal-Setting and Objective-Mapping
a. How are you defining the outcome c. Is this translation justifiable given the
(the target variable) that the system general purpose of the project and the
is optimising for? Is this a fair, potential impacts that the outcomes
reasonable, and widely acceptable of its implementation will have on the
definition? communities involved?
b. Does the target variable (or its d. Where appropriate, have you
measurable proxy) reflect a reasonable engaged relevant stakeholders to
and justifiable translation of the gather input on their views about
project’s objective into the statistical reasonableness and justifiability of the
frame? outcome definition and target variable
determination?
3. Possible Impacts on the Individual
a. How, if at all, might the use of your d. How, if at all, might the use of your
AI system impact the abilities of system impact freedoms of thought,
affected stakeholders to make free, conscience, and religion or freedoms of
independent, and well-informed expression and opinion?
decisions about their lives? How might
it enhance or diminish their autonomy? e. How, if at all, might the use of your
system infringe on the privacy rights
b. How, if at all, might the use of of affected stakeholders, both on the
your system affect their capacities data processing end of designing the
to flourish and to fully develop system and on the implementation
themselves? end of deploying it? When appropriate,
this question should supplement the
c. How, if at all, might the use of your completion of a Data Protection Impact
system do harm to their physical, Assessment.
Assessment
mental, or moral integrity? Have risks
to individual health and safety been
adequately considered and addressed?
4. Possible Impacts on Interpersonal Relationships, Society, and the

Biosphere
a. How, if at all, might the use of your system adversely affect each stakeholder’s fair
and equal treatment under the law? Are there any aspects of the project that expose
historically marginalised, vulnerable, or protected groups to possible discriminatory
harm? These questions should supplement the completion of an Equality Impact
Assessment.
Assessment

b. Does the project aim to advance the i. Have you sufficiently considered the
interests and wellbeingof as many wider impacts of the system on future
affected individuals as possible? Might generations and on the planet and
any disparate socioeconomic impacts biosphere as a whole?
result from its deployment?
j. How could the use of the AI
c. How, if at all, might the use of system you are planning to build or
your system affect the integrity of acquire—or the policies, decisions,
interpersonal dialogue, meaningful and processes behind its design,
human connection, and social development, and deployment—lead
cohesion? to the discriminatory harassment of
impacted individuals?
d. How, if at all, might the use of your
system affect freedom of assembly and k. How could the use of the AI
association? system you are planning to build or
acquire—or the policies, decisions,
e. How, if at all, might the use of your and processes behind its design,
system affect the integrity of the development, and deployment—lead to
information ecosystem, the right to the disproportionate adverse treatment
diverse and reliable information, and of impacted individuals from protected
access to a plurality of ideas and groups on the basis of their protected
perspectives? characteristics?
f. How, if at all, might the use of your l. How could the use of the AI
system affect the right of individuals system you are planning to build or
and communities to participate in the acquire—or the policies, decisions,
conduct of public affairs? and processes behind its design,
development, and deployment—lead
g. How, if at all, might the use of your to the discriminatory harassment of
system affect the right to effective impacted individuals?
remedy for violation of rights and
freedoms, the right to a fair trial
and due process, the right to judicial
independence and impartiality, and
equality of arms?
h. Have the values of civic participation,

inclusion, and diversity been
adequately considered in articulating
the purpose and setting the goals of
the project? If not, how might these
values be incorporated into your
project design?

SECTION 1B: Design Phase (Problem Formulation)
Sector-Specific and Use Case-Specific Questions
In this section, you should consider the sector-specific and use case-specific issues
surrounding the social and ethical impacts of your AI project on affected stakeholders.
Compile a list of the questions and concerns you anticipate. State how your team is
attempting to address these questions and concerns. Where appropriate, engage with
relevant stakeholders to gather input about their sector-specific and use case-specific
concerns.
SECTION 1C: Design Phase

Revisiting Project Summary Report
1. Revisiting Stakeholder Analysis and Positionality
a. Do the stakeholder groups outlined in c. Do the stakeholder groups currently

the report accurately reflect current identified as salient represent those
stakeholders of this project? Are there groups that are currently likely to
other stakeholder groups that should be most differentially impacted,
be considered. vulnerable, or marginalised?
b. Do the potential impacts outlined in d. Does the team positionality reflection

the report accurately reflect current accurately represent the relationship
SIA results? between team members and
stakeholders at this stage in the
project?
2. Revisiting Engagement Objectives and Methods
a. Considering the results of the SIA, c. Do any adjustments need to be

are there any new potential project made to your chosen engagement
impacts that may lead you to objectives and methods given the SIA
reconsider your engagement objectives results? If so, are there any additional
and methods? If so, how? practical considerations that need
to be addressed to ensure that your
b. Do your chosen engagement objectives engagement objectives and methods
and methods seem proportional to the are realised?
current identified impacts?

3. Revisiting the Process-Based Governance (PBG) Framework
a. Considering SIA results, does the PBG Framework for this project still accurately reflect
the human chain of responsibility and create the baseline conditions for the project
team to be actively accountable for system impacts? (For further details on the PBG
Framework, see Workbook 8, AI Accountability in Practice.)

Model Reporting
After reviewing the results of your initial SIA, answer the following questions:
a. Are the trained model’s actual b. Have any other areas of concern arisen
objective, design, and testing results with regard to possibly harmful social
still in line with the evaluations and or ethical impacts as you have moved
conclusions contained in your original from the Design to the Development
assessment? If not, how does your Phase?
assessment now differ?
Re-Assess Questions in the Project Summary Report.
You must also set a reasonable timeframe for Public Consultation and Development Phase
re-assessment:
Dates of Public Consultation Date of Planned Development

on Development Phase Impact Phase Re-Assessment:
Revisitation:

SECTION 3: Deployment Phase
System Use and Monitoring
Once you have reviewed the most recent version of your SIA and the results of the public
consultation, answer the following questions:
a. What steps can be taken to rectify c. Have the maintenance processes for
any problems or issues that have your AI model adequately taken into
emerged? account the possibility of distributional
shifts in the underlying population?
b. Have any unintended harmful Has the model been properly retuned
consequences ensued in the wake of and retrained to accommodate
the deployment of the system? If so, changes in the environment?
how might these negative impacts be
mitigated and redressed?
Re-Assess Questions in the Project Summary Report.
You must also set a reasonable timeframe for Public Consultation and Deployment Phase re-
assessment:
Dates of Public Consultation Date of Next Planned

on Deployment Phase Deployment Phase Re-
Impacts: Assessment:

Skills for Conducting Stakeholder
Impact Assessments
Weighing The Values and Considering

Trade-Offs
Different AI projects may give rise to circumstances when the SUM Values come into
tension with each other. A s part of conducting SIAs, there should be discussion around how
to weigh values against one another when they do not align with one another or conflict.
Discussions should include considerations about potential trade-offs between values. F or
instance, there may be circumstances where the use of an AI system could optimally
advance the public interest only at the cost of safeguarding the wellbeing or the autonomy
of a given individual. In other cases, the use of an AI system could preserve the wellbeing
of a particular individual only at the cost of the autonomy of another, or of the public
welfare more generally.
The issue of adjudicating between conflicting values has long been a crucial and thorny
dimension of collective life. The problem of discovering reasonable ways to overcome the
disagreements that arise as a result of the plurality of human values has occupied thinkers
for just as long. Nonetheless, over the course of the development of modern democratic
and plural societies, several useful approaches to managing the tension between conflicting
values have emerged.
Key Concepts Skills for Conducting Stakeholder Impact Assessments 21

Consequences-Based and Principles-Based
We can find a concrete and agent-centred approach to managing the tension between
conflicting values in two of the standard schools of modern ethics:
• consequences-based moral thinking or consequentialism; and
• principles-based moral thinking or deontology.
These positions offer tools for thinking through a given dilemma in weighing values.*
A Consequences-Based Approach
A consequences-based approach asks that, in judging the moral
correctness of an action, you prioritise considerations of the goodness
produced by an outcome. In other words, the consequences of your
actions and the achievement of your goals matter most. The goodness
of these consequences should be maximised. In this view, standards of
right and wrong (indicators of what one ought to do) are determined
by the goal served as a result of an action taken, rather than by the
principles or standards one applies when acting.
A Principles-Based Approach
A principles-based approach takes the opposite track. From this
standpoint, the rightness of an action is determined by the intentional
application of a universally applicable standard, maxim, or principle. This
approach does not base the morality of conduct on the ends served by
it. Instead, it anchors rightness in the duty or obligation of the individual
agent to follow a rationally determined (and therefore “universalisable”)
principle. Deontological or principles-based ethics holds that the integrity
of the principled action and intention matters most, and such constraints
must be put on the pursuit of the achievement of one’s goals when the
actions taken as means to achieve these ends come into conflict with
moral standards.
* Learn more about ethics and governance in Leslie, D., & Fischer, C. (2023). Introduction to
Normative Ethical Theories. In AI Ethics and Governance (Turing Commons Skills Track). The Alan
Turing Institute. https://alan-turing-institute.github.io/turing-commons/skills-tracks/aeg/chapter1/
normative/

Knowing when to prioritise consequences and when to prioritise principles in moral
deliberations is a tricky matter. This may make sense depending upon the context.
To take a familiar example, lying to a murderer who appears at your front door would
save an innocent victim whom you are concealing in your cellar. The prioritisation of
consequences makes more sense than the prioritisation of the principle of not lying.
However, in another situation, the principle matters. For instance, where you would be
constrained, on principle, from deceiving others by taking credit for someone else's work in
order to advance in your job.

Consider a more directly relevant example. In an overburdened sector, the introduction
of an automated system for making building sites available for development would vastly
expedite housing delivery. The implementation of this AI system would thus produce a
consequence that could be beneficial to the public, protecting public good.[6]
Yet, it may, among other things, simultaneously do damage to the value of Connect.
This value safeguards interpersonal dialogue, meaningful human connection, and social
cohesion. The implementation of the AI system would eliminate time intensive consultation
processes that contribute to interpersonal communication, trust building, and social
bonding between council staff and residents.[7]
How then could one go about weighing the

value of improving public welfare against
the value of respecting the integrity of
interpersonal relations?
One way would be to place each side of

this comparison under the rubric of either
consequences or principles, and then measure
them up against each other accordingly.
From one perspective, the publicly beneficial
consequences of improving service delivery
might outweigh the publicly harmful
consequences of impairing social cohesion.
From another perspective, such a trade-off
would be unacceptable, because the principle
of respecting the integrity of social cohesion
trumps any solidarity-harming but publicly
beneficial consequences whatsoever.
Getting clear on the consequences and the principles involved

in a specific case of conflicting values will allow you to get a
better picture of the practical and moral stakes at play in a
particular project. It will also help you gain a sharper idea of
the proportionality of using of an AI technology to achieve a
desired outcome given both its potential ethical impacts and
the social needs to which it is responding. When drawn upon
for guidance, consequentialism and deontology can provide you
with a procedural scale upon which to place, measure, and weigh
conflicting values. They are practical tools that can be used to
enable meaningful deliberation.

Ensuring Meaningful and Inclusive
Deliberation
The most general approach to ensuring meaningful and inclusive deliberation is to

encourage respectful, sincere, and well-informed dialogue. Through this approach, reasons
offered from all affected voices can be heard and considered. Deliberations that have
been inclusive, open, and impartial tend to generate better and more sound conclusions.
Approaching the adjudication of conflicting values in this manner will likely improve
mutual understanding of the rationales and perspectives which inform those values.[8] The
importance of cultivating a culture of innovation, which encourages respectful, open, non-
coercive, and accountable communication, must be stressed. The success of the modern
sciences has been built on the dynamic foundations of inclusive, rational, and democratic
communication. This is perhaps evidence enough to support the validity of this emphasis.
The rational exchange and assessment of ideas and beliefs plays a central role in
meaningful dialogue about balancing values. The validity of the claims we make in
conversations about values is bounded by practices of giving, and asking for, reasons. A
claim about values that is justified is one that convinces by the unforced strength of the
better or more compelling argument. Rational justification and persuasive reason-giving
are, in fact, central elements of legitimate and consensus-oriented moral decision-making.
And, along the same lines, claims made about moral value or properties need to be
carefully evaluated in terms of their inferential strengths and weaknesses.
Meaningful Dialogue and Grice’s Maxims[9]

In 1975, the British philosopher of language, Paul Grice, formulated
what he called the “cooperative principle” to capture the assumptions
that 'interacting people hold when engaging in meaningful
conversation'. He broke these assumptions down into four maxims:
Maxim of Quantity Maxim of Quality

Make your contribution as informative Be truthful, and do not provide
as is required, providing the information that is misleading, false
necessary information but no more. or that is not evidence-based.
Maxim of Relation Maxim of Manner

Be relevant. Be clear. Avoid ambiguity. Avoid
obscurity of expression. Be brief. Be
orderly.

There is another way to understand the importance of meaningful dialogue in balancing
values. This alternative focuses on the enabling conditions of meaningful deliberation. It
focuses on how an inclusive and open exchange of reasons about balancing or prioritising
values can be made possible without imposing substanive views about the values
themselves. Here, an emphasis on rational communication in deliberations looks to secure
a justified and equitable process of exchanging and evaluating reasons. It starts with the
question: what are the preconditions and assumptions of meaningful communication that
allow people, who are exchanging views on their values and beliefs, to come rationally
acceptable moral judgments and reason-based agreement?
To answer this question, moral thinkers over the past century have endeavoured
to reconstruct the practical assumptions behind, and presuppositions of, rational
communication (a summary of the most essential of such assumptions and presuppositions
is provided below).[10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] Creating a reflective and practicable
awareness of these assumptions and presuppositions among members of your team can
play a crucial role in creating an innovation environment that is optimally conducive to
meaningful and inclusive deliberation:
Preconditions of Meaningful Deliberation
Impartiality Consistency and Coherence
Interlocutors engaging in meaningful deliberation Arguments and positions offered in meaningful

must consider the interests of all those who are deliberation must be clear, free from contradictions,
affected by their actions equally. Thinking impartially and hold together collectively in an understable way.
involves taking on the view of others to try to put
oneself in their place.
Mutual Respect and Egalitarian Reciprocity
Non-Coercion
All interlocutors must be treated with respect
and given equal opportunity to contribute to
Meaningful deliberation must be free from any sort the conversation. All voices are worthy of equal
of implicit or explicit coercion, force, or restriction consideration in processes of exchanging reasons.
that would prevent the open and unconstrained
exchange of reasons.
Inclusiveness and Publicity
Sincerity
Anyone whose interests are affected by an issue
and who could make a contribution to better
Meaningful deliberation must be free from any sort understanding it must not be excluded from
of deception or duplicity that would prevent the participating in deliberation. All relevant voices must
authentic exchange of reasons. Interlocutors must be heard and all relevant information considered.
mean what they say.

Addressing and Mitigating Power Dynamics
Deliberation
The stewardship of meaningful and inclusive dialogue is critical to safeguarding the

collective weighing up of values. However, there is an important potential barrier to
meaningful deliberation that challenges its feasibility and must be addressed. As guiding
assumptions of rational communication, norms like sincerity, impartiality, non-coercion,
and inclusiveness may strike some as overly idealistic. In the real world, discussions are
rarely fully inclusive, informed, and free of assertive manipulation, coercion, and deception.
Rather, deliberation and dialogue are often steered by, and crafted to protect, the interests
of the dominant.[21] Likewise, differential power relationships (for instance, divergent
educational backgrounds that derive from differential socioeconomic privileges) create
power imbalances that fundamentally challenge the conditions of reciprocity and equal
footing that are needed for justified and equitable communication.[22] [23] [24]
Your project team should confront these obstacles to meaningful deliberation

head-on through a power-aware approach to facilitating collaborative reflection,
dialogue, and engagement. By kindling an awareness of, and sensitivity to,
the differential relationships of power that can suppress the full participation
of disadvantaged or marginalised voices, you can better encourage an inclusive,
open, and equal opportunity conversation between participants.[25] [26] Clear-headed
explorations of power dynamics between civil servants, scientists, citizens, domain
experts, and policymakers can assist you in avoiding the kind of deficiencies of
representation and empowerment that risk reinforcing existing power structures and
inequalities.[27] [28] This may involve active mitigation measures like the provision of
training, upskilling, and technical resources to those who have lacked access to them.
Above all, the norms of meaningful deliberation make you aware of the possible
distortions of communication (i.e. a lack of egalitarian reciprocity, non-coercion,
sincerity, etc.) that must be tackled and rectified for the hurdles of power
disparities to be scaled.[29]

Sustainability Throughout
the AI Lifecycle
The Need for Responsiveness Across

the AI Lifecycle
In its general usage, the word 'sustainability'

refers to the maintenance of, and care for,
! SIAs must pay continuous attention
an object or endeavour over time. In the AI
both to the dynamic and changing
innovation context, this implies that building
character of AI production and
sustainability into your project is not a “one-
implementation lifecycles, and to
off” affair. Carrying out a SIA at the beginning
the shifting conditions of the real-
of an AI innovation project is a critical step.
world environments in which AI
However, it is only a first step in a much
models are embedded.
longer, end-to-end process of responsive
evaluation and re-assessment.
Key Concepts Sustainability Throughout the AI Lifecycle 28

There are two sets of factors that necessitate this demand for
responsiveness in sustainable AI innovation:
1. Production and Implementation Factors
Choices made at any point along the design, development, and

deployment workflow may impact prior decisions and assessments. This
leads to a need for re-assessment, reconsideration, and amendment. For
instance, design and development choices could be made that were not
anticipated in the initial impact assessment. Such choices might include
adjusting the target variable, choosing a more complex algorithm, or
grouping variables in ways that may impact specific groups. Changes
may influence how an AI system performs or how it impacts affected
individuals and groups. Processes of AI model design, development, and
deployment are also iterative and frequently bi-directional. This often
results in the need for revision and update. For these reasons, sustainable
AI design, development, and use must remain agile, attentive to change,
and at-the-ready to move back and forth across the decision-making
pipeline as downstream actions affect upstream choices and evaluations.
2. Environmental Factors
During the time in which the system is in production or use, changes

in project-relevant social, regulatory, policy, or legal environments may
occur. These changes may have a bearing on how well the model works
and on how the deployment of the system impacts affected individuals
and groups. Likewise, domain-level reforms, policy changes, or changes
in data recording methods may take place in the population of concern.
This could affect whether the data used to train the model accurately
portrays phenomena, populations, or related factors in an accurate
manner. In the same vein, cultural or behavioural shifts may occur within
affected populations that alter the underlying data distribution. This can
hamper the performance of a model, which has been trained on data
collected prior to such shifts. All of these alterations of environmental
conditions can have a significant effect on how an AI system performs.
They can also have a significant effect on the way it impacts affected
individual and communities.

Example in Focus: Challenges to AI
Sustainability in Children’s Social Care
In the context of children’s social care (CSC), the performance and impact of AI models
over time can be influenced by amendments to underlying laws or procedures (i.e. changes
to legal thresholds or definitions), population shifts, and alterations in social work practices
and protocols.[30] Where new reforms or changes to law or service delivery procedures have
taken place, the performance of predictive risk models, whose fit to the data distribution is
based on prior/outdated social and legal structures, could potentially even undermine such
reforms. For instance, a shift in the policy that determines the procedural steps involved in
taking a child into care may work on a different set of assumptions about, or definition of,
risk than that which was programmed into a predictive risk model designed before such a
policy change took place. This could lead to the model underestimating or overestimating
risk in a manner that is at cross-purposes with the policy reform and that harms children
and families.
Changes and improvements in CSC services

can likewise adversely affect the predictive
qualities of models trained on data from
the past. For example, whereas a parent’s
placement in foster care as a child and a
child’s prior contact with child services have
both been found to be predictive of child abuse
or neglect, the predictive power of such input
features rests on the assumption (backed by
statistics) that foster care and child services
interventions have been ineffective in the
past. Successful reforms that improve the
effectiveness of these services would diminish
the predictive power of these variables, so
a model that retains the inferences from
the prior data will end up identifying risks
inaccurately and could inequitably impact
affected decision recipient(s).
Similarly, the thresholds for access to CSC,

as applied by different local authorities,
are often adapted to the changing demand
for services and to resource availability.[31]
They are also influenced by changes in the
central government’s policy strategies. These
are often external to, and independent of,
the policy strategies of the local authorities
providing CSC services. Furthermore,
thresholds for access are frequently influenced

by the outcome of the Office for Standards in Education inspections. All these factors are
subject to unpredictable and frequent transformation. AI models used in CSC, especially
ones deployed for risk assessments, should be continually monitored, re-assessed for
potential stakeholder impacts, and updated whenever policies, procedures, and practices
are changed.
The effectiveness of your project team’s ability to bring AI sustainability into practice will
largely hinge on the governance actions and procedures you set up to ensure that the AI
innovation workflow is sufficiently responsive to changing production, implementation, and
environmental factors. These procedures and mechanisms should involve both the public-
facing, engagement dimension of your project and internal processes of reassessment,
updating, monitoring, and deprovisioning.
Proportional Governance of Engagement

Goals and Methods
When significant changes occur in the production, implementation, and environmental

factors over the course of the AI innovation lifecycle, your team will have to re-evaluate
its engagement objectives and methods. This will ensure that affected stakeholders are
appropriately consulted and involved. This means that impacted stakeholder groups are
re-engaged to provide input on the relevant changes. It may also mean that your team
chooses different engagement methods to align the level of engagement with the scale
of impact that the changes may generate. For details on assessing appropriate levels of
participation, please consult the Assessing Stakeholder Engagement Needs section of the
AI Sustainability in Practice Part One workbook.
Deployment Phase Re-Assessment and

Other Necessary Monitoring, Updating, and
A pre-implementation re-assessment of your initial impact assessment is included as a
second part of the SIA. This re-assessment directs you and your team to revisit your initial
SIA to confirm that the trained or completed AI system is still in line with the evaluations
and conclusions of your original assessment. It is at this juncture of the workflow that
any changes occurring in production or environmental factors surrounding the project
should first be identified, discussed, and addressed. Beyond this, a reasonable timeframe
for monitoring, re-assessment, and updating (once the model is in operation) should be
established that accords with the specific use-context and domain in which the system will
operate as well as the scale of its potential impacts. Likewise, a plan should be generated,
setting out details of the timeframes and procedures for re-assessment.

The re-training of the model may be planned around these timeframes to help maintain a
high-level of performance. This type of updating can use the original model as a starting
point, in order to retune the model’s parameters or, where appropriate, to drop certain
features that are no longer predictive. However, there is also the option of entirely
deprovisioning (i.e. stopping use of) the model and system if performance simply drops
too low to be addressed by mere re-training. Deprovisioning may not always mean simply
removing a system. An existing, but retired, project may serve as a foundational input or
constraint into the planning stages of a new project—starting the cycle once more.
At all events, Deployment Phase re-assessment and other necessary monitoring, updating,
and deprovisioning activities should be determined by:
• the specific use-context of the systems;
• changes in production and environmental factors that may influence the system’s
performance; and
• changes in the scale or scope of system impacts.

AI Sustainability in Practice Part Two:
Sustainability Throughout the AI Workflow
Activities
34 Activities Overview 45 Stakeholder Impact Assessment

(Design Phase)
36 Interactive Case Study: AI in Urban
Planning 46 Project Proposal
40 Stakeholder Profiles 48 A Closer Look at the Model
54 Balancing Values
58 Revisiting Engagement Method

(Deployment Phase)
Activities 33
Activities Overview
In the previous sections of this workbook, we have presented an introduction to the core
concepts of AI Sustainability. In this section we provide concrete tools for applying these
concepts in practice. Activities related to AI Sustainability in Practice Part Two will help
participants conduct and respond to SIAs throughout the design and development of AI
systems. Your team will continue engaging with the interactive case study presented
in the previous workshop, playing the role of a local authority developing an AI model
aimed to identify suitable building sites for housing development. Your team will plan
stakeholder engagement activities, schedule impact assessments, and determine how to
incorporate results from these engagements and assessments. These activities are to be
conducted following the completion of activities from AI Sustainability in Practice Part
One. Although new participants may join this session, outputs from the previous workshop
One
board are necessary materials for the delivery of this workshop.
We offer a collaborative workshop format for team learning and discussion about the
concepts and activities presented in the workbook. To run this workshop with your team,
you will need to access the resources provided in the link below. This includes a digital
board and printable PDFs with case studies and activities to work through.
Workshop resources for AI Sustainability in Practice Part Two
A Note on Activity Case Studies
Case studies within the Activities sections of the AI Ethics and Governance in Practice
workbook series offer only basic information to guide reflective and deliberative activities.
If activity participants find that they do not have sufficient information to address an issue
that arises during deliberation, they should try to come up with something reasonable that
fits the context of their case study.
Note for Facilitators
In this section, you will find the participant and facilitator instructions required for
delivering activities corresponding to this workbook. Where appropriate, we have included
Considerations to help you navigate some of the more challenging activities.
Activities presented in this workbook can be combined to put together a capacity-building

workshop or serve as stand-alone resources. Each activity corresponds to a section within
the Key Concepts in this workbook. Some activities have pre-requisites, which are detailed
on the following page.
Activities Activities Overview 34

Practise answering key questions within SIAs.
Corresponding Sections Pre-Requisites

→ Introduction to Sustainability: Stakeholder Impact ↗ Key Concepts: AI Sustainability in Practice (Part
Assessments (page 10
10)) One)
→ A Closer Look at Stakeholder Impact ↗ Activity: Establishing an Engagement Objective

14)) from AI Sustainability in Practice (Part One)
→ Stakeholder Impact Assessment Template

(Design Phase) (page 15
15))
Balancing Values
Practise weighing tensions between values when assessing the ethical permissibility
of AI projects by considering consequence-based and values-based approaches and
engaging in deliberation.

→ Skills for Conducting Stakeholder Impact ↗ Key Concepts: AI Sustainability in Practice (Part
21)) One)
Revisiting Engagement Method
Practise undertaking practical considerations of resources, capacities, timeframes,

and logistics as well as stakeholder needs to establish an engagement method for the
following SIA.

→ Sustainability Throughout The AI Lifecycle ↗ Key Concepts: AI Sustainability in Practice (Part
(page 28
28)) One)
→ Determining Stakeholder Engagement Methods ↗ Activity: Stakeholder Impact Assessment (Design

for Stakeholder Impact Assessments from AI Phase) (page 45
45))
Sustainability in Practice (Part One)
Stakeholder Impact Assessment (Deployment Phase)
Practise using SIAs to formulate proportional monitoring activities for the development
and deployment of AI models.
Corresponding Sections → Stakeholder Impact Assessment Template

→ Introduction to Sustainability: Stakeholder Impact (Deployment Phase) (page 20
20))
10))
Pre-Requisites
→ A Closer Look at Stakeholder Impact ↗ Key Concepts: AI Sustainability in Practice (Part
14)) One)
Activities Activities Overview 35

Interactive Case Study Recap:
Your team is a local planning authority within a

borough facing a housing crisis. The local poverty
rate is higher than the national average and
residents complain of sub-optimal living conditions.
1/2 Around half of your residents are renters,

Renting 60% of whom live in private lets. The
private letting sector is becoming
increasingly unaffordable.
60%
Private lets
25% 53%
The number of homeless applications The number of households in
has risen by 25% in the past temporary accommodation has
three years. risen by 53%, with an unprecedented
number of applications submitted since
2020.
Activities Interactive Case Study 36

A recent council investigation found that
terminated private tenancy leases are the
single greatest cause of homelessness in
the borough.
10,000 50%
New homes Affordable homes
Your council has established a

10-year housing plan set out to deliver 10,000
homes, 50% of which will be affordable. The
objective of this plan is to improve the living
standards of residents by developing as many
high-quality affordable homes as possible
over the next ten years.
The council has offered to

subsidise new residential buildings
that deliver at least 50% affordable
housing.
To support housing developments, your team will

need to expand your list of sites permitted
for planning applications. Achieving your
target would mean doubling the number of local
homes. Your team will need to review a much
higher volume of planning applications, which
may not be obtainable through your current
process.

Model Proposal
Current Method
Your current method for allocating new development sites can take
up to ten months to complete and considers a limited number of
sites proposed by developers, landowners, and estate agents.
These sites are manually reviewed by your team to ensure they meet
policy standards (i.e. sites’ ability to provide basic amenities) and are
suitable for development in practice.
Sites that pass your review process are taken forward for a
public consultation. This gives residents the opportunity to object to
certain sites being open for planning applications. Your team considers
public input to help determine which site proposals are accepted.
Accepted sites are made available for planning applications.

Applications are detailed development proposals demanding in-depth
review. Your team manually reviews individual applications
in a process that includes a second tier of consultations with
neighbours of specific sites.
Granting planning permissions can take up to three months per

application.

Proposed Method
Your council has suggested you automate this process by using an

machine learning (ML) model to automatically review every site in
the local area, classifying them as suitable or unsuitable for housing
development.[32] This approach would allow your team to scale-up the
number of sites considered for development. Whereas your current
method captures a number of submitted proposals, the model would
capture all local sites. This model would consider sites that are
outside the reach of your current method, such as council owned
buildings that could be repurposed, and private parcels that could
accept purchase offers.
Sites categorised as suitable would be reviewed by your team. Those

that pass this review process would be brought forward for a
three month public consultation which your team would consider
when accepting a final list of sites for development. Accepted
sites would be made public in a digital map and approved for
development, forgoing the additional three-month application
review process.
The proposed method would remove neighbour consultations

from the application review process. By removing time-consuming
steps, your team would be able to verify applications’ compliance with
building design standards and grant approvals or request adjustments
within two working weeks.
Your team conducted a Stakeholder Analysis and advised your council on how to engage
stakeholders throughout the Design and Development process. The council has reviewed
your advice and has decided to open your assessment to the public.
If the team advised on Partnering or If the team advised on Informing or

Empowering stakeholders... Consulting with stakeholders...
They have approved your engagement They deemed that this model would
objective, as they deemed that this model have significant social impacts and have
would have significant social impacts. Your decided that your team should partner
team is now to (partner with or empower) with stakeholders as an engagement
stakeholders when conducting SIAs for this objective. Your team is now to engage with
model. stakeholder when conducting SIAs for this
model.

Stakeholder Profiles
George
60 He/him Impacted Stakeholder
Profile
George is a 60-year-old black British man. He is a member of the Local Small Business
Association and owns a popular restaurant at one of the local high streets.
Goals and Aspirations

Locals love George’s restaurant. However, because of a shortage of supplies, he has had to
temporarily close the shop twice in the past year. Having opened back up, he is hoping to
increase sales especially given that the rent for his commercial space might increase when
his contract is up.
Alex
35 He/him Project Team Member
Profile
Alex is a 35-year-old Chinese man who lives in another borough. He is the Planning
Authority Lead and has been working for the council for the past six years. Alex traditionally
leads the site searching process for council developments, and will be involved in the
Housing Delivery Plan.

Alex is committed to delivering projects that directly involve the local community throughout
the process. Criticism of previous uses of ML for urban planning worry him, and he wants to
ensure that the housing plan values the knowledge of his team and the local community.
Activities Stakeholder Profiles 40

Hayley
40 She/her Impacted Stakeholder
Profile
Hayley is a 40-year-old white British woman on the housing register. She lives in an
overcrowded flat with her family of five, she is waiting for a bigger home ideally in proximity
to affordable childcare and a specialist school for her son who has autism spectrum disorder
(ASD).

Hayley and her husband’s living situation has been overcrowded since the birth of their
third child a year ago, and greatly worsened when they started working from home. The
lack of space has been extremely challenging for their eldest son in particular, with ASD.
Their current flat is in close proximity to their eldest son’s specialist school, and affordable
childcare for their two youngest. They are hoping to move to a bigger home that provides
this level of access as soon as possible.
Ali
Profile
Ali is a mixed-race 17-year-old local. He moved to the UK from Jamaica with his family when
he was two and has lived locally since. His family rents a house near a community garden he
helps run.

Ali spends his leisure time at a local community garden. The garden is small but run by a
committed group of community members. It sits within a greater green space where young
people like to gather. Ali is concerned with the high levels of development in the area. Not
only are the construction noises overwhelming, but he has noticed an increasing number
of green spaces being used for development over the past years. He is hoping this space
remains run by the community.

Terry
Profile
Terry is a 27-year-old black British man. He was born and raised locally and works at a local
corner store owned by a family friend.

Terry grew up in a local council estate where his neighbours helped raise him, as his mother
worked two jobs. After the estate was demolished for replacement by mixed income homes,
residents were relocated. In order to stay close to his job, Terry has since rented a room in
a private letting. He aspires to open his own shop but is struggling to make enough money
to get by. In recent years he has noticed more buildings being repurposed for mixed income
housing and rent prices going up. Terry isn’t certain he will be able to afford staying in the
neighbourhood for much longer.
Katherine
73 She/her Impacted Stakeholder
Profile
Katherine is a 73-year-old white British woman. She is a regular at the local library, leisure
centre, and church. She currently lives with her daughter and grandchildren but has recently
been placed on a priority waiting list within the housing register in order to move into council
housing that supports her mobility needs.

Katherine has a very social lifestyle which she loves and intends to maintain. She wants to
make sure her new home is close to leisure facilities and public transport in order to be able
to see her friends and family regularly.

Mia
28 She/her Project Team Member
Profile
Mia is a 28-year-old British Indian woman and data scientist who rents an apartment in
another borough. She has been a council employee for two years, and has experience using
a variety of ML techniques.

Mia is passionate about applying ML techniques to improve the quality of life of residents.
Having recently finished her work utilising predictive analytics to identify families in need of
support, Mia is eager to continue applying ML across council services, and is excited about
the prospect of supporting your housing delivery plan.
Nick
Profile
Nick is a 55-year-old white British man and electrician who has recently been placed in
emergency accommodation after losing his job and after his private tenancy agreement
wasn’t renewed.

Although Nick is receiving immediate support, he is finding it difficult to cope with the
difficult situation he finds himself in. He is a transgender man and is afraid of facing
harassment at the emergency shelters located in neighbourhoods he is unfamiliar with.
His goal is to find employment again and move into a home where he feels safe as soon as
possible.
Jamie
Profile
Jamie is a 31-year-old white British man and graphic designer for a creative agency.

Jamie moved into the area three years ago and has been renting privately since. He has
fallen in love with the neighbourhood and is a regular at the variety of coffee shops and
restaurants that have opened up in the past couple of years. He and his husband are looking
to buy a home locally.

Tom
Profile
Tom is a 59-year-old Black French real estate owner. He inherited a property portfolio that
includes a variety of local commercial properties, which he has been managing for around 10
years.

Tom’s business has experienced a lack of consistency in rental payments in the last two
years. Not all business were able to pay and some went bankrupt and terminated their
leases. Tom has stopped extending some leases and is looking to sell a portion of his
property.
Michael
31 He/him Project Team Member
Profile
Michael is a 31-year-old white British man and the product manager for the proposed
project.

Michael has wanted to make a career move towards working in AI projects for a while, and
has proposed this project as a way to support the housing delivery plan.

40 mins Participant Instructions
Stakeholder Impact
Assessment (Design Phase)
Objective
Practise answering key questions within SIAs.
Role Play
In this activity, your team will conduct a Design Phase SIA. Your group will be assigned
stakeholder profiles in order to consider a variety of perspectives that may be present in
stakeholder engagements.
Team Instructions
1. This activity will start with your facilitator 3. Once team members have read over the
reading out the activity context. They Project Proposal, the team will have
will split the team into groups, each with some minutes to answer the questions on
assigned personas. your assigned section of the Stakeholder
Impact Assessment (Design Phase).
2. Once groups have been assigned, take a Consider how each persona might respond
few minutes to individually read over the differently to questions.
Project Proposal. Team members are to
consider the note on case studies at the 4. A group member is to volunteer to write
beginning of the activities section of this answers on the board and report back to
workbook, imagining how stakeholders the team.
might relate to the content.
5. You will then reconvene as a team, having
volunteer note-takers share each group’s
answers to the questions and discussing
the answers.
6. Having discussed as a team, individually

use sticky notes to write answers to the
questions under the Sector-Specific and
Use Case-Specific Questions section.
Participant Instructions Stakeholder Impact Assessment (Design Phase) 45

Project Proposal
Design Development Deployment
1 2 3 4 5 6 7 8 9 10 11 12
Project Data Extraction Preprocessing & Model Testing & System System Use
Planning or Procurement Feature Engineering Validation Implementation & Monitoring
Problem Data Model Model User Model Updating or

Formulation Analysis Selection & Training Reporting Training Deprovisioning
Stages in focus
Problem Formulation
The proposed system is a visual interface that classifies sites as

suitable or unsuitable. The target variable of suitability would
indicate that sites classified as suitable would be reviewed by
your team. Those that pass the review process would be brought
forward for public consultation. Your team would consider public
input to help determine which suitable sites are made public and
DESIGN
pre-approved for planning applications for developments that
include at least 50% affordable housing.
Data Extraction
The model would use pre-labelled data from sites currently

allocated as accepted for residential use (suitable sites) and an
equal amount of non-allocated sites (unsuitable sites). DEVELOPMENT
Model Selection & Training
This system would use a Random Forest Classification Model (a

supervised ML model, described on page 49 49). The model is
detailed in the second part of this proposal.

System Implementation
When deploying the model, your team would import data

representing the entire region, gathered from Open Street
Maps (OSM), satellite imagery databases, land registry parcel
databases, and your council’s urban databases. This data would
be pre-processed by the model, organising it into features that
match the training data. The pre-trained model would then
classify sites and illustrate outputs in a digital map highlighting
suitable sites and providing key information for each.
DEPLOYMENT
Site Validation
Your team would review suitable sites and adjust your list as
deemed appropriate based on local policy, landowners’ interest
in development, and a public consultation. This process would
take no longer than three months since suitable sites will
reflect features of currently accepted sites, and key information
for validating sites will be found in a centralised web interface.
Outcome
Accepted sites will be made publicly available in the council
website. Planning applications for these sites will be deemed
pre-approved, waiving time-consuming elements of
planning application reviews, such as consultations
with neighbours. Your work reviewing applications would be
reduced to verifying compliance with building standards (i.e.
compliance with accessibility, health and safety) and requesting
any necessary adjustments. Application results are to be
delivered within two working weeks or deemed approved in the
event of no response. Your public map will be automatically
updated as permissions are granted, reflecting availability.

A Closer Look at the Model
The model is a Random Forest Classifier

trained to map patterns between features
(individual characteristics or properties within
the data) and suitability (target variable),
using a pre-labelled database of 1300 local
sites.[33]
This database would be split in a 70/30 ratio,

where 70% of the data (1000 evenly split
sites) would be used to train the model,
while 30% (300 evenly split sites) would be
concealed from the model during training, and
later used for testing.
Each site in the dataset is represented by features used in the Random Forest to determine
whether the site is suitable:
Area/dimensions (m2) Restricted location (listed buildings,

conservation areas, safety hazard areas)
Location (coordinates) (within local area)
Building materials
Address (zip code)
Market value (£)
Road access and presence of parking area
Current use
Access to energy, utilities, water,
waste management Current ownership
Policy area Proximity to leisure, recreation,

entertainment, green spaces (km)

Site
Tree 1 Tree 2 Tree n

Unsuitable Suitable Suitable
Majority Voting
Suitable
Random Forest models determine classifications based on the majority vote of a large
number of individual decision trees (flow charts analysing features that lead to a
classification).

40 mins Facilitator Instructions
Stakeholder Impact
Assessment (Design Phase)
1. Read the following statement to the team:
Our team has conducted an iteration of the Stakeholder Engagement Process and advised
our council on an engagement objective for this project. The council has determined we
will conduct our Design Phase SIA by (partnering with or empowering, depending
on your team’s engagement objective, determined in Part One of this workshop)
stakeholders and engaging them in a citizens’ panel. They have provided a model proposal
containing further details about the model and its intended placement within our team.
2. Give the team some minutes to read the Group 1 (Goal Setting and Objective
instructions for this activity. When they Mapping):
finish reading the instructions, ask them if
• Terry Impacted Stakeholder
they have any questions.
• Mia Project Team Member
3. Split team members into groups, each

assigned a section within the SIA Group 2 (Horizon Setting and the
(Design Phase) template on the board. Decision to Design):
Each group will answer two key questions
• Hayley Impacted Stakeholder
within each section.
• George Impacted Stakeholder
If your team has chosen Partner

• Tom Impacted Stakeholder
as an objective, assign the
profiles marked with either
Group 3 (Possible Impacts on the
Impacted Stakeholder or Project Team Member .
Individual):
If your team has chosen Empower • Alex Project Team Member
as an objective, only assign the profiles

• Katherine Impacted Stakeholder
marked with Project Team Member .
Group 4 (Possible Impacts on Society and

Interpersonal Relationships):
• Jamie Impacted Stakeholder
• Ali Impacted Stakeholder
• Nick Impacted Stakeholder
• Michael Project Team Member
Facilitator Instructions Stakeholder Impact Assessment (Design Phase) 50

4. Let the team know they have some 8. After all questions are shared, lead
minutes to individually read over the a discussion about the answers. Ask
Project Proposal and then to go through participants if there is anything they
the questions together. would like to adjust, take away, or add to
the answers.
5. Facilitators and co-facilitators should
• Co-facilitator: adjust any necessary
check in with each group, using the
answers to the questions based on the
Considerations section of this activity to
team discussion.
help answer any questions.
9. When the group discussion is up, ask the

6. When the time is up, ask the team to
team to use sticky notes to individually
reconvene.
write answers to the questions under the
Sector-Specific and Use Case-Specific
7. Give volunteer note-takers a few minutes
Questions section of the board, placing
to share their teams’ answers to the
their answers on the board.
questions in their section.

Facilitator Considerations Key Discussion Points for Potential Harms
When facilitating the discussion on potential harms, it may be useful to refer back to
the 'Origins of the SUM Values: Drawing principles from real-world harms' section of AI
Sustainability in Practice Part One.
One In particular, the mapping of risks that emerge from
the use of AI/ML technologies to the ethical concerns underwriting responsible AI/ML
research and innovation provides a helpful starting point for examining potential negative
impacts:[34]
Risks that Emerge From the Use of Ethical Concerns Underwriting

AI/ML Technologies Responsible AI/ML Research and
Innovation
Loss of autonomy, interpersonal Human agency and social

connection, and empathy interaction
For instance: residents may not be sufficiently consulted about a development project
that results from the AI model’s use and may thus lose a sense of agency and autonomy;
meaningful community participation in decision-making about local affairs may be
circumvented by the use of the model, harming social solidarity and interpersonal
connection;[35] the significance of the professional judgment of city planners and local
officials may be diminished through this form of automation, thereby harming their
agency and decision-making authority; certain citizens may be displaced or severely
inconvenienced as the result of the automated classification of suitability without having a
say in the way suitability is being defined by the system, thereby harming their sense of
autonomy and agency.[36]

Innovation
Poor quality outcomes Wellbeing of each and all
For instance, poor quality data (i.e. city records or property and land use information that
contain human errors), gaps in measurement (poor/inconsistent recording of geographic
proximity to essential services and amenities), or out-of-date information (i.e. dated/
obsolete information about current property use, essential services and amenities, or
access to utilities), can lead to outputs that inaccurately indicate a site’s suitability for
development and that end up harming the wellbeing of future residents, who then have to
inhabit inhospitable or deprived living environments.[37] [38] [39] [40]

Innovation
Bias, injustice, inequality, and Social justice, equity, public

discrimination interest, and the common good
For instance, where past determinations of suitability contained discriminatory or biased

patterns (as in cases where features like location operated as a proxy for socioeconomic status
or race or, where deficient access to essential services and amenities disqualified deprived
populations of development opportunities), the model trained on data containing such patterns
could replicate or augment those discriminatory harms and injustices.[41] [42] [43] Moreover, trends
to toward gentrification and the displacement of local, and potentially vulnerable, social groups
in urban development processes my influence AI project design decisions and overall project
planning choices. Considerations of these potential ecosystem-level inequities should factor into
impact assessment.

Balancing Values
Objective
Practise balancing and navigating tensions between values when assessing the ethical
permissibility of AI projects. Learn to employ consequences-based and principles-based
approaches when engaging in deliberation.
Activity Context
When answering questions about the possible impacts of the proposed system, members
of your team may have noticed that, at times, different values come into tension with each
other. Decisions that are made around how to balance these tensions can both influence
the direction that AI projects take, and shape their outcomes.
In this activity, your team will consider three values that can come into conflict when
considerations are undertaken about the ethical permissibility of using a model to
automate the site selection process. Your team is to consider these tensions from both
consequences-based and principles-based approaches, establishing a plan for how you will
balance these values. This plan will be used both to inform your recommendation to the
council on whether to develop the system, and to specify any amendments you would need
to the Project Proposal in order to consider it ethically permissible.
Your group will be assigned a pair of conflicting values in this activity. The goal is for your
group to keep in mind the identities and circumstances of the stakeholder profiles in order
to consider how they might evaluate tensions discussed in this activity.
Team Instructions
1. Split into groups, each assigned a pair

of conflicting values. Take a look at your
group’s section of the Conflicting Values
Diagram.
2. Fill out the Consequence-Based

Approaches, Principle-Based
Approaches, and Balancing Plan
sections, answering the prompts listed. Conflicting Values Diagram
Participant Instructions Balancing Values 54

3. When the time is up, reconvene as a 5. Individually take a look at the Potential
team, having your group’s volunteer note- Project Benefits and the Sector-
taker share your conflicting values and Specific and Use Case-Specific
plans with the team. Questions in the Stakeholder Impact
Assessment (Design Phase) activity, then
4. Engage in a team discussion about the vote on the question:
extent to which your plans address
anticipated concerns or questions, and • Should our team develop this system
how you might adjust them to better to automate our site selection process?
address these. Discuss any promising
aspects of developing this system, 6. Choose the notes within these sections
(considering your plans). which best explain your decision, marking
them with a dot.
• Your co-facilitator will write your
answers on sticky notes, placing them
under the Potential Project Benefits
section.
Potential Project Benefits
Sector-Specific and Use Case-Specific Questions in

the Stakeholder Impact Assessment (Design Phase)
Participant Instructions Balancing Values 55

Balancing Values
1. Give the team a moment to read over 5. Next, lead a team discussion about the
the activity instructions, answering any extent to which these plans address any
questions. anticipated questions or concerns in the
Design Phase SIA, and how you might
2. Next, split the team into groups, each adjust the plans to address these.
assigned a pair of conflicting values. Let
• Co-facilitator: Write these answers
the team know that they will have some
on sticky notes, placing them under
minutes to answer the questions and
the Potential Project Benefits
come up with a plan.
section on the board.
3. Facilitators and co-facilitators are to touch

6. Give team members a moment
base with each of the groups in this time,
individually take a look at the Potential
using the Considerations section of this
Project Benefits and the Sector-
activity to support.
Specific and Use Case-Specific
Questions from the Stakeholder Impact
4. When time is up, ask the team to
Assessment (Design Phase) activity, then
reconvene. Have each group's volunteer
vote on the question:
note-taker take a few minutes to share
their conflicting values and plans with the • Should our team develop this system
team. to automate our site selection process?
7. When the team has voted, let them know

that this decision, alongside the plans
developed in this activity, will be used to
draft your recommendation to the council.
Facilitator Instructions Balancing Values 56

Facilitator Considerations Approaches Within the Model Proposal
When facilitating discussion on the role that principle-based and consequence-based

approaches to ethical deliberation play in the model proposal, you may want to stress that
some tensions might be better addressed by drawing one approach and other tensions
by drawing on the other. In other words, at times and in certain cases, consequences
and outcomes matter more for resolving tensions (say, for instance, when the benefits of
an action far outweigh the costs of abiding by a principle). At other times and in certain
other cases, abiding by principles may matter more (for instance, when, regardless of
the consequences, adhering to a principle is crucial for upholding a cherished right or
maintaining individual integrity).
Facilitator Considerations Balancing Plan
In stewarding discussions of balancing plans, you might want to point workshop

participants to the 'SUM Values in Focus: Respect, Connect, Care, and Protect' section
of the AI Sustainability in Practice Part One workbook. The details of the ethical values
contained in this section provide a good launching pad for in-depth dialogue about
balancing tensions between them. 'The Values Map' section of the same workbook may
also be helpful in this activity. On the following page are some useful guiding questions
to assist workshop participants in resolving value tensions and in weighing up principles
against the benefits of using the model to speed up delivery of development sites.
Facilitator Instructions Balancing Values 57

Revisiting Engagement
Method
Objective
Undertake practical considerations of resources, capacities, timeframes, and logistics as
well as stakeholder needs to establish an engagement method for the following SIA.
Activity Context
Your team has conducted the Design Phase SIA and shared it with your council along with
your recommendation regarding the development of the proposed project. The council has
chosen to move forward with the project but incorporated the following amendments to the
project proposal:
Amendments
• The target variable of suitability will now indicate that sites categorised as
suitable will, once passing your team’s review and public consultation, be made
publicly available for developers to submit planning applications.
• Your team will review individual applications in a process that will include
consulting with neighbours of specific sites. Your team will use the model’s web
interface to assess centralised information about each site, streamlining the
process while enabling human oversight.
• The model will be deployed in a small-scale area and expanded in increments,

subject to SIAs at each phase.
With the help of Mia, your team’s Data Scientist, your team has designed and
developed the model. You are now in the Model Reporting step within the
Development Phase of the lifecycle and are finishing up your Development
Phase SIA, which you are conducting through a Citizens' Jury. You are finishing
up your Design to Development Phase SIA and need to schedule proportional
Development Phase assessments and engagement activities.
Participant Instructions Revisiting Engagement Method 58

Full Team Instructions (Part One)
1. Take a moment to individually look 2. Your team will be split into two groups. Go
over the activity instructions and to your relevant team's instructions below
Development Phase SIA, the Model to continue the activity.
Performance Metrics Report, and the
updates on the Project Lifecycle section
on the board.
Project Lifecycle
Group 1 Instructions
1. Take a moment to individually look over • What, if any, practical considerations

the Engagement Method Cards as regarding resources, capacities,
well as the Overview of Engagement timeframes, and logistics may
Resources and Constraints. You will pose constraints when selecting
consider your established engagement engagement methods?
objectives along with practical
considerations of resources, capacities, • What engagement methods may be
timeframes, and logistics to determine most feasible within these constraints?
which engagement method to put forward
for the next SIA. • Considering the above questions,
which engagement method would you
2. Consider the following questions: establish for the next SIA?
• Which methods meet your established 3. Your group note-taker is to write out
engagement objective? answers within the Notes section of this
activity.
• What resources are available for
conducting engagements? 4. Once instructed to by your facilitator,
jump to the Full Team Instructions .
Engagement Method Cards Overview of Engagement Resources and

Constraints

Group 2 Instructions
1. In your group, discuss the results of the 4. Your group note-taker is to write out
Development Phase SIA. Consider the answers within the Notes section of this
questions: activity.
• How might the model update harm 5. Once instructed to by your facilitator,
stakeholders? jump to the Full Team Instructions .
• To what extent do the model’s

performance metrics safeguard salient
stakeholders against poor quality
outcomes?
• To what extent do the updates to the

project plan (including changes to the
target variable and the incremental
deployment of the model) mitigate
potential risks posed within this
SIA?
2. Next, take a moment to individually look

over the Engagement Method Cards.
You will consider SIA results along with
stakeholder needs to put forward an
engagement method for the next SIA.
3. Consider the following questions:
• What accessibility requirements might

stakeholders have? Engagement Method Cards
• Will online or in-person methods

(or a combination of both) be
most appropriate to engage salient
stakeholders?
• Which methods meet your established

engagement objective?
• Considering the above questions,

what engagement method would you
establish for the next SIA?

Full Team Instructions (Part Two)
1. Reconvene as a group, having group note- 5. Consider what feedback mechanisms will
takers present their chosen methods and be in place.
each group’s reasoning.
6. Your co-facilitator will place the
2. Have a group discussion about what established Engagement Method Card
method might be best suited to balance on the appropriate section of the Project
practical constrains with stakeholder Lifecycle on the board, and outline
needs. engagement details within the card.
3. Vote on an engagement method.
4. Next, have a group deliberation about

the design of the engagement method,
considering:
• How might the team make sure that

this chosen method accommodates
different types of stakeholders?
• How might the team ensure that the

PS report used to conduct the SIA is
accessible to stakeholders?
• How might the team ensure that

engagement method feeds useful
information to your SIAs?

Model Reporting
After reviewing the results of your initial SIA, answer the following questions:
• Are the trained model’s actual objective, design, and testing results still in line
with the evaluations and conclusions contained in your original assessment? If
not, how does your assessment now differ?
RESULTS
The model has been adjusted to address concerns with the original
assessment:
1. Concerns with the meaning attributed to the target variable of

suitability:
• The target variable of suitability will now indicate that

sites categorised as suitable will, once passing your team’s
review and public consultation, be made publicly available for
developers to submit planning applications for projects composed
of at least 50% affordable housing.
2. Concerns with the degree of monitoring in the Deployment Phase:
• The Deployment Phase of this project has been adjusted for the
model to be deployed in increments, expansion being subject to
SIAs.
Testing results meet acceptable performance metrics.

• Have any other areas of concern arisen with regard to possibly harmful social or
ethical impacts as you have moved from the Development to the Deployment
Phase of this project?
RESULTS
The model has been updated to account for a new planning policy that
enables commercial buildings to be repurposed for housing development.
The model now considers commercial buildings as potentially suitable
sites. This update has enabled the model to identify suitable sites
accurately under current local policy.
During model development, our team determined that the model was not
generating enough suitable sites due to a feature indicating the
percentage of green or public space within sites. The model was only
classifying sites with a low percentage of green or public spaces
as suitable. Our team removed this feature in order for the model to
generate a greater number of sites, irrespective of the percentage of
public or green space within these.
As these changes were not accounted for in the Design Phase SIA, our
team will closely monitor potential issues and feedback.
Revisiting Stakeholder Analysis and Positionality
Considering the current state of the project as reflected in the

Development Phase SIA:
• Stakeholder groups identified in the PS report accurately reflect

current stakeholders in the project.
• There are new potential impacts posed by updates to this model,

such as the potential harm to local businesses that may be caused
by the model categorising commercial sites as suitable for housing
development, and our team promoting them for housing development.
• Considering these new potential impacts, local business owners are to

be considered increasingly salient stakeholders at this phase in the
project.
• Given that there have been no changes to the team’s composition or

identified stakeholders, the positionality reflection conducted in
the Design Phase remains relevant at this phase in the project.

Revisiting Engagement Objective
The engagement objective of partnering with stakeholders is considered
proportional to the new potential impacts outlined in the SIA results
and the team positionality reflection.
Model Performance Report

The following metrics summarise the model’s performance when predicting suitability
within unseen testing data. These metrics were reported with a 95% confidence
interval and error bars.
Precision
97% Number of true positives divided by the number of all sites classified as
suitable (true positives and false positives)
Accuracy
97% Number of correct classifications divided by total number of
classifications made
Recall
95% Number of true positives divided by the number of all actual suitable
sites (true positives and false negatives)
True Positive: True Negative:

Model correctly classifies a site as Model correctly classifies site as
suitable. unsuitable.
False Positive: False Negative:

Model incorrectly classifies site as Model incorrectly classifies a site as
suitable. unsuitable.

Revisiting Engagement
Method
1. Give the team some minutes to 7. After the team discussion, ask the team to
individually look over instructions for this vote on an engagement method.
activity, as well as the Development
• Co-facilitator: place the established
Phase SIA and the Model Performance
Engagement Method Card on the
Metrics Report.
'System Use and Monitoring' step
within the Project Lifecycle on the
2. When time is up, ask the team if they
board.
have any questions.
8. Next, lead a group deliberation about the

3. Next, split the team into groups, asking
design of the engagement, considering
for a volunteer note-taker from each
the questions:
group. Note-takers will be responsible for
reporting back group findings to the team. • How might the team make sure that
this chosen method accommodates
• Group 1 will be responsible for
different types of stakeholders?
discussing practical considerations of
resources, capacities, timeframes, and • How might the team ensure that the
logistics that may pose constraints PS Report used to conduct the SIA is
when selecting engagement methods. accessible to stakeholders?
• Group 2 will discuss stakeholder • How might the team ensure that
needs. method feeds useful information to
your SIA?
4. Give each group some minutes to decide
on an engagement method by discussing 9. Consider what feedback mechanisms will
the questions in their group instructions. be in place.
• Facilitators should join and support one • Co-facilitator: write engagement

group, and co-facilitators another. details on sticky notes within the
Engagement Details section on the
5. When time is up, ask the team to board.
reconvene, giving each group's note-
taker a few minutes to share their team's
chosen engagement method, as well as
their reasoning behind this decision.
6. After the two groups have shared, lead

a team discussion about what method
might be better suited to balance practical
constrains with stakeholder needs.
Facilitator Instructions Revisiting Engagement Method 65

Facilitator Considerations Engagement Methods
In facilitating discussion of the re-evaluation and re-crafting of engagement methods

and of setting timeframes for Development Phase re-assessment, you should stress
how proportional monitoring acts on the need for responsiveness across sustainable AI
lifecycles. You might want to emphasise the points made in the relevant passages from AI
Sustainability in Practice Part One:
One
Stakeholder analyses may be carried out in a variety of ways that involve

more-or-less stakeholder involvement. This spectrum of options ranges
from analysis carried out exclusively by a project team without active
community engagement, to analysis built around the inclusion of community-
led participation and co-design from the earliest stages of stakeholder
identification. The degree of stakeholder involvement will vary from project
to project based upon a preliminary assessment of the potential risks and
hazards of the model or tool under consideration.
AI Sustainability in Practice Part One
Facilitator Instructions Revisiting Engagement Method 66

Stakeholder Impact
Assessment (Deployment
Phase)
Objective
Practise using SIAs to formulate proportional monitoring activities for the development and
deployment of AI models.
Activity Context
Your team has deployed the system within an initial deployment area, and you are due to
conduct your first Deployment Phase SIA.
Team Instructions
1. In this activity, your team will be split into Part One: Production, Implementation, and
three groups. Each group will be assigned Environmental Factors
samples of stakeholder feedback that
represent greater stakeholder reactions to 3. In your groups, discuss how your assigned
the deployment of the model. feedback samples may be connected to
changes in production and implementation
2. Each group will have an assigned note- factors, or to environmental factors.
taker who is to record team discussions
on the group’s section on the board and • Revisit the section Case Study:
report back to the team, considering: Challenges to AI sustainability in AI
for Urban Planning of the workbook for
• What was the feedback sample support in this discussion.
and how, if at all, was it connected
to production, implementation, or Part Two: SIA Question
environmental factors?
4. Next, turn to the Deployment Phase
• What harmful impacts were raised by SIA question assigned to your group
this sample? on your section of the board, and have
a group discussion to come up with an
• Did your team decide updating or answer to the question. Consider any
deprovisioning was a better option? harmful impacts that have arisen from the
What informed this choice? deployment of this system.
Participant Instructions Stakeholder Impact Assessment (Deployment Phase) 67

Part Three: Updating or Deprovisioning
5. Having assessed impacts, your team 8. Reconvene as a group, having volunteer

is to discuss whether updating the note-takers share each group's decision
model may serve to mitigate possible and discussing each decision as a team.
harmful impacts, and/or to amplify
beneficial ones. Consider the possibilities
identified on your group's Updating or
Deprovisioning section.
6. As a group, decide on whether you believe

updating the model would be a feasible
solution for addressing the harmful
impacts of this model, or if deprovisioning
the model is a better option.
7. Your volunteer note-taker is to indicate

your decision on your group’s Updating
or Deprovisioning section.
Updating or Deprovisioning
Participant Instructions Stakeholder Impact Assessment (Deployment Phase) 68

Stakeholder Impact
Assessment (Deployment
Phase)
1. In this activity, your team will be split 5. Give the team enough minutes to conduct
into groups. Each group will be assigned this activity. Inform the team of the
samples of stakeholder feedback that maximum allocated minutes for each part
represent wider stakeholder reactions of this activity:
to the deployment of the model, and a
relevant Deployment Phase SIA question. • Part One: Production, implementation,
and environmental factors
2. Harmful impacts raised by each feedback - 10 minutes
sample are connected to a production,
implementation,or environmental • Part Two: SIA question
factor. Each group will deliberate on - 10 minutes
what production, implementation, or
environmental change their assigned • Part Three: Updating or deprovisioning
feedback sample may be connected to: - 15 minutes
• Groups are to answer their assigned 6. Facilitators and co-facilitators are to

question, identifying harmful impacts. touch base with each group and provide
support, using the Considerations section
• Lastly, groups will decide if updating of this activity.
the model would be a feasible solution
for addressing harmful impacts, or if 7. When time is up, ask the team to
deprovisioning the model is a better reconvene.
option.
8. Give note-takers a few minutes to report
3. Give the team a moment to individually back to the team. After each volunteer
read over the activity instructions, shares, give the team a few minutes to
answering any questions. discuss the decision.
4. Next, split the team into groups, asking 9. Once all decisions have been shared and
for a volunteer note-taker per group that discussed, consider the overall group view
will report back to the team. on updating or deprovisioning the model.
Based on the overall group view, choose
the corresponding scenario from the
following section to read out to the group.
These scenarios represent the outcomes
of updating or deprovisioning the model.
Facilitator Instructions Stakeholder Impact Assessment (Deployment Phase) 69

Facilitator Considerations Updating
Model has been updated to include current datasets, and an updating protocol has been
set up to ensure data remains timely and relevant. The feature indicating the percentage
of green or public spaces within sites was re-integrated into the model to ensure it
doesn’t select sites that have a significant amount of green space, and that less sites
are consequently selected. Model updates that would result in outputs that are at odds
with planning policy were not permitted, but the council has taken note of stakeholders’
feedback for further consultation on the policy itself.
Local development has continued to grow at a pace that meets the target in our 10-year
housing plan, and the model is utilising datasets that accurately reflect real-world access to
essential services, safeguarding the quality of outputs. Residents have responded positively
to the shift to decrease the scale of development and limit the percentage of green and
public spaces being used, although rising prices continue to be an area of concern.
Our team does, however, continue to receive negative feedback regarding the impact of
the model on local businesses. Residents critique the time-consuming nature of updating
local policy compared to the speed at which commercial sites are being repurposed for
housing. We also continue to receive negative feedback regarding what constitutes the
definition of affordable housing.
Facilitator Considerations Deprovisioning
Model deployment has been stopped while there is public consultation regarding:
1. the percentage of affordable housing that is deemed appropriate by local residents;
2. the definition of “affordable”; and
3. the constraints that are to be put on what types of commercial sites can be
repurposed for housing.
The outputs of this consultation will serve to define objectives of a new project, for
which components of the current project (i.e. re-validated datasets, model) will serve as
a foundation. Residents are happy to be involved in defining outputs once these points
are democratically addressed. The new project is likely to provide outputs that reflect
residents’ self-articulated interests.
The pace of local housing development has, however, temporarily returned to the growth
rate it had prior to the deployment of this project. This is challenging our team’s ability to
meet the targets in our 10-year housing plan.

Group 1 Activity Considerations and Answers
Feedback Samples
• Quote from Alex, Planning Authority Lead, highlighting increase in housing

development in the deployment area:
The model has attracted much more development in a short time period. Our
team has been able to review planning applications faster while considering
residents’ input. Using the model has been useful, but we will need to review all
available feedback prior to assessing next steps.
• Quote from Terry, Local Resident, highlighting harmful impacts of the model, including
high rates of development pricing-out of local residents:
We are seeing development left right and centre, bringing people from outside
the area who can actually afford to rent or buy. It doesn’t seem like the council is
interested in those of us who have always been here. There are new shops
none of us can afford, public spaces turned private, rent prices going up. Your
plan is helping change our neighbourhood for the worse. I myself need affordable
housing, but this plan is kicking us out.
• Quote from Ali, Local Resident, highlighting harmful impacts of the model including
green spaces being built over:
A planning application has been submitted for a development to be built over our
community garden. We won’t let this happen. The council needs to protect the
green spaces that make this neighbourhood a community.
Relevant Production, Implementation, or Environmental Factor
• Production Factor: The meaning attributed to target variable of suitability includes

promotion of sites for developments of at least 50% housing.
Deployment Phase SIA Question and Potential Answers
• Q: How does the content of the existing SIA compare with the real-world impacts of
the AI system as measured by available evidence of performance, monitoring data,
and input from implementers and the public?
• A: The deployment of the model seems both to confirm many of the concerns
expressed in the original SIA and to uncover new harms that were previously
unanticipated. Concerns about affordability and the inequitable impacts of gentrification
and displacement have been validated in light of the rapid pace of development and
the influx of new residents. Concerns about the diminishment of public and green
spaces have been confirmed by spreading privatisation and the filing of new planning
applications, though there is disagreement among residents about the costs and
benefits of streamlined planning.

Considerations for Updating or Deprovisioning
• There are likely significant constraints to changing meaning attributed to the target
variable, namely significant stakeholder consultation and council approval.
Feedback Samples
• Quote from Katherine, Local Resident, highlighting model categorising sites without
access to essential services as suitable:
It was such a relief to hear I was one of the first people offered a disabled-
adapted home in these new houses, I can’t even walk up the flight of stairs
in my current flat! It is a shame that the only leisure within two kilometres of
the building was closed last year. I took the house because I simply cannot
stay here, but I don’t know what I will do without my exercise routine. This
something that needs to be thought about.
• Environmental Factor: Change in real-world proximity to service provision (used as

features in the model) have diminished the predictive power of these features. The
model has retained the inferences from outdated data and is inaccurately classifying
sites as suitable.
• Q: Have the maintenance processes for your AI model adequately taken into account
the possibility of distributional shifts in the underlying population? Has the model been
properly re-tuned and re-trained to reflect changes in the environment?
• A: Katherine's feedback indicates that the model has not been adequately updated
to keep pace with the relevant distributional shift (i.e. that the closing of the leisure
centre has changed certain people's access to essential services). This suggests that
more frequent model updating may be necessary.
• Extracting or producing up-to-date datasets that accurately reflect the state of

essential services in the deployment area, and establishing monitoring protocols that
ensure the data used by the model remains current are likely feasible avenues to
addressing this issue.

Feedback Samples
• Quote from Mia, Project Data Scientist, highlighting the model’s ability to identify sites
that meet requirements set out in current planning policy:
Having tested, validated, and verified the system, our team was happy to see the
model perform with strong performance and safety metrics. Having incorporated
new features, our model is also up to date with local policy.
• Quote from George, local business owner, highlighting harmful impacts of promoting
commercial sites for residential repurposing, namely, it’s correlation with local
businesses not receiving rental contract renewals:
Your model is closing down long-standing local businesses. More and more
property owners are refusing to renovate our contracts. By publishing commercial
buildings, you have attracted purchase offers that small business owners simply
cannot match.
• Change in Environmental Factors: Change in policy standards regulating the

repurposing of sites was incorporated into a model update.
• Q: Have any unintended harmful consequences ensued in the wake of the deployment
of the system?
• A: Incentives for building owners to sell to property developers, who are converting
commercial buildings to residential properties, are driving local businesses out of
their spaces. Though the pace of local development is allowing for the local authority
to meet its targets, negative impacts on local businesses have been an unintended
harmful consequence of this success.
• Updating the model for it to not categorise commercial sites as suitable would entail
significant stakeholder consultation and is likely to raise tensions as the system’s
categorisations would be at odds with policy.

Endnotes
1 Information Commissioner’s Office. (2021). 8 Hartz-Karp, J. (2005). A case study in
Guide to the General Data Protection deliberative democracy: Dialogue with the
Regulation (GDPR). https://ico.org.uk/for- city. Journal of Public Deliberation, 1(1), 1-15.
organisations/guide-to-data-protection/ https://doi.org/10.16997/jdd.27
2 Harrington, C., Erete, S., & Piper, A. M. 9 Grice, H. P. (1975). Logic and Conversation.
(2019). Deconstructing Community- In P. Cole, & J. L. Morgan. (Eds.), Syntax and
Based Collaborative Design: Towards More Semantics, Volume 3, Speech Acts (pp. 41-
Equitable Participatory Design Engagements. 58). Academic Press.
Proceedings of the ACM on Human-Computer
Interaction, 3(CSCW), 1-25. https://doi. 10 Arendt, H. (1958). The human condition.
org/10.1145/3359318 University of Chicago Press.
3 OECD. (2005). Evaluating Public 11 Bachrach, P., & Baratz, M. (1962). Two
Participation in Policy Making. https://doi. faces of power. American Political Science
org/10.1787/9789264008960-en Review, 57(4), 947–952. https://doi.
org/10.2307/1952796
4 OECD. (2013). Government at a Glance 2013.
https://doi.org/10.1787/gov_glance-2013-en 12 Bohman, J. (2000). Public deliberation:
Pluralism, complexity, and democracy. MIT
5 Dawkins, C. E. (2014). The principle of press.
good faith: Toward substantive stakeholder
engagement. Journal of Business Ethics, 121, 13 Gutmann, A., & Thompson, D. (1996).
283-295. https://doi.org/10.1007/s10551- Democracy and disagreement. Harvard
013-1697-z University Press.
6 Catapult. (2019, May 11). Building a 21st 14 Habermas, J. (1984). The theory of
century digital planning system: A quick start communicative action I: Reason and the
guide. UKRI Innovate UK. https://cp.catapult. rationalization of society. Beacon Press.
org.uk/news/building-a-21st-century-digital-
planning-system-a-quick-start-guide/ 15 Hindess, B. (1996). Discourses of power: From
Hobbes to Foucault. Blackwell Publishers
7 Leslie, D. (2019). Understanding artificial
intelligence ethics and safety: A guide for 16 Cohen, J. (1989). Deliberation and democratic
the responsible design and implementation legitimacy. In A. Hamlin, & P. Pettit (Eds.), The
of AI systems in the public sector. The Alan good polity: normative analysis of the state
Turing Institute. https://doi.org/10.5281/ (pp. 17–34). Basil Blackwell.
zenodo.3240529
17 Manin, B. (1987). On legitimacy and political
deliberation. Political Theory, 15(3), 338–368.
74
18 McLeod, J. M., Scheufele, D. A., Moy, P., 25 Parsons, M., Fisher, K., & Nalau, J.
Horowitz, E. M., Holbert, R. L., Zhang, W., (2016). Alternative approaches to co-
Zubric, S., & Zubric, J. (1999). Understanding design: insights from indigenous/academic
Deliberation: The effects of discussion research collaborations. Current Opinion in
networks on participation in a public forum. Environmental Sustainability, 20, 99-105.
Communication Research, 26(6), 743–774. https://doi.org/10.1016/j.cosust.2016.07.001
https://doi.org/10.1177/009365099026006005
26 Lupia, A., & Norton, A. (2017). Inequality is
19 Przeworski, A. (1998). Deliberation and always in the room: Language & power in
ideological domination. In J. Elster (Ed.), deliberative democracy. Daedalus, 146(3), 64-
Deliberative Democracy (pp. 140–160). 76.https://doi.org/10.1162/DAED_a_00447
https://doi.org/10.1162/DAED_a_00447
Cambridge University Press.
27 Tschakert, P., Das, P. J., Pradhan, N. S.,
20 Landwehr, C. (2014). Facilitating deliberation: Machado, M., Lamadrid, A., Buragohain, M.,
The role of impartial intermediaries in & Hazarika, M. A. (2016). Micropolitics in
deliberative mini-publics. In Grönlund, collective learning spaces for adaptive decision-
K., Bächtiger, A., & Maija Setälä (Eds.), making. Global Environmental Change,
Deliberative mini-publics: Involving citizens 40, 182-194. https://doi.org/10.1016/j.
in the democratic process (pp. 77-92). ECPR gloenvcha.2016.07.004
Press.
28 Garcia, A., Tschakert, P., Karikari, N. A.,
21 See for instance Nagoda, S., & Nightingale, A. Mariwah, S., & Bosompem, M. (2021).
J. (2017). Participation and power in climate Emancipatory spaces: Opportunities for
change adaptation policies: Vulnerability (re) negotiating gendered subjectivities and
in food security programs in Nepal. World enhancing adaptive capacities. Geoforum,
Development, 100, 85-93. https://doi. 119, 190-205. https://doi.org/10.1016/j.
org/10.1016/j.worlddev.2017.07.022 geoforum.2020.09.018
22 Lupia, A., & Norton, A. (2017). Inequality is 29 Leslie, D., Katell, M., Aitken, M., Singh, J.,
always in the room: Language & power in Briggs, M., Powell, R., Rincon, C., Perini, A.
deliberative democracy. Daedalus, 146(3), 64- M., & Jayadeva, S. (2022). Data Justice in
76.https://doi.org/10.1162/DAED_a_00447
https://doi.org/10.1162/DAED_a_00447 Practice: A Guide for Policymakers. The Alan
Turing Institute in collaboration with The Global
23 Mendelberg, T., Karpowitz, C. F., & Oliphant, J. Partnership on AI. https://doi.org/10.5281/
B. (2014). Gender Inequality in Deliberation: zenodo.6429475
Unpacking the Black Box of Interaction.
Perspectives on Politics, 12(1), 18–44. http:// 30 Glaberson, S. K. (2019). Coding over
doi.org/10.1017/S1537592713003691 the cracks: Predictive analytics and child
protection. Fordham Urban Law Journal, 46(2),
24 Mendelberg, T., & Oleske, J. (2000). 307-363. https://ir.lawnet.fordham.edu/ulj/
Race and public deliberation. Political vol46/iss2/3
Communication, 17(2), 169-191. https://doi.
org/10.1080/105846000198468
75
31 The Office for Standards in Education, 37 Kilkenny, M. F., & Robinson, K. M. (2018). Data
Children’s Services and Skills (Ofsted). (2023). quality: “Garbage in–garbage out”. Health
Inspecting local authority children’s services. Information Management Journal, 47(3),
https://www.gov.uk/government/publications/ 103-105. https://journals.sagepub.com/doi/
inspecting-local-authority-childrens-services- pdf/10.1177/1833358318774357
from-2018/inspecting-local-authority-
childrens-services#inspection-methodology 38 Babbage, C. (1864). Passages from the life
of a philosopher. Longman, Green, Longman,
32 Sideris, N., Bardis, G., Voulodimos, A., Roberts, and Green.
Miaoulis, G., & Ghazanfarpour, D. (2019).
Using Random Forests on Real-World City 39 Mellin, W. (1957). Work with new electronic
Data for Urban Planning in a Visual Semantic ‘brains’ opens field for army math experts. The
Decision Support System. Sensors, 19(10), Hammond Times, 10, 66.
2266. https://doi.org/10.3390/s19102266
40 Suresh, H., & Guttag, J. V. (2019). A
33 Sideris, N., Bardis, G., Voulodimos, A., framework for understanding unintended
Miaoulis, G., & Ghazanfarpour, D. (2019). consequences of machine learning. arXiv
Using Random Forests on Real-World City preprint arXiv:1901.10002, 2(8). https://doi.
Data for Urban Planning in a Visual Semantic org/10.48550/arXiv.1901.10002
Decision Support System. Sensors (Basel,
Switzerland), 19(10), 2266. https://doi. 41 O'Neil, C. (2017). Weapons of math
org/10.3390/s19102266 destruction: How big data increases inequality
and threatens democracy. Crown.
34 Learn more about risks in urban planning
through Koseki, S., Jameson, S., Farnadi, 42 Prince, A. E., & Schwarcz, D. (2020).
G., Rolnick, D., Régis, C., Denis. J., Leal, Proxy discrimination in the age of artificial
A., de Bezenac, C., Occhini,, G., Lefebvre, intelligence and big data. Iowa Law Review,
H., Gallego-Posada, J., Chehbouni, K., 105(3), 1257-1318. https://ssrn.com/
Molamohammadi, M., Sefala, R., Salganik, abstract=3347959
R., Yahaya, S., & Téhinian, S. (2022). AI and
Cities Risks, Applications and Governance. UN- 43 d’Alessandro, B., O’Neil, C., & LaGatta, T.
Habitat. https://unhabitat.org/sites/default/ (2017). Conscientious classification: A data
files/2022/10/artificial_intelligence_and_cities_ scientist’s guide to discrimination-aware
risks_applications_and_governance.pdf classification. Big data, 5(2), 120-134. https://
doi.org/10.1089/big.2016.0048
35 Ryan, R. M., & Deci, E. L. (2017). Self-
determination theory: Basic psychological
needs in motivation, development, and
wellness. Guilford Publications.
36 World Economic Forum (2019). AI Governance:

A Holistic Approach to Implement Ethics into
AI [White Paper]. https://www.weforum.
org/whitepapers/ai-governance-a-holistic-
approach-to-implement-ethics-into-ai/
76
Bibliography and Further
Readings
Stakeholder Impact Assessment

AI Now Institute. (2018). Algorithmic Impact Karlin, M., & Corriveau, N. (2018). The government
Assessments: Toward Accountable Automation of Canada’s algorithmic impact assessment: Take
in Public Agencies. Retrieved from: https:// two. Retrieved from: https://medium.com/@
ainowinstitute.org/publication/algorithmic-impact- supergovernance/a-canadian-algorithmic-impact-
assessments-toward-accountable-automation-in- assessment-128a2b2e7f85
public-agencies
Reisman, D., Schultz, J., Crawford, K., & Whittaker,
Diakopoulos, N., Friedler, S., Arenas, M., Barocas, M. (2018). Algorithmic impact assessments: A
S., Hay, M., Howe, B., Jagadish, H. V., Unsworth, practical framework for public agency accountability.
K., Sahuguet, A., Venkatasubramanian, S., Wilson, AI Now institute. Retrieved from: https://
C., Yu, C., & Zevenbergen, B. (n.d.). Principles ainowinstitute.org/publication/algorithmic-impact-
for accountable algorithms and a social impact assessments-report-2
statement for algorithms. Fairness, Accountability,
and Transparency in Machine Learning. Retrieved Vallor, S. (2018) An ethical toolkit for engineering/
from: http://www.fatml.org/resources/principles- design practice. Retrieved from: https://www.scu.
for-accountable-algorithms edu/ethics-in-technology-practice/ethical-toolkit/
Karlin, M. (2018). A Canadian algorithmic impact

assessment. Retrieved from https://medium.com/@
supergovernance/a-canadian-algorithmic-impact-
assessment-128a2b2e7f85
77
Cabinet Office, & Geospatial Commission. (2021). Sideris, N., Bardis, G., Voulodimos, A., Miaoulis, G.,
Planning and Housing Landscape Review — & Ghazanfarpour, D. (2019). Using Random Forests
Executive Summary. https://assets.publishing. on Real-World City Data for Urban Planning in a
service.gov.uk/government/uploads/system/ Visual Semantic Decision Support System. Sensors
uploads/attachment_data/file/965740/Planning_ (Basel, Switzerland), 19(10), 2266. https://doi.
and_Housing_Landscape_Review.pdf org/10.3390/s19102266
Geospatial Commission. (2019). Future The Open Data Institute. (2020, August 6). Case
Technologies Review. https://www.gov.uk/ study: Unlocking data on brownfield sites.
government /publications/future-technologies- https://theodi.org/article/case-study-unlocking-
review data-on-brownfield-sites/
Ministry of Housing, Communities & Local

Government, & The Rt Hon Esther McVey MP. (2019,
November 5). PropTech dragons form new expert
property innovation council. GOV.UK. https://www.
gov.uk/government/news/proptech-dragons-form-
new-expert-property- innovation-council
Resources Informing Activities

Boal, A. (2002). Games for actors and non-actors.
London: Routledge.
Coimbra, T. C., & Caroli, P. (2020).

FunRetrospectives: Activities and Ideas for Making
Agile Retrospectives More Engaging. Amazon Digital
Services LLC - KDP Print US. https://books.google.
co.uk/books?id=1KHMzQEACAAJ
78
To find out more about the AI Ethics and
Governance in Practice Programme please visit:
aiethics.turing.ac.uk
This work is licensed under the terms of the Creative Commons Attribution
License 4.0 which permits unrestricted use, provided the original author
and source are credited. The license is available at:
https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode

AI Sustainability in Practice

Uploaded by

Copyright:

Available Formats

AI Sustainability in Practice

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

AI Sustainability in Practice

Uploaded by

Copyright:

Available Formats

AI Ethics and Governance in Practice Programme

31 Deployment Phase Re-Assessment and

5 About the Workbooks

6 Intended Audience Activities

7 Introduction to This Workbook

36 Interactive Case Study: AI in Urban

15 Stakeholder Impact Assessment Template 48 A Closer Look at the Model

21 Skills for Conducting Stakeholder Impact

21 Ensuring Meaningful and Inclusive Deliberation

27 Addressing and Mitigating Power Dynamics Further Readings

Origins of the Workbook Series

Programme Curriculum: AI Ethics and Governance in Practice

1 AI Ethics and Governance in 5 Responsible Data Stewardship

3 AI Sustainability in Practice 7 AI Transparency and

   

There are two versions of each workbook:

• Participant Workbooks are intended for workshop participants to engage with in

• AI Sustainaiblity in Practice Part One: Foundations for Sustainable AI Projects.

• AI Sustainaiblity in Practice Part Two: Sustainability Throughout the AI Workflow

AI Sustainability in Practice Part Two: Sustainability Throughout the AI

Key Concepts Section

Introduction to Sustainability: A Closer Look at Stakeholder Impact

Skills for Conducting Stakeholder Impact Sustainability Throughout the AI

This section contains instructions for group-based activities (each corresponding to a

Stakeholder Impact Assessment (Design Phase)

Practise answering key questions within Stakeholder Impact Assessments (SIAs).

Revisiting Engagement Method

Practise undertaking practical considerations of resources, capacities, timeframes,

Stakeholder Impact Assessment (Deployment Phase)

Note for Facilitators

10 Introduction to Sustainability: Stakeholder 28 Sustainability Throughout the AI Lifecycle

21 Skills for Conducting Stakeholder Impact 31 Proportional Governance of Engagement Goals

21 Ensuring Meaningful and Inclusive Deliberation

27 Addressing and Mitigating Power Dynamics

Key Concepts Introduction to Sustainability 10

Stakeholder Impact Assessment (SIA)

Help to build public confidence Underwrite well-informed

Demonstrate forethought and

Shed light on unseen risks that

Key Concepts Introduction to Sustainability 11

SECTION 1: Design Phase

Carry out an initial SIA to determine the ethical

SECTION 2: Development Phase

Once your model has been trained, tested, and validated,

Key Concepts Introduction to Sustainability 12

After your AI system has gone live, your team should

Key Concepts Introduction to Sustainability 13

• To contextualise and corroborate these potential impacts in dialogue with stakeholders,

• To identify and analyse further potential impacts. By engaging in extended reflection

To illustrate how to implement an SIA, we have provided a we have provided a three-

Key Concepts A Closer Look at Stakeholder Impact Assessments 14

Stakeholder Impact Assessment for:

SECTION 1A: Design Phase (Project Planning)

Date completed: Team members involved: External stakeholders consulted:

1. Horizon-Scanning and the Decision to Design

c. the current state of available data;

Key Concepts A Closer Look at Stakeholder Impact Assessments 15

3. Possible Impacts on the Individual

4. Possible Impacts on Interpersonal Relationships, Society, and the