Guidelines for AI

in parliaments
D

December 2024
Table of contents
Forewords 3
Introduction 6
About the Guidelines 9

Key concepts 11

The role of AI in parliaments 12

Risks and challenges for parliaments 17
Alignment with national and international AI frameworks and standards 22
Inter-parliamentary cooperation for AI 25

Strategy 29

Strategic actions towards AI governance 30

Strategic actions towards AI governance: Policy and structure 34
Strategic actions towards AI governance: Strategy and innovation 37
Strategic actions towards AI governance: Stakeholder engagement 41
Strategic actions towards AI governance: Find out more 43
Generic risks and biases 45
Generic risks and biases: Categories of risk 50
Generic risks and biases: Cognitive bias types 53
Generic risks and biases: Data bias types 57
Generic risks and biases: Processing and validation bias types 60
D
Ethical principles 63
Ethical principles: Privacy 68
Ethical principles: Transparency 71
Ethical principles: Accountability 74
Ethical principles: Fairness and non-discrimination 76
Ethical principles: Robustness and safety 79
Ethical principles: Human autonomy and oversight 82
Ethical principles: Societal and environmental well-being 86
Ethical principles: Intellectual property 88
Introducing AI applications 90
Training for data literacy and AI literacy 93
Training for data literacy and AI literacy: Data literacy in an AI context 97
Training for data literacy and AI literacy: Developing AI literacy 100
Planning and implementation 103

Project portfolio management 104

Project portfolio management: The STEP approach 108
Data governance 112
Data governance: Data quality 115
Data governance: Personal data protection 118
Data governance: Parliamentary context 120
Data governance: Data management for AI systems 124

1
Guidelines for AI in parliaments

Security management 128

Security management: Parliamentary context 131
Security management: Threats 133
Security management: Good practices 138
Security management: Implementing cybersecurity controls 142
Risk management 146
Risk management: Risk assessment questionnaires 152
Systems development 155
Systems development: Systems life cycle and development frameworks 158
Systems development: Deployment and implementation 162
Systems development: Deployment patterns 166

Appendices 171

Glossary of terms 172

2
Forewords
I am pleased to present these Guidelines for AI in parliaments, which arrive at a
crucial moment in our democratic journey. We stand at the threshold of a
transformation that is reshaping how parliaments operate and serve their citizens.
Artificial Intelligence presents both extraordinary opportunities and significant
challenges for our institutions of democracy.

These Guidelines emerge from our recognition that parliaments must take a leading
role in governing the use of AI, not only through legislation and oversight but also
through their own adoption and implementation of these technologies. The
Guidelines represent a collaborative effort, drawing on the expertise and experience
of parliamentary staff and technology specialists from across our global community.

The IPU’s commitment to supporting parliaments in their digital transformation

journey has never been more vital. We continue to witness first-hand how
technological advancement is reshaping parliamentary work. These Guidelines build
upon our existing frameworks, including the IPU resolution on “The impact of artificial
intelligence on democracy, human rights and the rule of law” and the IPU Charter on
Ethics of Science and Technology (October 2024), as well as the Guide to Digital
Transformation in Parliaments and the World e-Parliament Reports developed
through our Centre for Innovation in Parliament.

The IPU resolution highlights both the tremendous potential of AI to enhance

D
parliamentary functions and the critical importance of ensuring its responsible
deployment, particularly in protecting our most vulnerable citizens. These Guidelines
therefore emphasise the fundamental principles of transparency, accountability and
fairness that must underpin any technological advancement in our institutions of
democracy, as well as more technical aspects.

The Guidelines offer a comprehensive framework for parliaments at all stages of

their AI journey, whether they are just beginning to explore AI’s potential or are
already implementing advanced applications. They provide practical guidance while
emphasising the importance of strong governance, ethical considerations and risk
management. Most importantly, they stress that AI should augment and enhance
human capability rather than replace it, particularly in the context of democratic
deliberation and decision-making.

These Guidelines are valuable for different audiences inside and outside of
parliament, including for members, especially those serving on modernization
committees, technology committees or committees of the future, as well as senior
managers and technical experts. They offer useful insights for parliamentarians as
they grapple with AI oversight and regulation in society, showing how AI can be
deployed responsibly within their own institutions. They provide a framework for
making informed decisions about AI adoption while ensuring robust democratic
oversight.

3
Guidelines for AI in parliaments

Working together, we can ensure that AI serves to strengthen rather than diminish
our democratic institutions, upholding the fundamental values that our parliaments
represent.

Martin Chungong
Secretary General
Inter-Parliamentary Union

4
Guidelines for AI in parliaments

Since the Brazilian Chamber of Deputies’ first experience with artificial intelligence
(AI) in 2013, we have been on a journey of continuous learning about this technology
and the extraordinary capabilities that it can offer to parliaments. Despite this, we are
still surprised by the exponential speed of advances in AI, and by its ubiquitous
nature in the daily lives of public organizations around the world.

Among the lessons learned over the years of using AI in the legislative branch, it is
worth highlighting the need for coordination encompassing multiple stakeholders, so
that the technology’s use can be well planned and managed. This observation
inspired us to hold the Parliamentary Data Science Hub meeting in Brasília (in April
2024) to discuss good practices for the use and development of artificial intelligence
in parliaments.

I am proud to note that the excellent work of the experts who participated in that
meeting has resulted in a set of guidelines with a user-friendly approach and the
flexibility needed to meet the different realities faced by parliaments. The guidelines
combine strategic actions and policies with examples of everyday practices, in a way
that recognizes the plurality of decisions involved in the use, development and
outsourcing of AI systems.

In addition to being an everyday reference document for IT professionals, the

Guidelines for AI in Parliaments are a reference for parliamentary leaders, directors
general, secretaries general and managers of parliaments, in defining strategies and
priorities relating to this topic. I hope that these guidelines will be instrumental in
reducing the technological gap between parliaments worldwide, so that they can
keep up with the advances that society expects.
D
Celso de Barros Correia Neto
Director General
Chamber of Deputies of Brazil

Coordinator of the Parliamentary Data Science Hub in the IPU’s Centre for
Innovation in Parliament

5
Introduction
Artificial intelligence (AI) presents significant opportunities for parliaments to
enhance their operations and to become more efficient and effective, enabling them
to better serve citizens. However, adopting AI introduces new challenges and
presents risks that must be carefully managed.

These Guidelines for AI in parliaments (the “Guidelines”) have been developed for
parliaments by parliamentary staff and the Inter-Parliamentary Union’s (IPU) Centre
for Innovation in Parliament (CIP). They provide comprehensive guidance to support
parliaments on their journey towards understanding and implementing AI responsibly
and effectively. By adopting a well-thought-through, strategic approach to AI,
parliaments can harness the technology’s full potential to drive innovation and
efficiency in the legislative process.

The Guidelines cover key areas, including the potential role of AI in parliaments,
related risks and challenges, suggested governance structures and AI strategy,
ethical principles and risk management, training and capacity-building, and how to
manage a portfolio of AI projects across parliament. They are complemented by a
set of use cases, shared by parliaments, that describe how AI can support specific
parliamentary actions.

The Guidelines stress the importance of a measured, risk-based approach to AI

adoption. Key recommendations include the following:
D
● Start with small pilot projects to build experience.
● Focus on use cases with clear benefits and manageable risks.
● Ensure robust human oversight of AI systems.
● Prioritize transparency and accountability.
● Invest in data and AI literacy across the organization.
● Engage with diverse stakeholders throughout the process.

Audience
The Guidelines have been written to support a range of parliamentary roles:

• For members of parliament (MPs), the Guidelines offer insights into the
potential impact of AI on legislative processes, constituent engagement and
parliamentary oversight. They provide a clear overview of AI capabilities and
limitations, helping MPs to make informed decisions about AI adoption and
regulation in parliament.

For members serving on modernisation committees, committees of the future

or similar bodies focused on technological advancements, the Guidelines
provide strategic insights into AI governance and implementation. This is
particularly valuable as parliamentarians increasingly face decisions about AI
deployment within their own institutions while simultaneously developing
legislation to govern AI use in society more broadly.

6
Guidelines for AI in parliaments

• For senior parliamentary managers, the Guidelines provide a high-level

overview, offering strategic advice on developing AI governance frameworks,
policies and oversight mechanisms. This includes establishing clear roles and
responsibilities, creating codes of ethics, and aligning AI initiatives with
parliament’s organizational goals.

• For staff involved in AI implementation, the Guidelines provide detailed

guidance on identifying use cases, managing projects, addressing technical
challenges, upholding ethical standards and managing risks throughout the AI
life cycle.

The Guidelines are designed to support parliaments of all sizes and levels of digital
maturity – from large, well-resourced legislatures with advanced digital
infrastructures, to smaller parliaments just beginning their digital transformation
journey.

The Guidelines can be tailored to allow parliaments to focus on areas most relevant
to their current needs and capabilities, and individual parliaments can adapt them to
suit their unique circumstances, culture and resources. While digitally mature
parliaments may be ready to implement more advanced AI applications, those at
earlier stages can use the Guidelines to build foundational governance structures
and develop AI literacy.

List of Guidelines
D
Audience
For senior For staff
parliamentary involved in AI
Guideline managers For MPs implementation
Key concepts The role of AI in
parliaments
Risks and challenges
for parliaments
Alignment with national
and international AI
frameworks and
standards
Inter-parliamentary
cooperation for AI
Strategy Strategic actions
towards AI governance
Generic risks and
biases
Ethical principles
Introducing AI
applications
Data and AI literacy
Planning and Project portfolio
implementation management

7
Guidelines for AI in parliaments

Data governance
Security management
Risk management
Systems development

8
Guidelines for AI in parliaments

About the Guidelines

Background
In April 2024, the Chamber of Deputies of Brazil – as the host of the CIP’s
Parliamentary Data Science Hub – convened, in Brasilia, a panel of experts from
parliaments and the IPU to develop an outline of what would become the Guidelines
for AI in parliaments. The intention was to develop comprehensive, flexible, scalable
guidance to support parliaments as they embrace and responsibly engage with the
rapidly emerging technologies associated with AI, whatever their level of digital
maturity.

Acknowledgements
The CIP, the Parliamentary Data Science Hub, the IT Governance Hub and the
editors are grateful for the contributions of the many parliamentary staff who helped
shape, write and review these Guidelines, and for the support provided by the
respective parliaments.
Editors
● Patricia Gomes Rêgo de Almeida, Chamber of Deputies of Brazil
● Ludovic Delépine, European Parliament
● Andy Williamson, Centre for Innovation in Parliament, Inter-Parliamentary
Union
Contributors
D
• Patricia Rêgo de Almeida, Chamber of Deputies of Brazil
• Francisco Edmundo Andrade, Chamber of Deputies of Brazil
• Javier de Andrés Blasco, Chamber of Deputies of Spain
• Álvaro Carmo, National Assembly of Angola
• Virginia Carmona, Chamber of Deputies of Chile
• Giovanni Ciccone, Chamber of Deputies of Italy
• Ludovic Delépine, European Parliament
• Claudia di Andrea, Chamber of Deputies of Italy
• Michael Evraire, House of Commons of Canada
• Marcio Fonseca, Chamber of Deputies of Brazil
• José Andrés Jiménez Martín, Chamber of Deputies of Spain
• Vinicius de Morais, Chamber of Deputies of Brazil
• Rune Mortensen, Parliament of Norway
• Neemias Muachendo, National Assembly of Angola
• Jurgens Pieterse, Parliament of South Africa
• Manuel Pereira González, Senate of Spain
• Peter Reichstädter, Parliament of Austria
• Frode Rein, Parliament of Norway
• Esteban Sanchez, Chamber of Deputies of Chile
• Luciana Silo, Chamber of Deputies of Italy
• Paul Vaillancourt, House of Commons of Canada

9
Guidelines for AI in parliaments

• Kim van Dooren, Senate of Netherlands

• Marieke van Santen, Senate of Netherlands
• Rodolfo Vaz, Chamber of Deputies of Brazil
• Ricardo Vilarins, Chamber of Deputies of Brazil
• Andy Williamson, Centre for Innovation in Parliament, Inter-Parliamentary
Union
Reviewers
• Patricia Rêgo de Almeida, Chamber of Deputies of Brazil
• Avinash Bikha, Centre for Innovation in Parliament, Inter-Parliamentary Union
Ludovic Delépine, European Parliament
• Andy Richardson, Centre for Innovation in Parliament, Inter-Parliamentary
Union
• Andy Williamson, Centre for Innovation in Parliament, Inter-Parliamentary
Union
Graphic design
• Ana Paulla Diniz, Chamber of Deputies of Brazil
• Kelly Lima Cardoso, Chamber of Deputies of Brazil

About the Centre for Innovation in Parliament

The CIP was established in late 2018 to support parliaments in their digital
transformation efforts. It facilitates knowledge-sharing and collaboration through a
network of regional and thematic hubs, each hosted by a parliament or partner
organization. The CIP has made significant contributions to parliamentary
modernization by undertaking research and providing guidance on digital strategies,
D
by organizing capacity-building activities, and by developing resources on topics
such as open data, cybersecurity and emerging technologies, including AI. Through
its work, including the World e-Parliament Report series and the Guide to digital
transformation in parliaments, the CIP has become a key platform for parliaments
worldwide to exchange good practices and innovative solutions, fostering more
efficient, transparent and accessible legislative institutions.

Contact
For more information about this work, please contact [email protected]. We are
always keen to learn how the Guidelines have been used. We welcome all feedback
and suggestions.

The Guidelines for AI in parliaments are published under the Creative Commons Attribution-
NonCommercial-ShareAlike 4.0 International licence. Please see the IPU’s terms of use for more
details.

Suggested attribution: Inter-Parliamentary Union (IPU), Guidelines for AI in parliaments (Geneva: IPU,
2024): www.ipu.org/AIguidelines.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the.
It may be freely shared and reused with acknowledgement of the IPU. For more information about the IPU’s work on artificial
intelligence, please visit www.ipu.org/AI or contact [email protected].

10
Guidelines for AI in parliaments

Key concepts
D

11
Guidelines for AI in parliaments

The role of AI in
parliaments
Audience
This high-level guideline is intended for parliamentary leadership and senior
parliamentary managers, as well as for parliamentary staff and MPs who are
interested in gaining a broad understanding of where AI can impact upon the work of
parliaments.

About this guideline

This guideline offers an overview of parliamentary functions and suggests potential
applications for AI within these functions. It explores how AI can transform various
aspects of parliamentary work, from streamlining administrative tasks to enhancing
legislative research and improving public engagement. It looks at how parliaments
can leverage AI effectively, while emphasizing the importance of upholding
democratic principles and values throughout the implementation process.

Potential uses for AI

There are opportunities to implement AI-based systems in many areas of parliament.
AI can support administrative, legislative and public engagement processes, as well
as parliamentary transparency efforts. While a small number of parliaments are
already developing AI-based applications in some of the areas below, their usage in
many other areas remains unexplored at this stage, or is only being tested on a pilot
or prototype basis.

For examples of potential applications of AI in parliaments, refer to the

accompanying use cases, which are intended to help parliaments develop
their own proposals.

The first step in the journey is to understand where opportunities lie for a given
parliament. It is important to recognize that every parliament is different and will want
to seize the opportunities on offer differently, based on its own unique culture,
working practices, resource availability and timing constraints. A framework to help
identify opportunities and support the adoption of AI can be found in the following
guidelines: Project portfolio management and Strategic actions towards AI
governance.

12
Guidelines for AI in parliaments

Improving legislative research and analysis

AI systems can assist parliamentary staff in conducting comprehensive legislative
research and analysis. Machine-learning algorithms can analyse vast volumes of
legislative documents, identifying patterns, trends and relevant ideas. Additionally,
AI-driven data analysis platforms can facilitate evidence-based policy formulation by
synthesizing disparate sources of information and highlighting key findings for
decision makers.
Analysing legislative trends
AI systems can analyse large volumes of legislative data (live or archived) in order to
identify trends and patterns in proposed and enacted legislation. These insights
enable MPs to better understand legislative priorities, areas of interest and areas
requiring attention. They can also assist in identifying legislative key performance
indicators (KPIs), helping parliamentary committees to determine the impact of
legislation once enacted.
Identifying similarities and differences in legislation
Machine-learning algorithms can compare and contrast proposed draft legislation
with legislation that already exists. Identifying similarities, disparities and areas of
overlap (or contradiction) in this way can help to avoid conflicts between laws and
ensure legislative coherence. This comparison also makes it possible to determine if
two similar laws could have different impacts, and to identify the reasons for this.

Supporting the legislative function

AI systems can be used to assist the legislative work of parliament, supporting tasks
such as assessing the impact of bills, and mapping amendments in order to
understand their impact on a bill.
Analysing the impact of proposed legislation
AI systems can be used to analyse the potential impact of proposed legislation,
helping MPs to make informed decisions about the viability and implications of a bill
and to identify any unexpected consequences of the proposal.
Mapping bills and amendments
AI systems can be used to analyse amendments and voting to ensure that they are
correctly matched up to bills, and that the changes (and the impact of those
changes) are transparent and can be easily understood.
Assisting with post-legislative scrutiny
AI systems can be used to gather comprehensive data and provide a detailed
analysis of the impact of legislation. This analysis enables MPs to understand
whether a piece of legislation has met its aims and objectives, and to assess
whether (and how) it might need revising.

13
Guidelines for AI in parliaments

Identifying stakeholders, experts and resources

AI systems can be used to analyse databases and online sources to identify experts,
key stakeholders, research reports and other relevant resources related to specific
legislative topics, ensuring that MPs’ legislative work is supported by up-to-date and
reliable information.
Analysing public submissions
The digital parliament is opening up opportunities for more people to submit their
views to parliamentary committees and inquiries. While this is a positive
development for democracy, it increases pressure on parliamentary staff, who have
to manage and make sense of ever-larger volumes of qualitative data. AI systems
can be used to assist with this task, and can be highly effective at identifying themes
and grouping submissions.

Automating administrative tasks

AI technologies such as natural language processing (NLP) and robotic process
automation (RPA) can automate routine administrative tasks such as scheduling
meetings, drafting agendas and managing documentation. Offloading these
repetitive tasks to AI systems allows more time and resources to be allocated to
high-value activities, thus increasing productivity and efficiency.
Scheduling parliamentary meetings and sessions
AI systems can be used to analyse MPs’ schedules, identify available slots, and
automatically schedule meetings and parliamentary sessions, taking into account the
availability of participants and meeting rooms and, therefore, avoiding coordination
issues and inefficiencies.
Managing documents
Parliaments generate and handle large amounts of documents, such as bills,
committee reports, communications and session minutes. AI systems can be used to
automatically organize, label and archive these documents, facilitating quick search
and retrieval when needed.
Automating translation
In multilingual parliaments, automatic document translation can be an invaluable
tool. AI systems can be used to automatically translate legislative documents,
allowing MPs to access information in their preferred language.
Managing digital communication
MPs receive large volumes of emails and other electronic communications from their
constituents and colleagues. AI systems can be used to automatically classify these
messages, identifying the most urgent or relevant ones and assigning them to the
relevant person for processing.
Generating reports and analysing data
Report generation and data analysis are key supporting tasks in parliaments. AI
systems can be used to collect and analyse relevant data, using data visualization

14
Guidelines for AI in parliaments

and predictive models to generate detailed and easily understandable reports for
MPs.

Improving transparency
AI systems can play a pivotal role in promoting transparency and accountability
within parliaments. Automated transcription and translation tools can generate
accurate and timely transcripts of parliamentary debates and discussions, making
legislative proceedings more accessible to the public, while AI-powered sentiment
analysis tools can gauge public sentiment towards legislative proposals, enabling
MPs to better understand and address their constituents’ concerns.
Automating transcription of parliamentary debates
AI systems can be used to automatically transcribe parliamentary debates in real
time. These accurate, rapidly produced transcripts can then be made available to the
public and to specific users or departments within parliament, allowing citizens and
key officials to access parliamentary proceedings without having to consult complete
audiovisual recordings. Real-time speech-to-text transcription and translation – a
possibility offered by some language models – could also facilitate effective
communication in the context of large multinational events.
Visualizing legislative data
AI systems can be used to create interactive visualizations of legislative data, such
as an MP’s legislative activity or the progress of a bill through parliament. These
visualizations make it easier for the public to understand and evaluate the work of
parliament.
Accessing legislative information
AI-powered search tools can allow constituents to easily find information about bills,
votes, committees and other aspects of parliamentary work. This promotes
transparency by making legislative information more accessible and understandable
to all.
Analysing economic data
AI systems can be used to analyse economic data related to parliamentary spending
and the financial interests of MPs, helping to identify potential conflicts of interest,
and promoting transparency and accountability by ensuring that MPs are subject to
public scrutiny.
Producing plain-language summaries
AI systems can be used to summarize bills, reports and transcripts in plain language,
making them easier to understand for ordinary citizens. Making such summaries
available can enhance public participation in the legislative process and foster
communication between MPs and their constituents.

Enhancing public engagement

AI-powered chatbots, sentiment analysis tools and multimedia content production
systems can enhance public engagement with parliament by facilitating
communication between citizens and MPs, offering insights into public opinion and
helping to create accessible content.

15
Guidelines for AI in parliaments

Helping citizens connect with parliament

AI-powered chatbots and virtual assistants can enhance public engagement by
providing fast, personalized responses to enquiries, facilitating communication
between MPs and constituents, and disseminating information about parliamentary
procedures and initiatives. These AI systems empower constituents to participate
actively in the democratic process while relieving parliamentary staff of the burden of
managing large volumes of enquiries manually.
Analysing sentiment in public spaces
AI-powered sentiment analysis tools can be used to monitor social media and other
online platforms in order to assess public sentiment towards legislative topics, as
well as towards MPs and their legislative decisions. These insights can help MPs
understand their constituents’ concerns and opinions, allowing them to better reflect
these positions in parliament.
Producing multimedia content
AI systems can be used to automatically produce short video summaries that can
then be posted on social media. These clips would typically focus on the most
important part(s) of a speech or other intervention, with multilingual subtitles
provided.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

16
Guidelines for AI in parliaments

Risks and challenges for

parliaments
Audience
This high-level guideline is intended for parliamentary leadership and senior
parliamentary managers, as well as for parliamentary staff and MPs who are
interested in gaining a broad understanding of the risks and challenges of AI.

About this guideline

This guideline examines the risks involved in the introduction of AI in parliaments,
both from a strategic level and in terms of what it means operationally for
parliaments, with a particular focus on risks that are unique to legislatures. For each
of these areas, it provides brief checklists that senior parliamentary managers can
use as a starting point to fully understand the impact that AI is likely to have.

As AI adoption gains traction and the use of this technology becomes more
commonplace, parliaments must understand the implications and closely examine
the risks and challenges associated with the implementation of AI.

Technologies such as generative AI, with its ability to create content based on vast
amounts of data, promises productivity gains and potentially transformational change
in parliamentary operations. However, it also introduces new complexities and risks
that must be carefully managed.

For a discussion of the more generic risks and biases that AI introduces, refer to the
guideline Generic risks and biases. For a discussion of the potential uses of AI in
parliamentary settings, refer to the guideline The role of AI in parliaments.

Strategic considerations
At the strategic level, parliaments face several key challenges in adopting AI.
Foremost among these is the development of comprehensive AI governance
frameworks that ensure strong ethical principles, transparency and accountability in
AI systems.

Parliaments must also address potential biases in AI algorithms, ensuring that these
systems do not inadvertently amplify existing societal inequalities or underrepresent

17
Guidelines for AI in parliaments

minority views. This is particularly crucial in parliamentary contexts, where fair

representation is a fundamental principle.

Public trust and perception present another strategic challenge. Parliaments must
effectively communicate their use of AI to constituents, managing expectations and
addressing concerns about the role of AI in democratic processes. This requires a
delicate balance between showcasing the benefits of AI adoption and reassuring the
public that human judgement remains central to parliamentary functions.
Strategic considerations checklist:
● Develop a comprehensive AI governance framework and policies that reflect
parliament’s ethical principles.
● Establish protocols for ensuring AI transparency and accountability.
● Expand the remit of existing data committees or similar bodies to encompass
AI.
● Develop a communication strategy to inform the public about AI use in
parliament.
● Regularly assess and mitigate potential biases in AI systems.

Unique parliamentary considerations

Several aspects of AI adoption are uniquely relevant to parliaments, with the rapid
evolution of this technology requiring legislatures to develop flexible, future-proof AI
strategies. This is particularly challenging given the typically slower pace of change
in parliamentary institutions.

Unlike in other contexts, the successful implementation of AI in parliament requires

buy-in across political divides. Different parties may have varying views on the role
and extent of AI use in parliamentary functions, necessitating careful negotiation and
compromise.

Perhaps most fundamentally, parliaments must balance the pursuit of efficiency

through AI with the preservation of core democratic values. While AI can enhance
many aspects of parliamentary work, it is crucial that it is used to augment – not
replace – the essential human elements of democratic representation and decision-
making.

The adoption of AI in parliaments has direct implications for legislative processes.

There is the potential for AI to significantly alter how debates are conducted and how
legislation is drafted. While AI can provide valuable insights and efficiencies, it is
essential to maintain human oversight in any move towards AI-assisted lawmaking.
The nuanced and often politically sensitive work of parliaments requires a level of
judgement and ethical consideration that current AI systems cannot replicate.

Constituency engagement is another area where AI could have a profound impact.

AI-powered tools could support deeper analysis of public sentiment on proposed
legislation, potentially providing more real-time feedback. However, this must be
balanced with the importance of direct constituent interactions to ensure that the
human element of representation is not lost.

18
Guidelines for AI in parliaments

Unique parliamentary considerations checklist:

● Develop guidelines for maintaining human oversight in AI-assisted lawmaking
and ensuring scrutiny of AI-based decisions.
● Establish rules for disclosing AI use in legislative processes.
● Develop a flexible AI strategy that can adapt with rapid technological change.
● Establish mechanisms for equitable access to AI resources across all of
parliament.
● Develop protocols for international cooperation on AI in parliaments.

Operational challenges
On the operational front, implementing and integrating AI systems into existing
parliamentary procedures and processes poses significant challenges, especially
since these are often complex and steeped in tradition. Moreover, parliaments must
ensure that AI adoption does not disrupt the essential human elements of political
discourse and decision-making.

Data management and security are also critical concerns. Parliaments handle
sensitive information and are prime targets for cyberattacks. AI systems could
potentially create new vulnerabilities if they are not implemented with robust security
measures.

Capacity-building and change management present another set of operational

challenges. Developing AI literacy and data literacy among MPs and staff is crucial
for effective use and oversight of these systems. However, this introduces a unique
challenge in balancing the need for traditional parliamentary skills and knowledge
with new AI competencies. Moreover, ensuring that AI systems are trained on high-
quality data is essential to prevent biased or inaccurate outputs.

There is also the potential for job displacement within parliamentary staff,
necessitating careful management of role redefinition and retraining.
Operational challenges checklist:
● Conduct a thorough assessment of existing parliamentary procedures for AI
integration.
● Implement robust cybersecurity measures for AI systems.
● Develop and implement AI literacy and data literacy programmes for staff and
MPs.
● Create a data quality assurance process for AI training data sets.
● Establish a change management plan to address potential job displacement
and role changes.

Mitigation strategies
Given these challenges, a cautious and measured approach to AI adoption is
advisable for most parliaments. The IPU’s Centre for Innovation in Parliament
recommends a step-by-step, risk-based approach.

Creating safe “lab environments” for AI experimentation is a prudent first step. This
allows parliaments to explore potential use cases, such as producing summaries of

19
Guidelines for AI in parliaments

texts or creating records of debates, without risking core parliamentary functions.

However, parliaments should be extremely cautious about introducing AI into core
legislative systems at this stage.

Maintaining human scrutiny and control is paramount. Any AI-generated outputs

must be explainable and subject to expert validation, and the entire system must be
auditable. This is essential not only for ensuring accuracy but also for maintaining
public trust in parliamentary processes.

Collaboration with other parliaments and external experts can be invaluable in

building capacity and sharing best practices. The complexity of AI systems means
that individual parliaments working alone may initially lack the skills and knowledge
to implement AI-based systems safely and effectively.

AI policies and practices should be regularly reviewed and updated to account for
the rapid pace of technological change. Parliaments must remain agile, continuously
assessing the impact of AI on their operations and adjusting their approaches
accordingly.
Mitigation strategies checklist:
● Create a safe environment for AI experimentation.
● Implement a step-by-step, risk-based approach to AI adoption and monitoring.
● Adopt and, where necessary, adapt these Guidelines to support the safe
introduction of AI in parliament according to its organizational culture.
● Develop partnerships with other parliaments and external experts for
knowledge-sharing.

Conclusion
The adoption of AI in parliaments offers significant potential benefits but also
presents unique challenges. By carefully navigating the strategic, operational and
legislative risks, parliaments can harness the power of AI to enhance their
effectiveness while safeguarding the essential human elements of democratic
representation.

Parliaments have a dual responsibility as both users and regulators of AI technology.

They must lead by example in the responsible adoption of AI within their own
institutions while also shaping the legislative frameworks that will govern AI use in
broader society.

Parliaments therefore have an opportunity to set a standard for responsible AI use

that could inform its adoption in other areas of government and in society at large.
This positions parliaments at the forefront of defining how AI can be leveraged to
strengthen – rather than undermine – democratic processes in the digital age.

The journey of AI adoption in parliaments is just beginning, and the path forward will
require ongoing dialogue, rigorous oversight and a commitment to preserving the
fundamental values of democratic governance.

20
Guidelines for AI in parliaments

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

21
Guidelines for AI in parliaments

Alignment with national

and international AI
frameworks and
standards
Audience
This high-level guideline is intended for parliamentary leadership and senior
parliamentary managers, as well as for parliamentary staff and MPs. It may also be
useful for technical staff involved in the implementation of AI-based systems.

About this guideline

As parliaments consider adopting AI technologies, it is important to be aware of and,
where appropriate, to adhere to government standards and relevant national and
international frameworks for AI use. These standards and frameworks may
significantly influence how parliaments implement AI and, in some cases, may
require parliaments to adapt the approaches outlined in these Guidelines.

Compliance with AI regulations

Parliaments must take steps to ensure that their use of AI conforms to national laws,
regulations and, where possible and appropriate, good practice for AI use in the
public sector.

It should be noted that, while these Guidelines serve as an important starting point,
parliaments may need to go above and beyond these in order to address their
unique needs and uphold democratic principles. They should be used in conjunction
with national and international standards to create a comprehensive approach to AI
governance in parliament.

As part of its AI governance framework, parliament should stay informed about

national and international AI-related policies and ensure that it remains compliant
with these at all times. For example, the Chamber of Deputies of Italy has developed

22
Guidelines for AI in parliaments

a code of conduct for the use of generative AI, which emphasizes that the use of this
technology must align with various multilevel strategies and regulations:

[The Code of Conduct] for the Use of Generative Artificial Intelligence Tools has therefore
been adopted, taking into account the Principles for the Use of AI in Support of
Parliamentary Work, as laid down by the Supervisory Committee on Documentation
Activities of the Chamber of Deputies, and having regard to the recommendations set
forth in the 2024–2026 Three-Year Plan for ICT in the Public Administration, the
Hiroshima Process International Code of Conduct for Advanced AI Systems as agreed
by the G7, as well as the Guidelines for Secure AI System Development, promoted at
the international level by the National Cyber Security Centre and signed on 27 November
2023 by the National Cybersecurity Agency.

In order to ensure compliance with applicable regulations, parliaments are advised to

take the following steps:

● Research and identify relevant national and international AI frameworks or

standards.
● Consult with government bodies responsible for AI oversight.
● Assess how these standards impact parliamentary AI implementation.
● Adjust internal AI policies and practices to ensure compliance.
● Regularly monitor national and international AI frameworks for updates.

Laws and regulations impacting upon the use of AI

In addition to AI-specific regulations, it is important to consider the implications of AI
systems in terms of other laws and regulations that might be in place. These may
cover the following matters, among others:

● Access to information and/or freedom of information

● Cybersecurity
● Data protection and privacy
● Discrimination and equality
● Employment
● Human rights and accessibility
● Intellectual property
● Procurement and competition

International standards and frameworks for AI

Parliaments may find the following international standards and frameworks helpful
when defining their own governance processes, rules and regulations regarding the
use of AI:

● OECD Principles on Artificial Intelligence: principles promoting innovative and

trustworthy AI that respects human rights and democratic values
● UNESCO Recommendation on the Ethics of Artificial Intelligence: the first
global standard-setting instrument on the ethics of AI
● Ethically Aligned Design: A Vision for Prioritizing Human Well-being with
Autonomous and Intelligent Systems: guidance for ethical AI produced as part

23
Guidelines for AI in parliaments

of the Institute of Electrical and Electronics Engineers (IEEE) Global Initiative

on Ethics of Autonomous and Intelligent Systems
● Standards by ISO/IEC JTC 1/SC 42: a list of standards developed by the
International Organization for Standardization (ISO) and International
Electrotechnical Commission (IEC) joint technical committee on AI, including
a data quality governance framework
● Council of Europe Framework Convention on Artificial Intelligence and Human
Rights, Democracy and the Rule of Law: a framework convention developed
by the Council of Europe’s Committee on Artificial Intelligence
● White Paper on Artificial Intelligence: A European approach to excellence and
trust: a European Commission document that, while specific to the European
Union, provides a framework for trustworthy AI that could be informative for
parliaments globally
● G7 Hiroshima Process on Generative Artificial Intelligence (AI): Towards a G7
Common Understanding on Generative AI: an OECD report that helped
inform and structure discussions around the G7 Hiroshima Process
● AI Governance: A Holistic Approach to Implement Ethics into AI: a framework
providing guidance on implementing ethical AI in practice
● Risks, Harms and Benefits Assessment Tool: a United Nations Global Pulse
tool to help assess the risks and benefits of using AI in development and
humanitarian contexts
● ITU/WHO Focus Group on Artificial Intelligence for Health (FG-AI4H): a joint
International Telecommunication Union (ITU) and World Health Organization
(WHO) focus group that has produced guidance and other outputs on AI
governance in a critical public-sector domain

Parliaments should regularly check for updates to these frameworks and for new
parliament-specific AI guidance documents as they emerge. The unique role of
parliaments in democratic societies may necessitate the development of more
tailored international guidance in the future.

National government frameworks and strategies for AI

For a list of national government frameworks, strategies and programmes relating to
AI, refer to the sub-guideline Strategic actions towards AI governance: Find out
more, which is part of the guideline Strategic actions towards AI governance.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

24
Guidelines for AI in Parliaments

Inter-parliamentary
cooperation for AI
Audience
This high-level guideline is intended for parliamentary leadership and senior
parliamentary managers, as well as for senior IT staff who are interested in the
adoption and application of AI in their parliament.

About this guideline

This guideline explores how parliaments can work together to support good practice
and the ethical use of AI at a time when legislatures are increasingly embracing the
opportunities and addressing the challenges presented by this technology. It looks at
various forms of parliamentary cooperation and collaboration, providing guidance
and recommendations for parliaments in this area.

Establishing inter-parliamentary AI networks

Every parliament will likely apply these Guidelines in unique ways, adapting them to
its specific context, culture and needs. However, the universal nature of many AI
challenges presents a compelling case for inter-parliamentary cooperation.
Collaboration is encouraged at all levels, whether this is global, regional or
subregional.

The foundation of effective collaboration lies in robust networks. Parliaments should

create formal AI knowledge-sharing networks for continuous exchange. Networks
such as the IPU’s Centre for Innovation in Parliament (CIP) already exist as vehicles
for these conversations, where parliamentary staff can discuss the latest
developments and challenges in AI governance.

Sharing use cases, case studies and good practices

Parliaments are encouraged to document successful AI implementations, analyse
challenges and share lessons learned. A centralized repository of use cases, case
studies and solutions that is accessible to all parliaments – such as the use cases
that accompany these Guidelines – can serve as a valuable resource for parliaments
at various stages of AI adoption.

25
Guidelines for AI in parliaments

Collaborative AI project development

Joint research initiatives and shared pilot projects offer an opportunity for
parliaments to pool resources and expertise, and to learn together. Parliaments
could coordinate efforts, focusing on developing open-source AI tools specifically
designed for parliamentary use. For instance, a collaborative project to create an AI-
powered legislative drafting assistant could benefit many parliaments, including
those that might not have the resources to undertake such an endeavour on their
own.

Harmonizing AI governance frameworks

While every parliament operates within its unique context, there is value in aligning
internal approaches to AI governance. Parliaments have already collaborated to
create these Guidelines, and this process could be extended to grow the Guidelines
and keep them up-to-date and relevant. Co-created resources such as these will
provide a solid foundation for parliaments to build upon, adapting them to local
needs while ensuring a baseline of good practice.

Capacity-building and training

The rapid evolution of AI necessitates ongoing training and capacity-building.
Parliaments can create efficiencies by jointly developing training programmes for
members and staff. They might even consider developing a certification in AI
governance for parliaments. Exchanges of AI experts and specialists between
parliaments will foster cross-pollination of ideas and expertise.

Data-sharing and standardization

Parliaments can jointly establish protocols for secure data-sharing and develop
common data standards for AI applications, following the example of Akoma Ntosa,
which was developed as an international standard for parliamentary data. This could
include collaborative efforts to improve data quality and consistency across
parliaments, enhancing the potential for AI-driven insights.

Ethical AI evaluation and auditing

Parliaments can work together to develop shared frameworks for AI system audits,
facilitating peer-review processes for AI implementations. Together, they can
develop common metrics for measuring AI impact and effectiveness, ensuring that AI
serves the needs of democratic institutions and citizens.

Addressing common challenges

Parliaments could coordinate on collaborative approaches to common issues such
as mitigating biases, ensuring AI transparency or managing AI-related privacy
concerns. By pooling knowledge and resources, parliaments could develop more
effective solutions to these shared challenges.

26
Guidelines for AI in parliaments

Engaging with international initiatives

As AI governance becomes a global concern, parliaments should engage with
international initiatives to ensure their interests are represented in global forums. By
taking these steps, parliaments can facilitate collaboration on AI-related legislation
and regulation, thus helping to ensure that democratic principles are upheld in the
global AI landscape.

Future-proofing AI governance
Parliaments should be involved in collaborative foresight and scenario planning
exercises, helping to prepare for potential future developments in AI. This could
include joint research on emerging AI technologies and their implications for
parliamentary work, as well as the development of adaptive governance frameworks
that can evolve alongside AI technology.

The role of the IPU’s Centre for Innovation in Parliament

Throughout these collaborative efforts, the CIP acts as a crucial facilitator and
coordinator, uniquely positioned to perform the following roles:

● Acting as a central hub for the sharing of knowledge and good practices, and
supporting regional networks
● Providing a neutral platform for discussing challenges and developing
solutions
● Representing parliamentary interests in global AI governance discussions
● Fostering a community of practice among parliaments, and encouraging
ongoing dialogue and collaboration
● Offering expertise and resources to support parliaments at various stages of
AI adoption

By leveraging its network and resources, the CIP plays a pivotal role in ensuring that
parliaments worldwide are well-equipped to harness the benefits of AI while
mitigating its risks.

Conclusion
As AI continues to transform parliamentary work, collaboration becomes not just
beneficial, but essential. Through shared efforts in areas such as exchanging
knowledge, developing projects and governance frameworks, and addressing
common challenges, parliaments can navigate the complex landscape of AI more
effectively.

The path forward is clear: through collaboration and coordination, parliaments can
lead the way in ethical and effective AI governance, setting a standard for
responsible AI use that extends to other sectors.
Actions
● Join the CIP’s thematic and regional networks to get peer support.
● Document and share AI use cases and implementation case studies.

27
Guidelines for AI in parliaments

● Collaborate on developing common AI governance frameworks, training

programmes and data standards for parliaments.
● Engage in international AI initiatives and forums in order to represent
parliamentary interests in global AI governance discussions.
● Participate in or initiate joint AI projects, focusing on open-source tools
specific to parliamentary needs.
● Conduct collaborative foresight exercises to prepare for future AI
developments and their implications for parliamentary work.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

28
Guidelines for AI in parliaments

Strategy

29
Guidelines for AI in parliaments

Strategic actions towards

AI governance
Audience
This high-level guideline is intended for senior parliamentary managers, including
those working at a strategic or board management level, and in particular those who
have, or will have, a role in governing the use of AI.

About this guideline

This guideline outlines key actions for effective AI governance in parliaments. It
focuses on creating an ethical framework, establishing robust governance structures
and developing comprehensive policies for the use of AI-based technologies. It
emphasizes stakeholder engagement, strategic alignment with parliamentary goals,
and capacity-building.

This guideline promotes responsible AI innovation, balancing technological

advancement with ethical use to deliver benefits including improved data reliability
and risk management, and more effective AI implementations.

Individual parliaments should adapt the recommendations contained in this guideline

to suit their unique circumstances and resources, using them as a foundation for
developing tailored guidance aligned with their existing strategies and methods.

Why strategic AI governance matters

The journey towards effective AI governance begins with an understanding of its
importance. AI governance is about more than simply managing new technology: it is
about creating a framework that maximizes the benefits of AI while minimizing its
risks.

By taking a strategic approach to AI governance, parliaments can build a robust AI

ecosystem that balances operational needs with innovation and ethical
considerations, enhancing data and system reliability, and creating a trustworthy
ecosystem for AI-based digital services.

Strategic governance ensures that AI initiatives align closely with business

requirements, increasing the overall effectiveness of AI implementations. It promotes

30
Guidelines for AI in parliaments

fair and inclusive practices, upholding the democratic principles that parliaments
embody. By encouraging responsible innovation, it positions parliaments at the
forefront of technological advancement in governance.

Moreover, a strategic governance approach improves compliance with still-evolving

legislation governing AI use in the public sector, helping parliaments to stay ahead of
the curve and reducing the risk of future non-compliance.
Key benefits of strategic AI governance:
● Maximizes the benefits to be obtained from AI while better managing risk
● Ensures clear alignment with business processes
● Increases the reliability of data and systems
● Promotes fair practices and responsible innovation
● Increases parliament’s preparedness for evolving regulatory landscapes
● Leaves parliaments well-placed to cooperate and collaborate, sharing good
practices relating to AI

Governance structure and policy

The first step in this journey is to establish a solid governance structure, which
requires a multidisciplinary approach across parliament involving executive boards,
legal departments, business units and the IT function. Some parliaments may
choose to add AI capacity to existing governance boards, while others might create
an AI-specific board. Regardless of the chosen structure, the responsibilities of such
a body are to approve AI policies, monitor strategies, oversee projects and address
ethical issues as they arise.

With this structure in place, the next task is to develop a comprehensive AI

governance policy. This policy should guide all AI-related activities within parliament.
A dedicated working group should lead this effort, defining objectives, outlining
ethical principles and ensuring legal compliance. The policy will clearly delineate
roles and responsibilities for key stakeholders, as well as establish processes for risk
management, information security and data governance.
Actions:
● Establish an AI governance structure and policy
● Define roles, responsibilities and key processes

Ethical and responsible use of AI

Since the adoption of AI is both challenging and filled with potential, adopting these
strategic measures will help parliaments lead the way in responsible AI use. It is
possible to leverage AI’s immense potential while upholding democratic principles
and ethical standards.

As parliaments embark on this journey, they must do so not just for their own benefit,
but also for the benefit of the citizens they serve. By governing AI responsibly, they
can pave the way to a future where technology and democracy work hand in hand,
enhancing governance and improving lives.

31
Guidelines for AI in parliaments

Key points:
● Responsible AI governance benefits both parliaments and citizens.
● Parliaments can lead the way in responsible AI use.

Ethical foundations
At the heart of AI governance lies a robust code of ethics. This code serves as a
declaration of values, guiding the use of AI and the management of the associated
risks. It should reflect parliament’s commitment to privacy, transparency,
accountability, fairness and societal well-being. Importantly, this code must align with
existing laws, regulations and parliamentary procedures. It must be adaptable and
responsive to change. It should also include real-world examples, in order to aid
understanding and adoption and to demonstrate how these ethical principles apply in
practice.
Actions:
● Create an AI code of ethics reflecting parliamentary values.
● Ensure alignment with national laws, standards, international good practice
and parliamentary procedures.

AI strategy and capacity-building

With the foundational elements in place, parliament can now focus on creating a
comprehensive AI strategy. This strategy serves as a road map, aligning AI use with
the institution’s broader goals and objectives. It should include a clear vision, set
measurable goals, outline specific actions and define key performance indicators
(KPIs). The strategy must also address ethical principles, regulatory issues and
infrastructure requirements.

Once the decision to adopt AI has been made, building capacity within parliament
becomes crucial for success. This involves developing a robust plan for staff training
and skills development.

Thinking about how to approach the introduction of AI is also important at this stage.
It is often beneficial to start with small, manageable pilot projects, which help to build
confidence and demonstrate the value of AI in a controlled environment.
Encouraging experimentation and cross-functional collaboration can foster
innovation and drive successful AI initiatives.
Actions:
● Develop a comprehensive AI strategy aligned with parliamentary goals.
● Build internal capacity through training and skills development.
● Look for small pilot projects to build confidence and demonstrate value.

Stakeholder engagement
Even the best governance structure and code of ethics will prove ineffective without
proper stakeholder engagement. Identifying and involving stakeholders from various
levels and departments is crucial. Engaging these stakeholders early and often helps

32
Guidelines for AI in parliaments

to manage risks, identify potential pitfalls and significantly increase the chances of
successful AI implementation.
Action:
● Engage stakeholders from all relevant departments and levels.

Find out more

● For a list of national government guidelines, strategies and programmes
relating to AI, refer to the sub-guideline Strategic actions towards AI
governance: Find out more.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

33
Guidelines for AI in parliaments

Strategic actions towards

AI governance: Policy and
structure
About this sub-guideline
This sub-guideline is part of the guideline Strategic actions towards AI governance.
Refer to the main guideline for context and an overview.

This sub-guideline provides guidance and recommendations for the design and
development of AI governance policies and structures in parliaments.

Background
When a parliament decides to adopt AI-based systems and services, it embarks on a
journey that requires careful planning and a multidisciplinary approach. This journey
begins with the recognition that AI governance is not just an IT issue: it is a matter
that touches every aspect of parliamentary operations.

The first step is to assemble a diverse team. Executive boards, legal departments,
and business and IT units all have crucial roles to play. This team will work together
to create a governance structure that integrates business needs, legal and regulatory
considerations, and technological insights. The exact nature of this structure will vary
from parliament to parliament, reflecting each institution’s unique culture and existing
working methods.

Developing an AI governance structure

When it comes to establishing a governance structure, parliaments have two main
options: either they add AI capacity to an existing board, or they create a dedicated
AI-specific board. Both approaches have their merits, and the choice will depend on
each parliament’s specific circumstances and resources.
Regardless of the chosen approach, the AI governance structure will have a wide-
ranging set of responsibilities, including the following:

• Approving and monitoring AI policy and strategy

• Overseeing the development of an AI code of ethics

34
Guidelines for AI in parliaments

• Managing budgets for AI research and development

• Supervising AI projects from initiation to completion
• Dealing with ethical issues as they arise
• Monitoring AI systems throughout their life cycle, including defining criteria for
when these systems should be decommissioned

Within this framework, it is crucial to define key bodies. These typically include the
following:

• A high-level “governing body” (which may be part of IT or corporate

governance)
• A body responsible for IT and data science
• Business units that act as stakeholders in the AI system life cycle
• An ethics committee – either a new entity or an adapted existing data
committee – to manage the unique ethical challenges posed by AI

While a central group should oversee these roles, day-to-day responsibilities can be
distributed across various areas of parliament. This might involve assigning tasks to
existing functional areas or creating new units specifically to manage AI-related
work. This approach ensures comprehensive governance while allowing for flexibility
in implementation.

By taking these steps, parliaments can create a robust governance structure that
enables them to harness the benefits of AI while effectively managing its risks and
ethical implications.
Actions:
● Assemble a multidisciplinary team from across parliament to lead AI
governance efforts.
● Choose and implement either an integrated or a dedicated AI governance
board structure.
● Define and assign key roles and responsibilities for AI governance, including
policy approval, ethical oversight and project management.
● Establish or adapt an ethics committee to address AI-specific ethical
challenges.
● Develop a clear AI life cycle management process, from project initiation to
system decommissioning.

Establishing an AI governance policy

The purpose of a parliamentary AI governance policy is to establish a unified
approach to AI use within the parliamentary environment. This policy will set
measurable standards and provide top-level monitoring for AI implementation.

The policy development process involves creating a dedicated working group,

defining objectives, and establishing ethical principles aligned with national
regulations and international best practices. It must address legal compliance, data
protection and digital public service legislation.

35
Guidelines for AI in parliaments

The policy will outline roles and responsibilities for all stakeholders involved in the AI
life cycle, from data scientists to legal experts. It will also define AI-supported
business processes, prioritizing them while considering risk tolerance and regulatory
requirements. This includes specifying conditions for AI use, prohibited areas, and
approval processes for certain AI applications.

Guidelines for AI sourcing, development and outsourcing will be established, along

with clear communication strategies to inform all parliamentary staff and MPs about
the policy.

Regarding generative AI, the policy will provide clear usage guidelines and detail
necessary precautions. While recognizing potential risks, it will also encourage
innovative experimentation in a controlled manner, avoiding outright bans that might
lead to unauthorized use on personal devices.
Actions:
● Establish an AI policy working group to lead the development process.
● Outline roles, responsibilities and processes for AI governance and
implementation.
● Create guidelines for AI use, including prohibited areas and approval
processes.
● Define mechanisms for ensuring regulatory compliance.
● Develop a communication strategy to inform all staff and MPs about the AI
policy.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

36
Guidelines for AI in parliaments

Strategic actions towards

AI governance: Strategy
and innovation
About this sub-guideline
This sub-guideline is part of the guideline Strategic actions towards AI governance.
Refer to the main guideline for context and an overview.

This sub-guideline provides guidance and recommendations on AI strategy and

innovation for parliaments.

Background
The journey towards effective AI governance begins with an understanding of its
importance. AI governance is about more than simply managing new technology: it is
about creating a framework that maximizes the benefits of AI while minimizing its
risks.

By taking a strategic approach to AI governance, parliaments can build a robust AI

ecosystem that balances operational needs with innovation and ethical
considerations.
Key points:
● AI governance is crucial for maximizing benefits and minimizing risks.
● A strategic approach balances operational needs, innovation and ethics.

Creating an AI strategy
Once the foundations are established in terms of strong governance and an AI code
of ethics, and once robust engagement with key stakeholders has identified where AI
can add value to the work of parliament, it is time to turn to developing an AI strategy
grounded in this work.

An AI strategy is a comprehensive plan that outlines how parliament will use AI to

achieve its goals and address challenges. It can be part of parliament’s broader

37
Guidelines for AI in parliaments

strategy or a specific document that is aligned with this strategy. It is a corporate-

level strategy that has goals which depend on AI, not a technology strategy. The
main aim is to develop a road map for AI solutions that is aligned with business
needs.

Focusing on parliament’s goals, an AI strategy encompasses AI systems and the

requirements for developing, deploying or purchasing them, with consideration given
to ethical principles, as well as to regulatory and legislative issues. Base
requirements to drive the strategy forward include workforce planning and
infrastructure.

As a high-level document prepared for, and agreed by, senior parliamentary leaders,
an AI strategy should use business language and arguments for business decision
makers. Many senior managers will therefore already be familiar with its format and
structure, and parliament should adopt a structure and approach it already uses, if
appropriate. Alternatively, parliament can follow the sample structure outlined below
and illustrated in Figure 1 (below):
Figure 1: Examples of goals, actions and KPIs in a parliamentary AI strategy

Vision
Formulate a clear vision statement indicating what parliament’s needs are for AI. The
nature of this statement will depend on whether the AI strategy is a stand-alone
document specific to AI, or if it is integrated into a broader parliamentary strategy. In
the latter case, there should be a single, overarching vision.
Goals
Include measurable goals that parliament can achieve using AI systems. These
goals can focus on processes, practices and resources aimed at improving or driving
AI adoption or mitigating AI-related risks.

38
Guidelines for AI in parliaments

Actions or drivers to achieve the goals

Detail specific projects and initiatives that will be implemented to achieve the stated
vision and goals. These projects can affect multiple goals at once. Likewise, a single
goal can be impacted by multiple projects. Projects can directly address business
needs, prerequisites for business needs, or processes and practices to mitigate
risks.
Key performance indicators (KPIs)
State what metrics will be used to measure progress towards the goals. It is often
useful to set a target for each KPI.
Risks
Identify the main risks associated with the inappropriate use of AI, which can justify
specific actions within the strategy.

Adopting an agile approach

An agile approach enables rapid iteration and continuous improvement, making it a
practical way to work when innovating with new technologies such as AI:

• Regularly reviewing and adjusting AI projects based on feedback and

changing business needs is important.
• Highlighting quick wins and sharing successes helps to build momentum and
show the value of AI.
• Publicizing early successes and lessons learned encourages wider adoption
across the organization.
• Engaging leadership by securing agreement from top executives and aligning
AI initiatives with strategic business goals ensures ongoing support and
commitment.
• Keeping leadership informed and involved in AI projects is crucial.

Managing change
Adopting rigorous change management practices within the iterative development
process helps parliaments to manage resistance and ensure the smooth adoption of
AI technologies. It is essential to develop a clear change management plan and to
transparently communicate the goals of AI adoption, as well as the technology’s
impact on the organization, its staff and members. By understanding and carefully
navigating the traditionally conservative culture of parliament and demonstrating
clear, tangible benefits from the adoption of AI, it is possible to foster innovation and
drive successful AI initiatives.

Promoting innovation
AI, as a new technology with immense potential, is very much about innovation. A
good AI governance regime will include ways to promote innovative practices, taking
a strategic and nuanced approach.

The first step is to build a strong case for AI by clearly articulating its benefits and
focusing on how it can address specific business needs and challenges.

39
Guidelines for AI in parliaments

Demonstrating successful AI implementations in similar organizations through data

and case studies can be persuasive. Likewise, selecting projects with a high
likelihood of success can build confidence and show the value of AI.

Starting with small, manageable pilot projects that have clear objectives and
measurable outcomes is crucial. This approach builds knowledge and experience,
helps to develop familiarity and trust in AI-based systems, and can demonstrate
potential, serving as a catalyst for further innovation. Of course, because pilots are
also about experimenting and testing ideas, it is important to accept that some will
inevitably fail. In other cases, it may be determined that the pilot is not worth
pursuing. Building a reflective learning process into the innovation cycle will help
parliaments to realize value and learn lessons as they go.
Innovation can be supported through the following approaches:

● Planning for education and awareness-raising

● Building cross-functional teams
● Encouraging a culture of experimentation
● Leveraging external expertise and partnerships

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

40
Guidelines for AI in parliaments

Strategic actions towards

AI governance:
Stakeholder engagement
About this sub-guideline
This sub-guideline is part of the guideline Strategic actions towards AI governance.
Refer to the main guideline for context and an overview.

This sub-guideline provides guidance and recommendations on engaging with

stakeholders from across parliament as part of the AI governance and adoption
process.

Identifying and engaging with internal stakeholders

Establishing good AI governance in parliament involves many stakeholders from
across the institution. These will come from a range of levels and from across many
different functional areas or organizational units.

Building strong engagement is crucial for creating buy-in within parliament as well for
identifying risks, challenges and opportunities. Identifying which stakeholders need
to be engaged with and then building that engagement with them early in the
process of AI adoption is vital, helping to build support, knowledge and
understanding across parliament.

Since the organizational structure and size of parliaments varies greatly, certain
roles may not exist in specific parliaments (especially in smaller or less well-
resourced legislatures). However, as a general rule, parliament should consider
engaging on the following matters with the bodies, units or teams listed below:

● Approve AI policy, strategy and budget: Senior leadership

● Develop and support AI policy: Legal staff
● Identify legal risks and implications: Legal staff
● Identify areas for AI value addition: Business staff
● Implement data literacy and AI literacy programmes: Training staff
● Ensure practical grounding of strategic actions: IT and data staff
● Manage AI implementation risks: Risk management team

41
Guidelines for AI in parliaments

● Ensure compliance with regulations: Compliance team

● Address ethical considerations: Ethics and data committee
● Plan financially for AI projects: Finance department
● Manage communications about AI initiatives: Communications team
● Oversee AI project implementation: Project management office

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

42
Guidelines for AI in parliaments

Strategic actions towards

AI governance: Find out
more
About this sub-guideline
This sub-guideline is part of the guideline Strategic actions towards AI governance.
Refer to the main guideline for context and an overview.

National government guidelines, strategies and

programmes
Below is a list of national government AI guidelines, strategies and programmes
relating to AI:

● Argentina: Presidencia de la Nación: Plan Nacional de Inteligencia Artificial

(available in Spanish only)
● Australia: Australian Government: Australia’s Artificial Intelligence Action
Plan
● Brazil: Chamber of Deputies: Digital Transformation Strategy of Brazilian
Chamber of Deputies 2021–2024
● Brazil: Ministry of Science, Technology and Innovation: Summary of the
Brazilian Artificial Intelligence Strategy (EBIA) 2021
● Canada: Government of Canada: Pan-Canadian Artificial Intelligence
Strategy
● Denmark: Agency for Digital Government: The Danish National Strategy for
Artificial Intelligence.
● Germany: German Federal Government: Artificial Intelligence Strategy of the
German Federal Government – 2020 Update
● Hungary: Ministry for Innovation and Technology: Hungary’s Artificial
Intelligence Strategy 2020–2030
● Italy: Italian Artificial Intelligence Strategy: Italian Strategy for Artificial
Intelligence 2024–2026
● Japan: Strategic Council for AI Technology: Artificial Intelligence Technology
Strategy

43
Guidelines for AI in parliaments

● Norway: Ministry of Digitalisation and Public Governance: The National

Strategy for Artificial Intelligence
● Spain: Government of Spain: National Artificial Intelligence Strategy (ENIA)
● Switzerland: Federal Council: Guidelines on Artificial Intelligence for the
Confederation: General frame of reference on the use of artificial intelligence
within the Federal Administration
● United Kingdom: Government of the United Kingdom: National AI Strategy

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

44
Guidelines for AI in parliaments

Generic risks and biases

Audience
This high-level guideline is intended for senior parliamentary managers, as well as
for parliamentary staff and MPs interested in gaining a broad understanding of
generic risk and biases associated with AI.

About this guideline

This guideline describes a range of generic risks and biases related to the
implementation of AI technologies, which parliaments will need to understand before
embarking on AI projects and initiatives.

For a discussion of risks that relate more specifically to the unique work of
parliaments, refer to the guideline Risks and challenges for parliaments.

Why inappropriate AI use is a risk

Inappropriate AI use can entail risks at various levels, from the individual to the
global:

• Unintended consequences, such as reinforcing existing biases through biased

systems, resulting in unfair treatment of individuals or groups
• A lack of accountability and transparency – which are crucial for building user
trust – owing to poor understanding of the complexity of AI systems and the
underlying decision-making processes
• The manipulation of public opinion through deepfakes, misinformation and
automated propaganda
• The creation of echo chambers, which amplify biased views and extremism
• Psychological profiling, which allows for the targeted manipulation of
individuals
• Fake content, which can potentially contribute to eroding trust in genuine
information
• Behavioural nudging, which can subtly influence opinions and actions, often
without people’s full awareness, thus potentially undermining democratic
processes and informed discourse
• Physical and psychological harm through the use of AI systems in health care,
autonomous vehicles and industrial automation, which can lead to accidents
or malfunctions

45
Guidelines for AI in parliaments

• Issues such as addiction, anxiety and depression, caused by AI-driven social

media algorithms that promote harmful content or create unrealistic social
comparisons
• Stress, privacy invasion and discrimination through the use of AI systems for
surveillance and profiling, exacerbating mental health problems and social
tensions

Categories of risk
The integration of AI introduces new types of risk that may not be familiar to
parliaments. These can include the following:

● Lack of AI literacy
● Bias and discrimination
● Privacy invasion
● Security vulnerabilities
● Lack of accountability
● Job displacement
● Ethical dilemmas
● Shadow AI
● Lack of data sovereignty
● Lack of trust

For further discussion of these categories, refer to the sub-guideline Generic risks
and biases: Categories of risk.

Identifying biases in a parliament

Bias is a systematic difference in the treatment of objects, people or groups
compared to others, leading to an imbalance in the distribution of data.

Biases are part of people’s lives. They usually start with habits or unconscious
actions (cognitive biases) which, over time, materialize as technical biases (data
biases and processing biases). Such a scenario increases or creates risks that could
result in untrustful AI systems.

Biases in AI systems arise from human cognitive biases, the characteristics of the
data used or the algorithms themselves. Where AI systems are trained on real-world
data, there is the possibility that models can learn from, or even amplify, existing
biases.

In a statistical context, errors in predictive systems are the difference between the
values predicted as model output and the real value of the variables considered in
the sample. When the error occurs systematically in one direction or for a subset of
data, bias can be identified in the data treatment.

Cognitive biases
Cognitive biases are systematic errors in judgements or decisions common to
human beings owing to cognitive limitations, motivational factors and adaptations

46
Guidelines for AI in parliaments

accumulated throughout life. Sometimes, actions that reveal cognitive biases are
unconscious.

For a list of cognitive biases, refer to the sub-guideline Generic risks and biases:
Cognitive bias types.
Data biases
Data biases are a type of error in which certain elements of a data set are more
heavily weighted or represented than others, painting an inaccurate picture of the
population. A biased data set does not accurately represent a model’s use case,
resulting in skewed outcomes, low accuracy levels and analytical errors.

For a list of cognitive biases, refer to the sub-guideline Generic risks and biases:
Data bias types.

Processing and validation biases

Processing and validation biases arise from systematic actions and can occur in the
absence of prejudice, partiality or discriminatory intent. In AI systems, these biases
are present in algorithmic processes used in the development of AI applications.

For a list of cognitive biases, refer to the sub-guideline Generic risks and biases:
Processing and validation bias types.

Interrelationship between biases

Cognitive biases are part the culture of many societies and organizations. They are
often present, unconsciously, in the work processes and decisions that underpin the
functioning of institutions. Over the years, cognitive biases are transformed – often in
combination – into data biases and processing biases.

The underrepresentation or omission of a particular type of data in a data sample

can therefore be the result of one or more of the following factors (among others):

● Systems were built by teams that unconsciously did not involve other
organizational units owing to incorrect judgements regarding their
participation.
● Important stakeholders were not involved in the design of data-entry systems
because they had a different view than project managers.
● System interfaces favoured individual points of view or confirmed
preconceived ideas.
● Irrelevant or incomplete databases were used to train AI systems simply
because they were easy to obtain and avoided the need for negotiation
between managers from different departments.
● AI system projects that revealed decisions based on inappropriate variables
were launched anyway in order to justify the costs already incurred.
● AI system developers were so used to working with certain models that they
used them in situations where they were inappropriate.

47
Guidelines for AI in parliaments

Figure 1: Bias path from the unconscious to untrustful AI systems

Source: Adapted from NIST Special Publication 1270 and the Oxford Catalogue of
Bias

Some biases can multiply the impact of others

Below are some examples of how cognitive biases can influence and, in some
cases, even compound data or processing biases in parliamentary settings:

● Parliament feeds data sets with information from surveys and questionnaires
completed only by people sharing the same political party ideology. Here,
there is a high likelihood of existing affinity bias. Moreover, if this data set
contains data such as “opinion regarding a specific theme”, and it is used to
train an AI algorithm, there is a high possibility that such biases could be
reproduced in that AI system.
● Parliament uses only data sets from a very small number of committee
meetings to train an AI algorithm. In this case, there is a likelihood of
interpretation biases because some terms may have different meanings or
importance to different committees.
● Parliament has spent its entire innovation budget but the project team has
failed to find the best AI algorithm to solve the original problem. The team
implements an AI system anyway, launching it as a successful innovation, in
an attempt to justify the costs. This is a funding bias that results in an AI
system that is not reliable.

As the examples below show, with generative AI tools, all cognitive biases contained
in a vast data set can be combined together and exposed directly to the user:

48
Guidelines for AI in parliaments

● A generative AI tool replicates bias against female job applicants when asked
to draft letters of recommendation. Letters for male applicants often use terms
like “expert” and “integrity” while female candidates are described using terms
such as “beauty” and “delight”.
● Male researchers using a generative AI tool to create avatars receive diverse,
empowering images portraying them as astronauts and inventors. However, a
female researcher receives sexualized avatars – including topless versions,
reminiscent of anime or video-game characters – that she did not request or
consent to.
● A generative AI system fails to create appropriate images of people with
disabilities.

Find out more

● The State of Data & AI Literacy Report 2024
● Rethinking Privacy in the AI Era: Policy Provocations for a Data-Centric World
● Crowdsourcing Moral Machines
● Embedded ethics: some technical and ethical challenges
● UNESCO Recommendation on the Ethics of Artificial Intelligence
● Oxford Catalogue of Bias
● NIST Special Publication 1270: Towards a Standard for Identifying and
Managing Bias in Artificial Intelligence
● ISO/IEC TR 24027:2021: Information technology – Artificial intelligence (AI) –
Bias in AI systems and AI aided decision making

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

49
Guidelines for AI in parliaments

Generic risks and biases:

Categories of risk
About this sub-guideline
This sub-guideline is part of the guideline Generic risks and biases. Refer to the
main guideline for context and an overview. For a discussion of risks that relate more
specifically to the unique work of parliaments, refer to the guideline Risks and
challenges for parliaments.

This sub-guideline explores new types of risk arising from the integration of AI that
may not be familiar to parliaments and that, if not addressed effectively, can
undermine democratic processes and public trust in parliamentary institutions.

Lack of AI literacy
AI literacy is an understanding of the basic principles, capabilities and limitations of
AI – something that is crucial for informed decision-making about AI adoption and
oversight in parliaments. It involves the ability to recognize AI applications, grasp
fundamental concepts like machine learning and data analysis, and critically
evaluate AI’s potential impacts. Without adequate AI literacy, users may misinterpret
AI results, fail to recognize discriminatory patterns, become overly reliant on flawed
AI systems, and overlook ethical and legal implications. This can lead to poor
decision-making and potential harm.
Bias and discrimination
AI systems used in parliamentary functions, such as for automated decision-making
or policy analysis, can reflect and reinforce cognitive and other biases present in
their training data. This can result in skewed policy recommendations and
discriminatory legislative outcomes, adversely affecting minority groups, and
undermining the principles of equality and fairness that underpin democratic
institutions.
Privacy invasion
Parliamentary systems often handle sensitive personal and political data. Improper
data-protection measures can lead to privacy infringements when using AI for data
analysis and decision-making. Unauthorized access to, or misuse of, this data can

50
Guidelines for AI in parliaments

compromise the privacy of citizens, MPs and other stakeholders, eroding trust in
parliamentary processes.
Security vulnerabilities
AI systems, particularly those used in parliamentary settings, are potential targets for
cyberattacks. These attacks can lead to the manipulation or theft of sensitive
legislative data, but can also disrupt parliamentary operations or compromise the
integrity of legislative processes. This poses significant risks to national security and
public safety.
Lack of accountability
The opaque nature of AI decision-making – often termed the “black box” problem –
presents challenges in parliamentary contexts where transparency and
accountability are paramount. Decisions made or influenced by AI without clear
explanations can lead to difficulties in holding the right entities to account for
legislative outcomes, diminishing public trust in democratic institutions.
Job displacement
While AI can improve efficiency, the automation of administrative tasks within
parliamentary functions can lead to job and task displacement, particularly for
support and administrative staff. As AI becomes increasingly adept at handling
routine tasks such as scheduling, document processing and data analysis, the need
for human involvement in these roles may decrease. This reduction in demand can
lead to workforce downsizing, resulting in unemployment and economic disruption
for those affected.

Aside from the loss of jobs, the nature of remaining roles may change significantly.
Tasks that were once performed by human workers may be automated, leading to a
shift towards more complex, decision-oriented or creative responsibilities that require
a higher level of expertise. This evolution in job tasks can be challenging for
employees who may not have the skills or experience needed to adapt, creating
further risks of job insecurity and potential displacement.

The shift towards AI-driven processes also has the potential to increase job
polarization, where low-skill, routine jobs are automated, leaving a gap that may not
easily be filled by existing employees. This could exacerbate social and economic
inequalities, particularly if the affected workers are unable to transition into new roles
that require different skills.
Ethical dilemmas
AI applications in parliamentary settings raise ethical questions, particularly
regarding the delegation of decision-making authority. Relying on AI for policy
recommendations, legislative drafting or constituent services can lead to ethical
dilemmas, especially if AI decisions conflict with human values or lack the necessary
contextual understanding. Different AI services may report varying values depending
on the country in which the underlying model is defined and trained.

51
Guidelines for AI in parliaments

Shadow AI
Shadow AI, which is related to the concept of shadow IT, can be defined as the
unsupervised or unsanctioned use of generative AI tools within an organization or
institution outside of its IT and cybersecurity framework. Shadow AI can expose
organizations to the same risks as shadow IT: data breaches, data loss, non-
compliance with privacy and data protection regulations, lack of oversight from IT
governance, misallocation of resources, and even new risks stemming from a lack of
understanding of the technology, such as the creation of AI models with biased data
that can produce incorrect results.

Lack of data sovereignty

Training and deploying AI systems demands massive computing and storage
resources, often requiring the use of public cloud systems. In some cases, these
cloud systems are located in a different country and are therefore subject to the laws
and regulations of that country. Without appropriate risk-mitigation strategies, such
as encryption or data minimization, it may be difficult for parliaments to maintain
effective control over such AI systems.
Lack of trust
The adoption of AI systems in parliamentary functions carries significant risks related
to a lack of trust. One of the primary concerns is the complexity and opacity of these
systems, which can lead to uncertainty about whether they are providing accurate
and reliable information.

The absence of clear information on how these systems respect privacy or the
nature of the data used for training further exacerbates distrust. Users may be
concerned that their data could be misused or that the AI system’s decisions are
biased or flawed owing to inadequate or biased training data. This lack of trust can
hinder the effective integration of AI in parliamentary operations, as stakeholders
may be reluctant to rely on systems they do not fully understand or trust.

The overall risk is that without trust, the benefits of AI may not be fully realized, as
users may resist or underutilize these systems, potentially leading to inefficiencies
and a failure to achieve the intended improvements in parliamentary processes.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

52
Guidelines for AI in parliaments

Generic risks and biases:

Cognitive bias types
About this sub-guideline
This sub-guideline is part of the guideline Generic risks and biases. Refer to the
main guideline for context and an overview. For a discussion of risks that relate more
specifically to the unique work of parliaments, refer to the guideline Risks and
challenges for parliaments.

This sub-guideline focuses on cognitive biases, which are systematic errors in

judgements or decisions common to human beings owing to cognitive limitations,
motivational factors and adaptations accumulated throughout life. Sometimes,
actions that reveal cognitive biases are unconscious.

Automation bias
Automation bias occurs when conclusions drawn from algorithms are valued more
highly than human analyses. For example, people will often blindly follow satellite
navigation systems and arrive at the wrong place, or cross dangerous streets and
put their life at risk.

Group attribution bias

Group attribution bias refers to the belief that everything that occurs for one
individual is true for everyone. An example is when people stereotype professions
with statements such as “all lawyers are manipulative” or “all artists are eccentric”.

Implicit bias
Implicit bias refers to the practice by which people unconsciously associate
situations with their own mental model of representing that situation. For example,
people often assume that a younger colleague cannot be experienced enough to be
a good manager, or that an older employee is not able to learn new skills.

In-group favouritism
In-group favouritism occurs when someone acts with partiality towards the existing
aspects of the group to which they belong. For example, people may systematically

53
Guidelines for AI in parliaments

recommend someone from their “group” for a job, while sports fans will always view
their team as the best.

Out-group favouritism
Out-group favouritism refers to the favouring of groups outside the group to which a
person belongs. For example, a manager who does not recognize the talent
available in their own team will always turn to someone from another team for advice
or support.

Affinity bias
Affinity bias happens when someone prefers individuals who are similar to them in
terms of ideology, attitudes, appearance or religion. For example, a hiring manager
might prefer a candidate who went to the same university as they did, overlooking
other qualified applicants.

Social bias
Social bias occurs when many individuals in a society or community share the same
bias. The simplest examples are religion and politics. Some people are so closed in
a belief system that they are incapable of seeing both sides of an argument. They
seek only information that supports their belief and negate anything that counters it,
demonstrating their bias in their every action.

Rules and systems bias

Rules and systems bias refers to the fact that, when developers are used to
particular rules embedded in systems, they try to reproduce the same rules to
represent other situations. For example, developers sometimes choose solutions
based on examples they readily remember. Controlled laboratory studies have
identified the harmful effects of specific cognitive biases on several aspects of
software development such as defect density, requirements specification, originality
of design and feature design.

Requirement bias
Requirement bias refers to the assumption that all people or situations are capable
of meeting, or meet, the same technical requirements (hardware and/or software). It
is a subset of “rules and systems bias”.

Anchoring bias
Anchoring bias occurs when people rely too much on pre-existing information, or on
the first information they find, when making decisions. For example, if someone sees
a computer that costs $5,000, and then see a second one that costs $2,000, they are
likely to view the second computer as cheap. This type of bias can impact
procurement decisions.

Availability bias
Availability bias is a mental shortcut whereby people tend to ascribe excessive
weight to what is readily “available” – i.e. what comes easily or quickly to mind –

54
Guidelines for AI in parliaments

when making judgements and decisions. For example, people remember vivid
events like plane crashes over more common incidents such as car crashes, despite
the latter being much more common. As a result, they often overestimate the
likelihood that a plane will crash and might even choose to drive rather than fly, even
though they are much more likely to be involved in a road traffic accident. This type
of bias can occur when business staff are describing business rules to developers.

Confirmation bias
Confirmation bias refers to the fact that people tend to prefer information that
confirms their existing beliefs. It affects how people design and conduct surveys,
interviews or focus groups, and analyse competition. Essentially, people construct
questions in a way that will produce the answers they want. For example, if someone
types the question “Are dogs better than cats?” into an online search engine, articles
that argue in favour of dogs will appear first. Conversely, the question “Are cats
better than dogs?” will produce results in support of cats. This applies to any two
variables: the search engine “assumes” that the person thinks variable A is better
than variable B, showing them results that agree with their opinion first.

User interaction bias

User interaction bias occurs when a user imposes their own self-selected biases and
behaviour when interacting with data, output or results. For example, when a system
is trained using streaming data from a live group discussion, it instils the bias that
exists in that group.

Groupthink
Groupthink refers to the fact that people in a group tend to make non-optimal
decisions based on their desire to conform to the group, or for fear of dissenting. For
example, when the leader of a group tells everyone that they need to ban all
members of a particular ethnic group from joining them, the members of the group
accept that decision without questioning it.

Funding bias
Funding bias occurs when biased results are reported in order to support or satisfy
the organization funding a piece of research. For example, a study published in a
scientific journal found that drinks containing high-fructose corn syrup did not
increase liver fat or ectopic fat deposition in muscles. However, the
“acknowledgements” section shows that one of the researchers received funding
from a major soft-drinks company. The results may therefore have been skewed to
paint the funding organization in a positive light.

Sunk cost fallacy

The sunk cost fallacy is a human tendency to continue with an endeavour or
behaviour because resources such as money, time or effort have already been
invested, regardless of whether the costs outweigh the benefits. For example, in AI,
an organization that has already invested significant time and money in a particular
AI application will pursue it to market rather than deciding to cancel the project, even
in the face of significant technical or ethical debt.

55
Guidelines for AI in parliaments

Rashomon effect
The Rashomon effect is a term derived from the classic 1950 Japanese film
Rashomon, which explores the concept of subjective reality and the nature of truth
by presenting differing accounts of a single event from the perspectives of multiple
characters. This bias occurs when there are differences in perspective, memory and
recall, interpretation, and reporting on the same event from multiple witnesses. For
example, people who attended a legislative committee meeting might have different
perceptions regarding the debate and, therefore, provide a different summary of the
event.

Streetlight effect
The streetlight effect refers to the fact that people tend to search only where it is
easiest to look, such as when data scientists develop an AI algorithm using only a
small data set (i.e. only the data they have access to) instead of considering
obtaining more complete data from other organizations.

Ranking bias
Ranking bias is a form of anchoring bias. It refers to the fact that, in a list of search
engine results, people believe that the highest-ranked results are the most relevant
and important. They will still tend to click more on the top result than others, even if
the results are ranked randomly.

Ecological fallacy
The ecological fallacy refers the drawing of conclusions about individuals based on
group-level data. For example, if a specific neighbourhood has a high crime rate,
people might assume that any resident living in that area is more likely to commit a
crime.

Survivorship bias
Survivorship bias is when people focus on the items, observations or people who
“survive” (i.e. make it past a selection process), while overlooking those who do not.
For example, by assessing only “surviving” businesses and mutual funds, analysts
record positively biased financial and investment information – omitting the many
companies that failed despite having similar characteristics as the successful ones.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

56
Guidelines for AI in parliaments

Generic risks and biases:

Data bias types
About this sub-guideline
This sub-guideline is part of the guideline Generic risks and biases. Refer to the
main guideline for context and an overview. For a discussion of risks that relate more
specifically to the unique work of parliaments, refer to the guideline Risks and
challenges for parliaments.

This sub-guideline focuses on data biases – a type of error in which certain elements
of a data set are more heavily weighted or represented than others, painting an
inaccurate picture of the population. A biased data set does not accurately represent
a model’s use case, resulting in skewed outcomes, low accuracy levels and
analytical errors.

Selection bias
Selection bias occurs when selecting data.

In one example, an AI system for detecting Parkinson’s disease was trained using a
data set containing only 18.6% women. Consequently, the rate of accurate detection
of symptoms was higher among male than female patients even though, in reality,
the symptoms in question are more frequently manifested by female patients.

In another example, an AI system for detecting skin cancer was not able to detect
the disease in people of African descent. Researchers observed that, as rates of skin
cancer were increasing in Australia, the United States and Europe, the data set used
to train the system consisted largely of people of European descent.

Sampling bias
Sampling bias is a form of selection bias in which data is not randomly selected,
resulting in a sample that is not representative of the population. For example, if a
poll for a national presidential election targets only middle-class voters, the sample
will be biased because it will not be diverse enough to represent the entire
electorate.

57
Guidelines for AI in parliaments

Coverage bias
Coverage bias is a form of sampling bias that occurs when a selected population
does not match the intended population. For example, general national surveys
conducted online may miss groups with limited internet access, such as the elderly
and lower-income households.

Participation bias
Participation bias is a form of sampling bias that occurs when people from certain
groups decide not to participate in the sample. It occurs when the sample consists of
volunteers, which also creates a bias towards people who are willing and/or available
to participate. The results will therefore only represent people who have strong
opinions about the topic, omitting others.

Omitted variable bias

Omitted variable bias is a form of sampling bias that occurs when an important
variable is omitted during data collection, compromising an expected result. For
instance, when designing an algorithm that determines the price of second-hand
cars, the developers include the following variables: make, number of seats, accident
history, distance on the clock and engine size. However, they forget to include the
car’s age. The algorithm is likely to give biased estimates because two cars with
exactly the same values for the other variables will probably have different prices
according to their age.

Popularity bias
Popularity bias is a form of sampling bias that occurs when items that are more
popular gain more exposure, while less popular items are underrepresented. For
example, recommendation systems tend to suggest items that are generally popular
rather than personalized picks. This happens because the algorithms are often
trained to maximize engagement by recommending content that is liked by many
users.

Data inaccuracy
Data inaccuracy is a result of failures in data entry. For example, with a system that
registers temperature automatically, if there is a failure in the sensor, the data set will
not be trustful for using temperature as a variable. Sometimes, systems are not rigid
with data entry and accept data without standards or with errors.

Obsolete data
Obsolete data is data that is too old to reflect current trends. For example, a system
designed to predict how long a public procurement exercise will take is trained on an
excessively large data set, consisting mostly of procurement exercises that
happened 10 years ago under different legislation. As a result, this system will likely
produce inaccurate predictions.

58
Guidelines for AI in parliaments

Temporal bias
Temporal bias occurs when the training data is not representative of the current
context in terms of time. For example, census data – which is only collected once
every 10 years – is used for many predictions. However, if the last available census
data was collected in 2021, i.e. in the middle of the COVID-19 pandemic, then
algorithms that use this data may be biased in a number of ways.

Variable selection bias

Variable selection bias occurs when a chosen variable is not fit for purpose. For
example, a national health agency looking to give an additional benefit to citizens
selects, as the variable for allocation of the benefit, total health spending according
to age. The algorithm selects people of European descent and those on higher
incomes to receive the additional benefit. This biased outcome happened because
people in this group spent more on their health. The chosen variable was the seed of
the problem.

Confounding variable
A confounding variable, in research investigating a potential cause-and-effect
relationship, is an unmeasured third variable that influences both the supposed
cause and the supposed effect. For example, when researching the correlation
between educational attainment and income, geographical location can be a
confounding variable. This is because different regions may have varying economic
opportunities, influencing income levels irrespective of education. Without controlling
for location, it is impossible to determine whether education or location is driving
income.

Simpson’s paradox
Simpon’s paradox is a phenomenon that occurs when subgroups are combined into
one group. The process of aggregating data can cause the apparent direction and
strength of the relationship between two variables to change. For example, a study
shows that, within an organization, male applicants are more successful than
women. However, comparing the rates within departments paints a different picture,
with female applicants having a slight advantage over men in most departments.

Linguistic bias
Linguistic bias occurs when an AI algorithm favours certain linguistic styles,
vocabularies or cultural references over others. This can result in output that is more
relatable to certain language groups or cultures, while alienating others.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

59
Guidelines for AI in parliaments

Generic risks and biases:

Processing and validation
bias types
About this sub-guideline
This sub-guideline is part of the guideline Generic risks and biases. Refer to the
main guideline for context and an overview. For a discussion of risks that relate more
specifically to the unique work of parliaments, refer to the guideline Risks and
challenges for parliaments.

This sub-guideline focuses on processing and validation biases, which arise from
systematic actions and can occur in the absence of prejudice, partiality or
discriminatory intent. In AI systems, these biases are present in algorithmic
processes used in the development of AI applications.

Aggregation bias
Aggregation bias arises when a model assumes a one-size-fits-all approach for
different demographic groups that, in reality, may have different characteristics or
behaviours.

Amplification bias
Amplification bias occurs when several AI systems, each with separate biases
influenced by their training data and programming, interact and mutually reinforce
each other’s biases, leading to a more pronounced and persistent bias than what
any single system might display.

For instance, a system trained on historical hiring data, in which male candidates
have been predominantly selected, unintentionally favours male candidates during
CV screening. Another AI system, tasked with performance evaluation, has been
trained on data where female employees were often given lower scores owing to
latent human biases. As these two systems interact, the hiring AI system may
propose a larger number of male candidates, while the performance-evaluation AI
system continues to judge female employees more harshly.

60
Guidelines for AI in parliaments

Deployment bias
Deployment bias – perhaps more of an operational failing than a bias – occurs when
a system that works well in a test environment performs poorly when deployed in the
real world owing to differences between the two environments.

Evaluation bias
Evaluation bias is a type of discrimination in which the methods used to evaluate an
AI system’s performance are biased, leading to incorrect assessments of how well
the system is working.

Exclusion or sampling bias

Exclusion or sampling bias occurs when specific groups of user populations are
excluded from testing and subsequent analyses.

Feedback loop bias

Feedback loop bias arises when the output of an AI system influences future inputs,
potentially reinforcing and amplifying existing biases over time.

Model selection bias

Model selection bias is a technical term for confounding exploratory and hypothesis-
testing statistical analyses. If data is used to select the best-fitting model from a set
of candidates, that same data cannot then be used to test hypotheses about the
value of the estimated parameters of the best-fitting model.

Optimization bias
Optimization bias occurs when the objective function of an AI system is defined in a
way that leads to unintended consequences or unfair outcomes.

Overfitting or underfitting bias

Overfitting bias refers to a situation where a model is too complex and fits too closely
to the training data, potentially incorporating noise or outliers that do not represent
the true patterns in the data. Conversely, underfitting bias occurs when the model is
too simple to capture the true patterns in the data, leading to poor performance and
potentially biased results.

Proxy bias
Proxy bias occurs when variables used as proxies for protected attributes (such as
race or gender) introduce bias into the model.

Temporal bias
Temporal bias occurs when training data becomes outdated and no longer
represents current realities, leading to biased predictions. While this might be
considered a data bias, it is also a processing/validation bias because it often occurs
when systems fail to consider temporal aspects of the data validation process, or

61
Guidelines for AI in parliaments

when the process of updating and validating models fails to adequately account for
changes in the underlying data distribution over time.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

62
Guidelines for AI in parliaments

Ethical principles
Audience
This guideline is intended for senior parliamentary managers and senior IT
professionals in parliaments with responsibility for implementing AI and for its
ongoing governance and management.

About this guideline

Parliamentary use of AI needs to be grounded in strong ethical practices. While this
is true of AI use in any organization, it is particularly important for parliaments, which
must ensure that they maintain public trust and confidence. Although the potential
risks associated with AI use are generally agreed upon, managing and mitigating
these risks requires an understanding of ethical principles.

This guideline and its sub-guidelines present a range of ethical principles related to
AI. They discuss how AI can be implemented ethically across parliamentary
processes and practice, at all levels of the institution. Ethical principles for AI are
explored across eight areas:

• Privacy
• Transparency
• Accountability
• Fairness and non-discrimination
• Robustness and safety
• Human autonomy and oversight
• Societal and environmental well-being
• Intellectual property

Why ethical principles matter

In order to ensure that AI systems are trustworthy and used responsibly, parliaments
should establish a code of ethics for the use of AI. This code should be applied
during the use, development and deployment of AI systems, in order to manage or
mitigate the risks of these technologies while maximizing their benefits.

The code of ethics should be explicit about what parliament expects from the
operation of AI systems and from the people involved in their production and use. It
should align with relevant national and international laws, regulations and standards.
It should include recommendations, guidance and limitations for each ethical

63
Guidelines for AI in parliaments

principle, which should apply throughout the entire AI system life cycle – from
planning to decommissioning.

Developing ethical principles for parliaments

There are many resources available to parliaments to guide them in developing
ethical principles for AI use. Some parliaments may already have a framework in
place that can be adapted.

The following section presents a model that parliaments can adopt if they wish. In
this model, ethical principles for parliamentary AI use are broken down into eight
areas:

● Privacy: AI systems should respect and uphold privacy rights and data
protection.
● Transparency: People should be able to understand when and how they are
being impacted by AI, through transparency and responsible disclosure.
● Accountability: It should be possible to identify who is responsible for the
different phases of the AI system life cycle.
● Fairness and non-discrimination: AI systems should be inclusive,
accessible and not cause unfair discrimination against individuals,
communities or groups.
● Robustness and safety: AI systems should reliably operate in accordance
with their intended purpose.
● Human autonomy and oversight: AI systems should respect people’s
freedom to express opinions and make decisions.
● Societal and environmental well-being: AI systems should respect and
promote societal well-being and environmental good.
● Intellectual property: AI systems should respect intellectual property rights.

Figure 1: Ethical principles for parliaments

64
Guidelines for AI in parliaments

From minimizing biases and ensuring robust oversight to maintaining clear

communication and protecting personal data, these principles work together to
create a comprehensive framework for ethical AI use. By adhering to these
principles, parliaments can harness the benefits of AI while mitigating risks, fostering
public trust and upholding their democratic responsibilities.

Each of these areas is explored in turn in the remainder of this guideline, and in the
associated sub-guidelines, which describe the specific challenges and
considerations for parliaments, and offer practical guidance, actionable strategies
and recommendations.

Privacy
This sub-guideline explores the principle of privacy in AI governance for parliaments,
with a focus on personal data protection. It outlines specific privacy concerns in
various parliamentary work processes, including legislative, administrative and
citizen interaction contexts. It emphasizes the importance of justifying and limiting
the use of personal data in AI systems, and provides guidance on handling sensitive
information. Special attention is given to the challenges posed by generative AI in
processing personal and sensitive data.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that respect privacy and protect personal data.

For further guidance on the principle of privacy, refer to the sub-guideline Ethical
principles: Privacy.

Transparency
This sub-guideline explores the principle of transparency in AI governance for
parliaments. It defines transparency as the communication of appropriate information
about AI systems in an understandable and accessible format. The sub-guideline
addresses three key aspects of transparency: traceability, explainability and
communication.

Highlighting the importance of documenting the entire life cycle of AI systems, from
planning to decommissioning, it provides practical recommendations for
implementing transparency. These include risk assessment documentation,
standardized methods for explaining AI decisions, and clear communication about AI
system capabilities and limitations. The sub-guideline also offers specific guidance
on ensuring transparency in generative AI applications, acknowledging the unique
challenges they present.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that are transparent, accountable and aligned with democratic
values.

For further guidance on the principle of transparency, refer to the sub-guideline

Ethical principles: Transparency.

65
Guidelines for AI in parliaments

Accountability
This sub-guideline explores the principle of accountability in AI governance for
parliaments. It emphasizes that while AI systems themselves are not responsible for
their actions, clear accountability structures are essential.

The sub-guideline discusses the importance of auditability and risk management

throughout the AI system life cycle. It provides practical recommendations for
implementing accountability, including stakeholder identification and risk assessment
processes, and for preparing for both internal and external audits.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that are accountable and aligned with democratic values.

For further guidance on the principle of transparency, refer to the sub-guideline

Ethical principles: Accountability.

Fairness and non-discrimination

This sub-guideline explores the principle of fairness and non-discrimination in AI
governance for parliaments, including minimizing biases in legislative processes and
citizen interactions. It emphasizes the importance of trust and provides specific
recommendations for dealing with potential biases.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that are fair, non-discriminatory and free from biases.

For further guidance on the principle of transparency, refer to the sub-guideline

Ethical principles: Fairness and non-discrimination.

Robustness and safety

This sub-guideline explores the principle of robustness and safety in AI governance
for parliaments, emphasizing that, in order to be trustworthy, AI systems should be
robust to adversity and to changes within the environment for which they were
designed.

The sub-guideline presents the principle of robustness and safety through two
lenses: resilience to failures that could cause damage to people, organizations or the
environment or that could prevent traceability, and resilience to cyberattacks.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that are robust and safe.

For further guidance on the principle of transparency, refer to the sub-guideline

Ethical principles: Robustness and safety.

Human autonomy and oversight

This sub-guideline explores the principle of human autonomy and oversight in AI
governance for parliaments. It refers to the way in which AI systems interact with

66
Guidelines for AI in parliaments

humans, as well as the way in which information is stored, transmitted and secured.
It stresses that parliaments, as enablers of a democratic, flourishing and equitable
society, must support the user’s agency and uphold fundamental rights and that, in
an AI context, this requires human oversight.

In this sub-guideline, special attention is given to the challenges posed by generative

AI, emphasizing the need for robust feedback channels and frequent human checks.
Overall, this sub-guideline provides a framework for parliaments to develop and
maintain AI systems that promote human autonomy and allow for human oversight.
For further guidance on the principle of transparency, refer to the sub-guideline
Ethical principles: Human autonomy and oversight.

Intellectual property
This sub-guideline explores the principle of intellectual property in AI governance for
parliaments. It emphasizes that everyone involved in an AI system’s life cycle,
including users, must respect intellectual property in order to protect the investment
of rights-holders in original content. It covers copyrights, accessory rights, and
contractual restrictions on accessing and using content.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that respect intellectual property rights.
For further guidance on the principle of transparency, refer to the sub-guideline
Ethical principles: Intellectual property.

Societal and environmental well-being

This sub-guideline explores the principle of societal and environmental well-being in
AI governance for parliaments. It emphasizes that, owing to the ubiquitous nature of
AI in society, this technology should be used for people’s well-being. It further
stresses that applications of AI should not negatively affect people’s physical and
mental well-being or harm the environment or society at large.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that protect and promote societal and environmental well-being.

For further guidance on the principle of transparency, refer to the sub-guideline

Ethical principles: Societal and environmental well-being.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

67
Guidelines for AI in parliaments

Ethical principles: Privacy

About this sub-guideline
This sub-guideline is part of the guideline Ethical principles. Refer to the main
guideline for context and an overview.

This sub-guideline explores the principle of privacy in AI governance for parliaments,

with a focus on personal data protection. It outlines specific privacy concerns in
various parliamentary work processes, including legislative, administrative and
citizen interaction contexts. It emphasizes the importance of justifying and limiting
the use of personal data in AI systems, and provides guidance on handling sensitive
information. Special attention is given to the challenges posed by generative AI in
processing personal and sensitive data.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that respect privacy and protect personal data.

Why privacy matters

In the context of digital transformation, it is important – and often a legal requirement
– to protect personal data. AI systems are no exception to this rule, with parliaments
needing to comply with both legislation and internal standards on this subject.

Before embarking on an AI project, parliaments will therefore need to delve deeper

into the problem to be solved with AI in order to identify whether any type of personal
data will be collected, processed and potentially shared, clearly recording the
classification of the data and its precise meaning.

The following sections address specific privacy concerns raised by the diverse work
processes of parliaments.
Use of personal data
It is important to exercise caution and restraint when using personal data, and only to
do so where absolutely necessary. Parliaments should adhere to the following
principles:

• Where personal data is deemed essential to an AI system’s functionality, a

clear and compelling justification for its use must be made. This justification
should be subject to rigorous scrutiny and approval by a data protection

68
Guidelines for AI in parliaments

officer (or equivalent person) and by key decision makers within the AI
governance framework.
• If approval for the use of personal data is given, strict practices must be put in
place to safeguard privacy and to prevent misuse. Such practices must
protect individuals from exposure, even indirectly, especially when dealing
with biometric data or when combining information from multiple sources.
• AI systems must not profile individuals according to their behaviour or use
personal data in ways that could lead to discrimination, the manipulation of
opinions, or any form of harm, whether psychological, physical or financial.
• Explicit authorization should be required for the use of sensitive data, adding
an extra layer of protection and accountability.
• Special conditions may be required for the use of personal data for research
purposes or to support bills going through parliament, especially if parliament
already has internal regulations regarding the use of personal data.

Administrative processes
Where parliament is adopting or developing AI systems, it should identify,
understand and document what data is being used – both internal data, and
externally sourced or hosted data – and identify who the owner of that data is.
Citizens’ data
When interacting with citizens, parliaments must take special care to manage and
protect the personal data they collect, such as through an online digital service or a
manual data-collection process. They must also carefully consider what data is
stored in a system that is exposed to AI, and ensure that only essential data is
retained. More generally, when designing an AI system, parliaments need to
understand the parameters of data privacy, knowing what is admissible for release
into the public domain, what must be anonymized, and what is protected.
Sensitive data and generative AI
Parliaments must exercise extreme caution and appropriate scrutiny when feeding
personal and sensitive data into generative AI systems, as these systems will
process and use any data given to them. The institution should have in place
mechanisms to protect its personal and sensitive data from inadvertent or
inappropriate access by such tools. This is especially important if this data is
processed externally, as is the case with most generative AI systems.

Where a parliament does authorize personal data for use by generative AI systems,
it should actively implement processes to anonymize this data, as well as adopting
other mechanisms, established by internal rules, before submitting any personal data
to such tools. This practice minimizes the risk of personal data breaches and misuse.

Practising privacy
In order to ensure that AI systems respect and protect privacy, parliaments should
adopt a comprehensive approach. The components of this approach are detailed
below:

69
Guidelines for AI in parliaments

● Conduct a thorough assessment of AI systems to identify any use of personal

data, clearly documenting data classification and purpose.
● Implement strict data protection practices, including obtaining approval from
the data protection officer for any use of personal data in AI systems.
● Establish clear protocols for managing citizens’ data in AI-driven interactions,
ensuring that only essential data is collected and stored.
● Develop and enforce stringent safeguards for handling sensitive data,
particularly when using generative AI tools.
● Create a comprehensive data ownership and management system,
documenting both internal and external data sources used in AI processes.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

70
Guidelines for AI in parliaments

Ethical principles:
Transparency
About this sub-guideline
This sub-guideline is part of the guideline Ethical principles. Refer to the main
guideline for context and an overview.

This sub-guideline explores the principle of transparency in AI governance for

parliaments. It defines transparency as the communication of appropriate information
about AI systems in an understandable and accessible format. The sub-guideline
addresses three key aspects of transparency: traceability, explainability and
communication.

Highlighting the importance of documenting the entire life cycle of AI systems, from
planning to decommissioning, it provides practical recommendations for
implementing transparency. These include risk assessment documentation,
standardized methods for explaining AI decisions, and clear communication about AI
system capabilities and limitations. The sub-guideline also offers specific guidance
on ensuring transparency in generative AI applications, acknowledging the unique
challenges they present.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that are transparent, accountable and aligned with democratic
values.

Why transparency matters

Transparency involves communicating appropriate information about AI systems to
the right people and in a free, understandable and easily accessible format.

Transparency – throughout the entire life cycle of an AI system – encompasses three

key aspects: traceability, explainability and communication. These are discussed
below.

71
Guidelines for AI in parliaments

Traceability
Traceability implies the ability to follow and monitor the entire life cycle of an AI
system, from the definition of its purpose, through to planning, development, use and
ultimate decommissioning.

Architects, developers, decision makers and even users involved in the development
and evolution of AI systems are advised to use a combination of tools and
documentation to support traceability.

Explainability
Explainability is the ability for humans to understand and trust each decision,
recommendation or prediction made by an AI system.

As complexity increases in AI systems, explainability declines. Consequently, initially

simple AI systems become less explainable as new layers of functionality are added
over time.

Since different AI system stakeholders require different types of explanations,

parliaments must generate documentation aimed at decision makers and those
responsible for AI governance, in addition to the documents produced by the
development team.

Communication
Communication is important for transparency: humans must always know that they
are interacting with an AI system. As such, any AI system that interacts with humans
must identify itself unambiguously. It must be explained to users and practitioners, in
a clear and accessible manner, how the system functions and what its limitations
are.

Practising transparency in AI systems

In order to ensure that AI systems are transparent, parliaments should adopt a
comprehensive, life cycle-wide approach. The components of this approach are
detailed below:

• Risk assessment: Produce a comprehensive risk assessment to guide

project authorization, development and maintenance. This assessment should
consider all stakeholders and inform decisions from initiation to potential
decommissioning.
• Standardization: As part of the AI systems development process, adopt a
standardized transparency method, such as Explainable AI (XAI), to
document key aspects including problem definition, data selection criteria,
personal data usage, technical specifications, user feedback and oversight
results. Capture the rationale behind all significant decisions.
• Reporting: Maintain transparency through regular behaviour reports and
continuous data storage for auditing. Clearly communicate system
expectations, limitations and potential abnormalities to all relevant parties.

72
Guidelines for AI in parliaments

• Communication: Ensure that AI applications interacting with humans

disclose their artificial nature. Inform business managers about AI usage in
their areas of responsibility.
• Documentation: Tailor transparency documentation to the intended
audience, whether internal or external. For outsourced AI systems, clearly
communicate and enforce transparency requirements with external providers.
Practising transparency in generative AI systems
Parliaments using generative AI must prioritize transparency and responsibility,
recognizing that the learning process and data used by AI systems may not be
transparent:

• Label AI-assisted documents, specifying the tool and version used.

• Establish clear guidelines for permissible AI use in document creation.
• Document AI processes from design to deployment, including mechanisms for
ensuring trust in AI systems.
• Prioritize commercial AI tools aligned with human rights frameworks.
• Justify and document any use of personal or third-party data.
• Clearly communicate when AI outputs are probabilistic rather than factual.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

73
Guidelines for AI in parliaments

Ethical principles:
Accountability
About this sub-guideline
This sub-guideline on accountability is part of the guideline on the ethical principles
for the use of AI in parliaments.

It explores the principle of accountability in AI governance for parliamentary settings.

The sub-guideline discusses the importance of auditability and risk management
throughout the AI system lifecycle. It provides practical recommendations for
implementing accountability, including stakeholder identification, risk assessment
processes, and preparing for both internal and external audits.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that are accountable, fair, and aligned with democratic values.

Why accountability is important

The accountability principle in AI governance focuses on establishing clear
structures, processes, and tools to evaluate and hold AI systems accountable, as the
systems themselves cannot be responsible for their actions. Parliaments should
ensure clear accountability for all decisions and actions throughout an AI system's
lifecycle, from planning to decommissioning.

Effective accountability relies on two key elements:

1. Auditability: The ability to track the entire process of an AI system's lifecycle,

including planning, development, use, and maintenance. This depends
heavily on transparency and may involve both internal and external audits.
2. Risk Management: The identification, evaluation, documentation, and
minimisation of risks associated with AI systems. This proactive approach
helps identify potential vulnerabilities and assigns responsibility for risk
mitigation.

These practices are particularly important for parliaments that undergo frequent
audits. By implementing robust accountability measures, parliaments can ensure

74
Guidelines for AI in parliaments

their AI systems remain trustworthy and aligned with their democratic

responsibilities.

Practices towards accountability

To ensure accountability in AI systems in parliament, it’s crucial to adopt a
comprehensive approach throughout the system’s lifecycle. Begin by identifying all
the stakeholders affected by the AI system, whether directly or indirectly. This holistic
view helps anticipate potential impacts and concerns.

Next, implement a robust risk management process. This should encompass

identifying, evaluating, documenting, minimising and continuously monitoring risks
associated with AI systems. Such a process allows for the proactive management of
potential issues before they escalate.

Establish rigorous internal auditing processes for AI systems. These regular checks
help maintain system integrity and provide ongoing assurance of compliance with
ethical standards and operational requirements.

It’s equally important to prepare staff for external audits. Provide thorough training to
equip team members with the knowledge and skills needed to engage confidently
with third-party auditors, ensuring transparency and cooperation.

Finally, conduct a thorough assessment to identify which AI systems require

trustworthy certification. Once identified, develop a clear, actionable plan to achieve
this certification. This step not only enhances the system’s credibility but
demonstrates a commitment to maintaining high standards of AI governance.
By implementing these measures, parliaments can create a culture of accountability
around their AI systems, fostering trust and ensuring responsible deployment of this
powerful technology.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

75
Guidelines for AI in parliaments

Ethical principles:
Fairness and non-
discrimination
About this sub-guideline
This sub-guideline is part of the guideline Ethical principles. Refer to the main
guideline for context and an overview.

This sub-guideline explores the principle of fairness and non-discrimination in AI

governance for parliaments, including minimizing biases in legislative processes and
citizen interactions. It emphasizes the importance of trust and provides specific
recommendations for dealing with potential biases.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that are fair, non-discriminatory and free from biases.

Why fairness and non-discrimination matter

Fairness can be defined at the individual level (such ensuring similar individuals are
treated consistently) or at the group level. In the latter case, this involves grouping
people into categories and ensuring that these groups are treated equitably.

Fairness, in the context of AI, is the ability of AI systems to not discriminate or

reinforce biases against any individual or group. This principle is based on
impartiality and inclusion. Fair, non-discriminatory decisions therefore presuppose
bias-free data and algorithms.

Practising fairness and non-discrimination

In order to ensure that AI systems are fair and non-discriminatory, parliaments
should adopt a comprehensive approach to data management and bias mitigation.
The components of this approach are detailed below:

• Data quality management: Establish robust processes to manage data

quality, particularly for data sets likely to be used in AI systems. Implement
practices to ensure that there are no biases in the data and in the models that

76
Guidelines for AI in parliaments

will be used to train the algorithms. Such practices should consider not only
data biases and processing biases, but also cognitive biases (for further
guidance on this topic, refer to the guideline Generic risks and biases and its
associated sub-guidelines).
• Staff training: Provide staff with training in data ethics, focusing on identifying
and minimizing biases throughout the AI development process.
• Data governance: Implement a data governance process, with a clear
delineation of responsibilities between data owners and data stewards.
• Collaboration: Have IT and business units work closely together. Such
collaboration is vital for predicting, minimizing and monitoring biases
throughout the AI system life cycle.
• Data ethics committee or team: Establish a data ethics committee or a
multi-skilled team capable of analysing potential biases and communicating
them to both managers and IT teams for each AI project.
• Diversity and inclusivity: Prioritize diversity and inclusivity when forming
project teams and data ethics committees. By bringing together individuals of
different ages, genders, ethnicities and skill sets, parliaments can ensure that
a broad range of perspectives are heard, reducing the risk of that potential
biases could be overlooked and enhancing the overall fairness of AI systems.

Minimizing biases in parliamentary processes

When planning and developing AI systems for use in legislative processes,
parliaments should:

● Ensure that the data does not contain biases regarding political-party ideology
and previous value judgements
● Be aware of possible historical biases in data relating to committee meetings
and plenary sessions
● Establish partnerships with public organizations from which they regularly
source external data for AI-powered bill-drafting systems, in order to maintain
data quality
● Be aware of biases in text translation and speech-to-text transcription
● Confirm whether the information produced by generative AI systems is free
from biases before considering using them

When planning and developing AI systems for use in government oversight

processes, parliaments should:

● Identify data quality problems in government data and alert the government
agency in charge of the data
● Establish partnerships with government agencies in charge of the data in
order to improve data quality and minimize biases

When planning and developing AI systems for use in citizen interaction processes,
parliaments should:

● Identify biases coming from citizens

● Avoid internalizing biases presented by citizens
● Avoid exposing any biases when interacting with citizens

77
Guidelines for AI in parliaments

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

78
Guidelines for AI in parliaments

Ethical principles:
Robustness and safety
About this sub-guideline
This sub-guideline is part of the guideline Ethical principles. Refer to the main
guideline for context and an overview.

This sub-guideline explores the principle of robustness and safety in AI governance

for parliaments, emphasizing that, in order to be trustworthy, AI systems should be
robust to adversity and to changes within the environment for which they were
designed.

The sub-guideline presents the principle of robustness and safety through two
lenses: resilience to failures that could cause damage to people, organizations or the
environment or that could prevent traceability, and resilience to cyberattacks.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that are robust and safe.

Resilience to cyberattacks
Cyberattacks against AI systems exploit both algorithmic opacity and the strong
dependence of algorithms on data. Such attacks may be difficult to detect in a timely
manner, requiring AI systems security management practices that encompass
advanced prevention techniques and focus on restoring the system and the entire
environment to normal operating conditions.

Resilience to failure
Failures can occur in AI systems when variables take on unknown or false values
that the developer did not consider and did not programmatically act to prevent.
While AI systems are generally expected to be robust, if such failures do occur, there
must be a mechanism to restore the system to its normal state in a timely and
responsible manner, with minimal loss of data or impact on parliament.

79
Guidelines for AI in parliaments

Practising robustness and safety

In order to ensure that AI systems are robust and safe, parliaments should adopt a
comprehensive, dynamic risk management approach that adapts to the ever-
changing environment in which these systems operate. The components of this
approach are detailed below:

• Comprehensive testing: Identify and mitigate cyber threats specifically

targeting AI systems, while not neglecting other potential vulnerabilities.
• Security practices: Tailor security practices to address the unique
challenges posed by AI systems and the threats they face, including through
close and rapid communication between data teams and information security
experts. When developing AI systems, cybersecurity should be a primary
consideration, integrated from the outset, rather than added as an
afterthought.
• Training: Invest in continuous training for developers and information security
staff. These staff should be well-versed in techniques to prevent cyberattacks
on AI systems and equipped with disaster recovery strategies specific to
these technologies.
• Internal collaboration: Ensure that internal business units responsible for AI
systems work closely with IT departments to establish clear parameters for
monitoring system behaviour and defining thresholds for alerts regarding
suspicious activity.
• External partnerships: Forge partnerships with other public institutions.
These alliances facilitate swift and effective communication about emerging
threats and new attack categories. They also provide a platform for sharing
experiences – both successes and failures – in implementing various security
techniques and technologies.

By adopting this holistic approach, parliaments can create a resilient framework for
AI systems that can withstand threats, adapt to changes, and continue to serve their
intended purpose effectively and safely.

Maintaining safety when using generative AI

When implementing generative AI in parliamentary contexts, safety considerations
are paramount:

• Maintain strict control over data access, ensuring that AI systems and tools
only interact with data specifically authorized for their intended purpose. This
approach safeguards sensitive information and maintains the integrity of
parliamentary processes.
• Where data transfer to external cloud services raises security concerns or
presents other risks, explore alternative solutions. One viable option is to
employ open-source generative AI models that can run locally on a
parliament’s own systems. This strategy provides the benefits of generative AI
while offering full control over security, data management and integrity.

By adopting these measures, parliaments can harness the power of generative AI

while upholding the highest standards of data protection and operational safety.

80
Guidelines for AI in parliaments

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

81
Guidelines for AI in parliaments

Ethical principles: Human

autonomy and oversight
About this sub-guideline
This sub-guideline is part of the guideline Ethical principles. Refer to the main
guideline for context and an overview.

This sub-guideline explores the principle of human autonomy and oversight in AI

governance for parliaments. It refers to the way in which AI systems interact with
humans, as well as the way in which information is stored, transmitted and secured.
It stresses that parliaments, as enablers of a democratic, flourishing and equitable
society, must support the user’s agency and uphold fundamental rights and that, in
an AI context, this requires human oversight.

In this sub-guideline, special attention is given to the challenges posed by generative

AI, emphasizing the need for robust feedback channels and frequent human checks.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that promote human autonomy and allow for human oversight.

Human autonomy and protection of citizens’ rights

Humans should be free to express their opinions and make decisions about their
lives without interference, coercion or manipulation. In order to ensure that AI
systems do not negatively affect citizens’ rights, it is important for parliaments to
understand how AI systems interact with humans and how information is stored,
transmitted and secured.

Maintaining human autonomy is particularly challenging in certain areas, especially

web searches, content curation, content moderation, browsing activities, and email
and text communications stores in the cloud. When parliaments rely on AI systems
to execute core public functions, they should ensure that the design and operation of
these systems complies with international human rights standards, as part of their
duty to promote freedom of expression.

Moreover, in order to preserve human agency, direct interaction between human end
users and an AI system should be established in a way that avoids simulating social
relationships or stimulating potentially negative or addictive behaviour.

82
Guidelines for AI in parliaments

Human oversight
In the day-to-day running of organizations, human oversight is exercised through
human supervision of an AI system’s outputs. Users and managers responsible for
AI systems analyse this output to ascertain whether undesirable behaviours have
occurred, whether the rules established at the development stage need to be
modified, or whether there are any data biases that went unnoticed during the
development of the system.

The nature of human oversight depends on the type of AI application. Supervision

can occur during development, on an ongoing basis, or once the system is in
production, in order to gather feedback on the system’s output.

There are three types of human supervision that can be applied to AI systems:

● Human-in-the-loop (HITL): Under this model, a human mediates all

decisions made by the AI system. While this approach offers the highest level
of human control, it is not always desirable or feasible, particularly for systems
designed for rapid decision-making or high-volume data processing.

● Human-on-the-loop (HOTL): This approach allows for human intervention

during the project development phase. Once the system is operational, the
human’s role shifts to monitoring the system’s operation and decisions. This
approach balances automation with human oversight, allowing for intervention
when necessary.

● Human-in-command (HIC): This is the most comprehensive form of

oversight, extending beyond the AI system’s immediate functioning to
consider broader economic, social, legal and ethical impacts. Under this
model, oversight can even extend to society at large, with public feedback
gathered on the AI system’s behaviour providing a broader perspective on its
effects and implications.

The distinction between these three approaches – HITL, HOTL and HIC – lies
primarily in the level of autonomy granted to the AI system and the extent of human
oversight. These are summarized in Table 1 below:

Table 1: Models of Human-AI interaction

83
Guidelines for AI in parliaments

By implementing these oversight approaches, parliaments can harness the benefits

of AI while maintaining essential human control and accountability, thus upholding
democratic principles and public trust.

Practising human autonomy and oversight

In order to safeguard human autonomy and maintain proper oversight of AI systems,
parliaments should adopt a comprehensive approach. The components of this
approach are detailed below:

• Risk assessment: Conduct a thorough risk assessment of each AI system,

paying particular attention to those that interact directly with human end users.
For these systems, identify any potential for confusion about who or what is
engaging in the interaction.
• Rules: Establish clear rules for AI-human interactions in order to prevent any
manipulation or the formation of inappropriate social relationships. The type of
oversight required for each AI system should be determined according to its
specific risk profile.
• Standards and testing: Develop a set of clear, measurable criteria for
acceptable and unacceptable AI behaviours, and draw up an extensive testing
plan to explore the full range of system behaviours. Parliament’s AI policy
should designate a specific position or organizational body with the authority
to withdraw an AI system from operation if it cannot meet these behavioural
standards.
• Training: Provide thorough training on the assessment process to both
technical staff and managers, including on the use of any specific tools or
functionalities built into the AI system itself.
• Reporting: Produce regular oversight reports, appropriately tailored for both
technical staff and managers.
• Review: Establish a timely and efficient process for reviewing these human
oversight reports to ensure that any issues or concerns are addressed
promptly, thus maintaining the integrity and trustworthiness of the AI systems
in use.

By implementing these practices, parliaments can leverage the capabilities of AI

systems while ensuring that these remain under appropriate human control and
respect human autonomy.

Human oversight of generative AI

Parliaments should establish robust oversight mechanisms for generative AI:

• Create a digital channel for user feedback on AI outputs.

• Invest in staff training for effective AI oversight.
• Conduct more frequent human checks on AI-generated content, given the
rapid advancements in this technology.
• Carefully select generative AI tools and ensure that all users understand their
specific limitations.

84
Guidelines for AI in parliaments

These measures allow parliaments to leverage generative AI while maintaining

essential human control and ethical standards.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

85
Guidelines for AI in parliaments

Ethical principles:
Societal and
environmental well-being
About this sub-guideline
This sub-guideline is part of the guideline Ethical principles. Refer to the main
guideline for context and an overview.

This sub-guideline explores the principle of societal and environmental well-being in

AI governance for parliaments. It emphasizes that, owing to the ubiquitous nature of
AI in society, this technology should be used for people’s well-being. It further
stresses that applications of AI should not negatively affect people’s physical and
mental well-being, or harm the environment or society at large.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that protect and promote societal and environmental well-being.

Practising societal and environmental well-being

In order to ensure that AI systems promote and support societal and environmental
well-being, parliaments should adopt a comprehensive approach. The components
of this approach are detailed below:

● Stakeholder identification: Identify potentially impacted stakeholders

(individuals, organizations and the environment) during AI system planning.
● Procurement: Prioritize environmentally responsible suppliers in AI-related
public procurement exercises.
● Impact assessment: Assess the impacts of an AI system on all stakeholders
throughout its life cycle.
● Feedback: Implement user feedback mechanisms to gauge an AI system’s
impact on work processes.
● Citizen interaction: Ensure unbiased and accurate AI interactions with
citizens.
● Misinformation: Check AI-generated communication outputs for
misinformation.

86
Guidelines for AI in parliaments

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

87
Guidelines for AI in parliaments

Ethical principles:
Intellectual property
About this sub-guideline
This sub-guideline is part of the guideline Ethical principles. Refer to the main
guideline for context and an overview.

This sub-guideline explores the principle of intellectual property in AI governance for

parliaments. It emphasizes that everyone involved in an AI system’s life cycle,
including users, must respect intellectual property in order to protect the investment
of rights-holders in original content. It covers copyrights, accessory rights, and
contractual restrictions on accessing and using content.

Overall, this sub-guideline provides a framework for parliaments to develop and

maintain AI systems that respect intellectual property rights.

Protecting intellectual property in AI systems

In order to protect intellectual property in AI systems, parliaments should adopt a
comprehensive approach. The components of this approach are detailed below:

● Before starting to develop an AI system, check for copyrighted data and any
contractual conditions.
● When using generative AI, check whether the presented data sources
generate copyrighted content.
● Standardize the types of documents for which generative AI cannot be used
to generate content.
● Expressly inform readers if a document’s content has been written using
generative AI.
● Protect parliament’s unpublished or sensitive work by avoiding uploading it
into an online AI system unless there are assurances that the data will not be
reused.
● Train internal staff on the technical and ethical implications of intellectual
property rights.
● Create a specific peer-review process for researchers who are using
generative AI, in order to both maintain high standards of quality and protect
against breaches of intellectual property rights.

88
Guidelines for AI in parliaments

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

89
Guidelines for AI in parliaments

Introducing AI
applications
Audience
This high-level guideline is intended for senior parliamentary managers, as well as
for parliamentary staff and MPs who are interested in gaining a broad understanding
of where AI can impact upon the work of parliaments.

About this guideline

This guideline outlines considerations and recommendations for parliaments facing
new challenges in procuring, implementing and managing AI systems at a time when
this technology is becoming increasingly prevalent in standard software packages. It
addresses how parliaments can approach AI-enhanced off-the-shelf products while
adhering to the ethical principles and risk management strategies outlined in these
Guidelines.

Procurement considerations
Increasingly, many off-the-shelf software packages are being augmented with AI
functionalities. This AI integration is often opaque, and it is not immediately apparent
to the user what impact AI is having, or how it works behind the scenes:

● Microsoft 365 and Microsoft Edge products are starting to embed Microsoft
Co-pilot AI support.
● Google Docs already has new AI features such as “Help me write”, “Smart
compose”, “Summarization” and “Voice typing”.
● Adobe Acrobat has integrated several AI functionalities, including AI Assistant
for generating summaries and creating multi-document insights.
● Photo and video editing software often contains AI augmentation that makes it
easy to render manipulated images.

When procuring AI-enhanced products, parliaments must therefore exercise due

diligence to ensure alignment with their ethical standards and operational needs.

Checklist:
● Conduct thorough vendor assessments, focusing on AI ethics and data
practices.

90
Guidelines for AI in parliaments

● Evaluate AI features against parliamentary needs and ethical guidelines.

● Ensure contracts include robust clauses on data protection and AI
accountability.

Implementation strategy
A measured approach to implementation allows parliaments to assess the impact of
AI functionality and its alignment with existing processes. Starting with a pilot phase
provides an opportunity to develop clear protocols for AI feature usage and establish
monitoring mechanisms.

Checklist:
● Begin with a pilot phase to evaluate AI impact and alignment.
● Develop protocols for enabling or restricting AI features based on task
sensitivity.
● Establish monitoring mechanisms for AI-driven decisions or suggestions.

User training and awareness

Comprehensive training is crucial to ensure staff can leverage AI features effectively
while understanding their limitations and potential risks. This is particularly important
given the ethical considerations and potential biases inherent in AI systems.

Checklist:
● Develop training programmes highlighting AI benefits and risks (including data
literacy and AI literacy).
● Create user guidelines for the appropriate usage of AI functionalities.
● Educate staff on recognizing and critically evaluating AI-generated content.

Ongoing management and ethical considerations

As AI capabilities evolve, parliaments must regularly review and update their policies
and practices. This includes maintaining open communication with vendors,
conducting periodic audits, and ensuring ongoing alignment with ethical principles
such as fairness, transparency and human oversight.

Checklist:
● Regularly review and update AI usage policies.
● Maintain open communication channels with vendors.
● Conduct periodic audits of AI feature usage and impact.
● Assess and mitigate potential biases in AI-enhanced features.
● Ensure transparency and maintain human oversight in AI use.

Data governance and performance evaluation

Strict data governance is essential when using AI-enhanced products, particularly
given the sensitive nature of parliamentary work. Regular performance evaluations
help to ensure that AI features continue to meet parliamentary needs and ethical
standards.

91
Guidelines for AI in parliaments

Checklist:

● Implement strict data access controls for AI-enhanced features.

● Ensure compliance with data protection regulations and parliamentary
policies.
● Regularly assess the effectiveness and appropriateness of AI features.
● Gather user feedback to inform future procurement and implementation
decisions.

Conclusion
By following these recommendations and checklists, parliaments can harness the
benefits of AI-enhanced off-the-shelf products while mitigating risks and upholding
ethical standards. This approach aligns with the broader AI governance framework
outlined in the Guidelines, ensuring a consistent and responsible approach to AI
adoption across all areas of parliamentary work.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

92
Guidelines for AI in parliaments

Training for data literacy

and AI literacy
Audience
This guideline is intended for a diverse audience within parliaments, including staff
and MPs without prior AI expertise, senior staff responsible for AI implementation,
decision makers at various levels (particularly those responsible for training and staff
development budgets), users of parliamentary technologies, and technical staff
involved in AI production and procurement.

About this guideline

This guideline focuses on the importance of awareness-raising and training for both
data literacy and AI literacy in parliaments. It outlines the need for a comprehensive
training programme tailored to different roles within parliament, including decision
makers, users, technical staff and MPs, providing detailed recommendations for
training content in both areas.

The AI literacy sub-guideline includes the fundamentals of AI, machine learning

basics, ethical implications and practical applications in governance. The data
literacy sub-guideline, meanwhile, encompasses topics such as data management
principles, data collection, cleaning and analysis, and ethical considerations.

This guideline stresses the importance of understanding AI’s potential benefits and
risks in a parliamentary context, as well as the ethical considerations involved. It also
highlights the need for a multidisciplinary approach and the development of a data-
driven culture within parliaments.

Overall, this guideline provides guidance for parliaments on building capacity and
preparing their staff for the effective and responsible use of AI technologies.

The need for training

The transformative potential of AI in parliaments necessitates comprehensive staff
preparation and training. All parliamentary staff – from technical implementers to end
users – will benefit greatly from understanding AI in general and, more specifically,
its impact on their roles. Knowledge of the related ethical considerations and
complexities is also important.

93
Guidelines for AI in parliaments

By strategically building staff skills and capabilities, parliaments can maximize the
benefits of AI while effectively managing its risks, avoiding both overestimation of
AI’s impact and underestimation of its challenges.

Given that AI systems rely on data, parliamentary training programmes should cover
both AI literacy and data literacy – since good-quality, well-managed and understood
data is at the heart of successful AI implementations. It is important for parliaments
to start developing these programmes before working with AI.
What is data literacy?
Data literacy is the ability to read, understand, create and communicate data as
information. It involves understanding how to effectively collect, analyse, interpret
and present data in meaningful ways. Data literacy includes knowing where data
comes from, and grasping basic statistical concepts and data presentation and
visualization techniques. Data literacy is vital for critically evaluating data-driven
arguments and conclusions.

Those requiring a more advanced level of data literacy will need to understand the
following topics:

• Defining the strategic questions to be answered by data

• Identifying and analysing the dependencies between questions and data
• Designing data architectures
• Implementing data governance and data management
• Analysing and processing data, including practices for data ethics
• Building digital products based on data analysis, including dashboards and
data visualizations
What is AI literacy?
AI literacy is an understanding of the basic principles, capabilities and limitations of
AI – something that is crucial for informed decision-making about AI adoption and
oversight in parliaments. It involves the ability to recognize AI applications, grasp
fundamental concepts like machine learning and data analysis, and critically
evaluate AI’s potential impacts. An AI-literate workforce can better leverage AI to
enhance parliamentary functions while identifying and mitigating risks, ensuring
responsible AI use that aligns with democratic principles.

A more advanced level of AI literacy encompasses a deep understanding of the

following:

● The fundamental principles of AI

● The risks of AI systems, as well as the resulting business value and outcomes
● Technology and applications
● Analytical and algorithmic methods
● Data and most dimensions of data literacy
● Ethical considerations regarding AI

94
Guidelines for AI in parliaments

Relationship between data literacy and AI literacy

Data readiness is a prerequisite for AI readiness in parliaments and requires
multidisciplinary working teams across the product portfolio and throughout the
system’s life cycle. It is important for all AI initiatives to consider data governance
and management processes and, as Figure 1 below shows, for data literacy to be a
prerequisite for AI literacy training.

Figure 1: Possible roles to be considered in AI literacy and data literacy training

programmes

The figure above emphasizes the importance of starting a data literacy programme
before working with AI, and why it is important that parliaments have a plan to
ensure good levels of AI literacy.

In addition, the emergence of generative AI takes the technology to the end user and
gives them the power to harness it in their work. For this reason, it is important for
staff and MPs to understand the basic tenets of data literacy and AI literacy,
including the risks and downsides, before they utilize such tools in parliament.
Data literacy in an AI context
By establishing a training programme and building a solid foundation in data
management, parliaments can harness the full potential of their data assets to
support evidence-based decision-making, foster public trust and uphold democratic
principles.

For further guidance on the training requirements for data literacy when using AI,
refer to the sub-guideline Training for data literacy and AI literacy: Data literacy in an
AI context.
Developing AI literacy
AI literacy is crucial in parliaments because it enables MPs, decision makers and
staff to make informed choices about AI adoption, shape appropriate policies and
regulations, and effectively oversee AI-driven initiatives.

95
Guidelines for AI in parliaments

Having a well-trained workforce that is familiar with at least the basic tenets of AI will
help parliament to both leverage the opportunities AI presents for enhancing
parliamentary functions and mitigate the potential risks that can occur.

For further guidance on developing AI literacy in parliaments, refer to the sub-

guideline Training for data literacy and AI literacy: Developing AI literacy.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

96
Guidelines for AI in parliaments

Training for data literacy

and AI literacy: Data
literacy in an AI context
About this sub-guideline
This sub-guideline is part of the guideline Training for data literacy and AI literacy.
Refer to the main guideline for context and an overview.

Why data literacy matters

Data literacy is the ability to read, understand, create and communicate data as
information. It involves understanding how to effectively collect, analyse, interpret
and present data in meaningful ways. Data literacy includes knowing where data
comes from, and grasping basic statistical concepts and data presentation and
visualization techniques. Data literacy is vital for critically evaluating data-driven
arguments and conclusions.

In parliaments, data has emerged as a critical asset across all business domains. As
parliaments increasingly rely on data-driven insights to fulfil their mandate, the
importance of robust data management practices cannot be overstated. Becoming
data-literate empowers parliaments to harness the full potential of their data assets,
driving informed decision-making, enhancing transparency, fostering public trust and
strengthening the foundations of democratic governance. This is fundamental to the
adoption of AI.
Data literacy training for MPs and non-technical staff
Users are important actors in a data culture, since they add data into many systems
and are often the people who come face-to-face with the data. It is important for
them to understand basic data principles and to be attuned to possible errors or
problems that can arise.

Increasingly, too, users are extracting, combining and otherwise repurposing data,
often into top-line reports or business dashboards. In all cases, this requires quality
assurance, as well as an understanding of where the data comes from and what it
means.

97
Guidelines for AI in parliaments

AI can be a powerful tool for analysing and understanding data and trends. But it can
also be unreliable, which is why data literacy is especially important in this context.

Parliaments could develop or source training programmes on the following topics in

order to achieve a good level of data literacy across MPs and a wide range of non-
technical staff:

● Introduction to data literacy

● Data collection and management
● Data analysis fundamentals
● Data visualization and presentation
● Ethical data usage and privacy
● Data-driven decision-making
● Introduction to AI and machine learning
● Critical thinking with data

These courses could be offered at different levels (basic, intermediate or advanced)

to cater to learners in different roles and with varying levels of exposure to AI-based
systems within parliament.

Data literacy training for decision makers

AI-related data literacy is crucial for senior leaders and decision makers in
parliaments, as it enables informed decision-making, effective risk management and
strategic planning for AI adoption. Data-literate leaders are able to exercise oversight
over AI projects, optimize resource allocation and foster innovation, while ensuring
that AI is used ethically.

By equipping decision makers with these skills, parliaments can ensure that AI
adoption is guided by informed leadership, aligning with institutional goals while
adhering to ethical standards and best practices.

A targeted data literacy training programme for senior leaders and decision makers
could include the following:

• Dedicated workshop: This session covers the importance of data-driven

initiatives, AI readiness and associated risks. Participants gain an overview of
foundational data management concepts tailored to their level, reaching a
comprehensive understanding of the AI landscape in a parliamentary context.
• Just-in-time learning and self-paced resources: Additional learning
resources allow senior leaders and decision makers to embed their
knowledge and understanding of AI.

Data literacy training for technical staff

Data literacy training is of paramount importance for technical staff, as these
individuals are at the forefront of implementing and managing AI systems in
parliaments. Their expertise directly impacts the effectiveness, ethical use and
security of AI applications in parliamentary operations.

98
Guidelines for AI in parliaments

A data literacy training programme for this population could be structured as follows,
depending on the needs of parliament:

• Data management foundations: The programme begins with an overview of

the fundamental concepts of data management, ensuring participants have a
solid understanding of data types, integrity, governance and life-cycle
management. This foundational knowledge is critical, since the quality and
management of data directly affect the performance and reliability of AI
systems.
• Practical skills: The programme then progresses through practical skills in
data collection, cleansing and storage. These skills are essential for preparing
and maintaining the high-quality data sets that AI systems rely on. Advanced
topics such as database management, cloud storage solutions and data
visualization techniques could be included to ensure technical staff can
effectively handle and communicate insights from large, complex data sets.
• Non-technical aspects: Importantly, the programme also addresses key non-
technical topics including data ethics, legal compliance and the application of
data in parliamentary contexts. This training ensures that technical staff not
only have the skills to implement AI systems, but also understand the broader
implications and responsibilities of using AI in a parliamentary setting.

By providing comprehensive training in this way, parliaments can ensure that their
technical staff are well-equipped to lead the responsible and effective
implementation of AI technologies, ultimately enhancing the efficiency and
effectiveness of parliamentary operations.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

99
Guidelines for AI in parliaments

Training for data literacy

and AI literacy:
Developing AI literacy
About this sub-guideline
This sub-guideline is part of the guideline Training for data literacy and AI literacy.
Refer to the main guideline for context and an overview.

Why AI literacy matters

AI literacy is the ability to understand, critically evaluate and effectively interact with
AI technologies. In a parliamentary context, it encompasses knowledge of the
capabilities and limitations of AI, as well as an understanding of its potential impacts
on legislative processes and democratic governance. Moreover, AI literacy can help
parliaments to foster a culture of responsible innovation.

AI literacy is crucial in parliaments because it enables MPs, senior parliamentary

managers and staff to make informed choices about AI adoption, shape appropriate
policies and regulations, and effectively oversee AI-driven initiatives.

Having a well-trained workforce that is familiar with at least the basic tenets of AI will
help parliament to both leverage the opportunities AI presents for enhancing
parliamentary functions and mitigate the potential risks that can occur.

AI literacy training is needed across a range of parliamentary stakeholders. The

nature and content of this training will differ depending on their role. A potential
outline of training programmes for MPs, decision makers and technical staff is given
below.

AI literacy training for MPs

MPs face a dual challenge: they must understand AI in order to effectively use it and
oversee its implementation within parliament, while also grasping its broader
implications to inform their legislative work.

100
Guidelines for AI in parliaments

A flexible AI literacy programme, tailored to the unique culture of each parliament

and the demanding schedules of its MPs, would be a useful way to address this
need. A potential structure for such a programme is outlined below:

• AI fundamentals: The programme begins with a clear introduction to the

fundamentals of AI, exploring its core concepts and potential applications in
legislative work. MPs learn how AI can enhance parliamentary functions, from
improving constituent services to streamlining research processes.
• Risks and ethical implications of AI: Since understanding the potential of AI
is only half of the equation, the programme also delves into the associated
risks and ethical implications, covering topics such as privacy concerns,
potential biases and societal impacts.
• AI governance: This section of the programme covers AI governance and its
role in risk mitigation. For parliaments with existing AI policies, the programme
explains the relevance of these policies to MPs’ daily work. For parliaments
that already use AI systems in their legislative processes, the programme
offers practical insights into the operation and impact of these systems.
• Ethical principles: Throughout the programme, emphasis is placed on the
ethical principles that form the cornerstone of responsible AI use in
democratic institutions. MPs explore concepts such as transparency,
accountability and the preservation of human autonomy, all of which are
crucial for promoting societal well-being in an AI-driven world.

By the end of this programme, MPs are equipped with the knowledge to confidently
engage in AI-related policy discussions, make informed decisions about AI adoption
in parliamentary processes, and navigate the increasingly AI-influenced landscape of
modern governance. This comprehensive yet accessible approach ensures that MPs
can harness the benefits of AI while safeguarding democratic values and protecting
the public interest.

AI literacy training for decision makers

Parliamentary decision makers need a solid understanding of AI in order to guide
their institutions effectively. They must equip themselves with essential knowledge
about AI and its implications for parliamentary work. A possible structure for a
training programme for this audience is given below:

• AI fundamentals: The training begins with an explanation of the fundamental

concepts of AI and its potential applications within parliamentary contexts.
• Governance, opportunities and risks: Decision makers learn how AI can
enhance legislative processes and streamline administrative tasks, as well as
examining the potential risks of the technology. The programme also looks at
how effective AI governance can mitigate these risks and ensure responsible
use.
• Broader implications: The programme covers the broader legal, ethical and
social implications of AI, covering key concepts such as privacy,
transparency, accountability and fairness.
• Limitations of AI: Participants explore the limitations of AI – and of
generative AI in particular – to ensure they are equipped with the knowledge
needed for realistic implementation.

101
Guidelines for AI in parliaments

• AI legislative frameworks and policies: For parliaments with specific AI

legislative frameworks or policies, the programme covers the implications of
these frameworks or policies for parliamentary AI use cases.

AI literacy training for technical staff

Technical staff involved in AI production and procurement require a comprehensive
understanding of AI technologies and their parliamentary context and application. A
training programme for these staff could cover several key areas:

● AI fundamentals: The programme begins with an exploration of the history of

AI, key concepts, and the role of the technology in governance and legislation.
This foundational knowledge helps align AI implementations with
parliamentary needs.
● Machine learning and deep learning: Participants explore various learning
paradigms, common algorithms and frameworks. Hands-on training with
relevant tools provides practical knowledge for developing AI models that
address legislative challenges.
● Ethical implications: A significant portion of the programme focuses on the
ethical and social aspects of AI. Participants engage in critical discussions
about bias, fairness, transparency, and the impact of AI on privacy and human
rights. This ethical grounding ensures AI systems uphold democratic
principles.
● Governance-specific applications: Through case studies and practical
projects, technical staff learn to identify opportunities for enhancing legislative
processes with AI. They work on developing AI solutions for real
parliamentary challenges, considering both technical feasibility and ethical
implications.
● AI tools and resources: The programme introduces a range of AI
development tools, platforms and resources. Participants gain experience with
AI libraries, application programming interfaces (APIs), data sets and pre-
trained models. They also practice creating effective prompts for generative AI
tools while adhering to ethical principles.

For parliaments primarily using generative AI, the programme could be adapted to
emphasize avoiding inaccuracies, hallucinations and biases in AI outputs. It could
also stress the importance of clear guidelines to protect against adversarial prompts
and to maintain information security.

By the end of this comprehensive programme, technical staff are well-equipped to

lead AI initiatives within their parliaments. They possesses the technical expertise
needed to implement AI solutions, the ethical foundation necessary for ensuring
responsible use, and the contextual understanding required to align these
technologies with parliamentary needs and values.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

102
Guidelines for AI in parliaments

Planning and
implementation

103
Guidelines for AI in parliaments

Project portfolio
management
Audience
This guideline is intended for parliamentary staff including senior leaders who may
not have a technical background in AI but who are involved in the management or
oversight of AI projects.

About this guideline

This guideline examines the practice of managing a portfolio of AI projects.

The ability to effectively manage a diverse array of AI projects is increasingly

important as AI becomes more commonplace in parliaments. By strategically
prioritizing and aligning AI projects with organizational goals, parliaments can
maximize their impact.

This guideline therefore aims to equip parliaments with the knowledge and tools they
need to navigate the complexities of AI project portfolio management, ensuring that
AI technologies are implemented successfully and in line with ethical principles.

What is project portfolio management?

Project portfolio management (PPM) refers to the centralized management of a
parliament’s programmes, projects and related activities. It is used to meet strategic
objectives by optimizing resource allocation, balancing risks and maximizing overall
value.

Why project portfolio management is relevant to AI

governance
AI project portfolio management (AI PPM) entails overseeing a diverse array of AI
technologies and initiatives in order to optimize performance and achieve strategic
objectives as part of ongoing strategic governance.

One key consideration in AI PPM is recognizing that AI systems are designed to

evolve over time, adapting to changing needs and circumstances. With multiple large

104
Guidelines for AI in parliaments

language models (LLMs) available in the field of generative AI, parliaments must
carefully evaluate and select the most suitable AI technologies to support their
objectives.

Effective AI PPM involves ensuring that AI initiatives are aligned with organizational
goals, while also prioritizing transparency, accountability and ethical considerations.
By adopting a proactive approach to AI PPM, parliaments can harness the full
potential of AI to drive innovation, efficiency and effectiveness in legislative
processes and governance.

From segmentation to portfolio management

AI PPM provides a structured approach to strategically selecting, prioritizing and
overseeing AI projects within parliaments. Unlike traditional project management,
which focuses on individual projects, AI PPM takes a holistic view, considering the
collective impact of AI projects and their alignment with organizational goals and
priorities.

At its core, AI PPM involves identifying, evaluating and prioritizing AI projects based
on their potential value, feasibility and alignment with strategic objectives. Each
potential or existing AI project can be evaluated against seven key criteria:

1. Strategic alignment: Consider whether the project aligns with parliament’s

overarching strategy and goals. Projects that contribute directly to achieving
strategic objectives should be given higher priority.
2. Measurable impact: Prioritize projects with clear, objective and measurable
impact metrics in order to ensure a tangible return on investment.
3. Augment or replace: Determine whether the project will augment current
human operations or replace existing manual processes entirely. Projects that
enhance human capabilities or efficiency should be prioritized over those that
solely aim to automate existing processes.
4. Nature of the problem: Evaluate whether the problem being addressed is
suitable for AI-driven solutions. Projects that align with the capabilities of AI
technologies and have clear problem-solving potential should be prioritized.
5. Data availability: Assess the availability and quality of data required for the
project. Projects for which the necessary data is already available or can
easily be acquired should be prioritized, as data availability is essential for AI
model training and performance.
6. Technological capability and skills: Consider whether parliament
possesses the technological infrastructure and skill set required to develop,
deploy and scale up the solution successfully. Projects that align with existing
technological capabilities and expertise should be prioritized in order to
minimize implementation challenges.
7. Ethical considerations: Finally, ensure that all ethical considerations –
including bias mitigation, privacy protection and transparency – have been
thoroughly evaluated for each project and that the project aligns with agreed
organizational values.

105
Guidelines for AI in parliaments

Prioritization
Once parliament has identified and evaluated AI projects, the next important step is
to review the institution’s (potential) AI portfolio as a whole and to prioritize the
workstream:

● Evaluate how each AI project aligns with parliament’s overall strategic goals
and rank them according to potential value and impact.
● Assess resource availability, conflicts and constraints.
● Evaluate the potential benefits and risks of each AI project and prioritize those
with favourable risk-reward ratios.
● Identify projects that are prerequisites for others or that could create synergies
if implemented together, and consider prioritizing those that unlock value in
other projects or create a foundation for future initiatives.
● Understand the time sensitivity of each project, looking to balance quick wins
and long-term strategic value.
● Assess the impact of each project on key stakeholders (both internal and
external) and prioritize those with high levels of stakeholder support.

Methodologies and frameworks for PPM

There are many standards and frameworks that organizations can use to support the
implementation of a PPM approach. Some examples are given below:

• The Standard for Portfolio Management (SPM): a standard developed by the

Project Management Institute (PMI)
• Disciplined Agile (DA): a toolkit, also developed by PMI, that includes portfolio
management practices
• https://www.pmi.org/learning/library/pathway-organizational-project-
management-maturity-8221Organizational Project Management Maturity
Model (OPM3): another model developed by PMI
• Projects IN Controlled Environments (PRINCE2): a project management
method that also has implications for portfolio management
• Lean Portfolio Management: a method that is part of the Lean-Agile approach
• Hoshin Kanri: a strategic planning process that can be applied to portfolio
management
• Objectives and Key Results (OKRs): an approach that can be used to align
portfolios with organizational goals

Other approaches that can be used for PPM include Balanced Scorecard and
Theory of Constraints.

Parliaments can use their existing methodologies to support AI PPM, or they can
adopt an established external framework that fits well with their culture and working
methods.

For further guidance on developing or adopting a framework for PPM, refer to

the sub-guideline Project portfolio management: The STEP approach.

106
Guidelines for AI in parliaments

Find out more

● IPU Centre for Innovation in Parliament (CIP), IT Governance Hub: Framing
the development of IT governance for parliaments

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

107
Guidelines for AI in parliaments

Project portfolio
management: The STEP
approach
About this sub-guideline
This sub-guideline is part of the guideline Project portfolio management. The main
guideline should be read first for context and an overview.

This sub-guideline introduces an example of a project portfolio management (PPM)

methodology that parliaments could implement to support their AI programme.

The STEP approach

The STEP approach offers a structured way for parliaments to implement AI
technologies and leverage their potential to enhance productivity and efficiency.
Following this model allows parliaments to navigate the introduction of AI solutions
with precision and foresight, ultimately driving positive outcomes and organizational
success.

The STEP approach comprises four key stages:

● Segmentation involves identifying and segmenting tasks suitable for AI

integration, recognizing that no single AI system can fulfil all requirements.
● Transition is the process of gradually incorporating AI systems into existing
workflows, allowing for the deepening or upgrading of work roles in order to
maximize the benefits of AI technologies.
● Education plays a crucial role in ensuring staff readiness and competency in
effective AI use, encompassing training on AI functionality, potential biases
and best practices for integration.
● Performance evaluation measures the impact and efficacy of AI systems in
real-world contexts, providing valuable insights into their tangible benefits and
areas for improvement.

These stages are shown Figure 1 and discussed in turn below.

108
Guidelines for AI in parliaments

Figure 1: The STEP approach

Segmentation
Segmentation offers a systematic framework for analysing activities within
parliamentary proceedings by delineating related processes and tasks.

Segmentation entails categorizing tasks into three distinct types:

• Tasks that AI cannot or should not perform according to decisions made by

those in charge of AI governance
• Tasks where AI can augment staff actions (augmentation)
• Tasks that can be automated by AI (automation)

Crucially, this approach emphasizes that parliamentary staff should take the lead in
task segmentation, leveraging their domain expertise to identify suitable tasks for AI
integration. It also underscores the importance of staff experimentation with AI tools
before widespread adoption, enabling them to assess suitability, usability and
effectiveness in real-world contexts.

By adopting a proactive approach to task segmentation following a business analysis

or a job competency analysis, parliaments can maximize the benefits of AI systems
while ensuring alignment with organizational goals and priorities. Having a clear list
of tasks identified for augmentation or automation will be crucial for the ideation
phase. Within an AI PPM context, these tasks become the primary candidates for
digital transformation.
Transition
The transition phase of the STEP approach focuses on changes to parliamentary
staff members’ roles as tasks are augmented or automated through the integration of
AI systems.

During this phase, resources freed up by AI implementation are carefully considered

and assigned to one of three categories:

109
Guidelines for AI in parliaments

• Job elimination: resources are redeployed to other functions within parliament

• Job upgrading: tasks previously performed by more senior staff are now
supported by AI-enabled systems
• Job deepening: staff members have the opportunity to spend their time on
tasks with greater added value, gain deeper knowledge in their domain, or
engage in skill-building activities such as preparing training materials

By strategically managing this transition, parliaments can optimize resource

allocation, enhance staff productivity and capitalize on the transformative potential of
AI technologies.
Education
In the context of parliamentary AI integration, education is a multifaceted approach
aimed at fostering a culture of continuous learning and skill development, with a
particular focus on embedding an ethos of AI learning into the organizational culture.
As part of this, parliaments should frequently review and update learning materials to
ensure they keep pace with evolving AI trends and best practices, as well as
revisiting the segmentation process regularly to identify new opportunities for AI
integration.

Parliamentary staff can be enrolled on certifying courses that teach them the
necessary knowledge and skills to effectively leverage AI solutions in their roles.
These skills could include fine-tuning documents or data priority from the
organization, mastering prompt engineering to create effective commands or
prompts for AI systems, and evaluating the validity of predictions made by these
systems.
Performance evaluation
The performance evaluation phase acknowledges that the introduction of AI systems
in parliaments represents a shift rather than just a lift in operational dynamics.
Performance evaluation encompasses the following aspects:

• Recalibrating individual annual performance metrics to encompass factors

such as speed, efficiency, accuracy and creativity, reflecting the contribution
of AI systems and related training
• Identifying and measuring appropriate performance indicators for AI systems
to determine whether these systems are having a measurable impact that
aligns with the organization’s strategic objectives
• Regularly assessing the viability and effectiveness of AI systems, with
feedback loops that provide valuable qualitative metrics to evaluate adoption
rates and user satisfaction
• Evaluating the effectiveness of AI-related training programmes, and
continuously updating learning materials to ensure staff remain up to date with
the latest AI advancements and methodologies
• Closely monitoring the availability of new AI technologies and reviewing task
segmentation accordingly

110
Guidelines for AI in parliaments

Summary
The STEP approach provides a structured and adaptable AI PPM method for
parliaments. It helps ensure alignment with organizational goals, optimizing resource
allocation and maximizing the overall impact of AI investments. By adopting this
approach, parliaments can navigate the complex landscape of AI projects with
confidence, driving innovation, efficiency and effectiveness in legislative processes
and governance.

Find out more

● IPU Centre for Innovation in Parliament (CIP), IT Governance Hub: Framing
the development of IT governance for parliaments

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

111
Guidelines for AI in parliaments

Data governance
Audience
This guideline is intended for parliamentary staff involved in the oversight and
implementation of AI-based systems, including business managers, chief information
officers, chief technology officers and IT managers. It will also be useful for senior
parliamentary managers responsible for AI governance.

About this guideline

This guideline outlines the desirable characteristics of data for safe AI systems
development. It highlights data-quality issues to avoid during the planning and
development of AI systems, and explains how to implement privacy principles to
protect personal data – an essential factor for AI development. These practices
should be guided by data governance that aligns with the organization’s needs.

Since data exists independently of AI systems and is used by parliaments, data

management practices should be established before any AI initiatives commence.
This lays the foundation for trustworthy AI systems in legislative bodies.

Why is data governance relevant to AI governance?

The quality and reliability of AI system outputs depend heavily on two factors:

● The quality of data used to train the AI model

● The quality of data the AI system uses during its operation

It is crucial to protect this data from unauthorized access and misuse. Improving data
quality and enhancing data protection are key components of an organization’s AI
governance strategy. Achieving these improvements requires coordinated actions
and agreements among various stakeholders involved in data-related decisions and
processes.

Data governance plays a crucial role here. It involves coordinating and managing the
efforts of all stakeholders to enhance data quality and protect privacy. By effectively
implementing data governance practices, organizations can ensure they are
developing AI systems that are reliable and trustworthy.

112
Guidelines for AI in parliaments

Data quality
Data quality refers to certain features of data that make it accessible, useful and
reliable to support effective decision-making. Data must be demonstrably accurate,
complete, consistent, accessible, relevant and secure.

For further discussion of this subject, refer to the sub-guideline Data governance:
Data quality.

Personal data protection

Personal data protection refers to practices, policies and legislation designed to
safeguard individuals’ personal data from unauthorized access, misuse or exposure.
It encompasses various measures to ensure that personal data is collected, stored,
processed and shared in a way that respects individuals’ privacy and complies with
relevant laws and regulations.

For further discussion of this subject, refer to the sub-guideline Data governance:
Personal data protection.

Data governance in a parliamentary context

Data governance is crucial for coordinating efforts to enhance data quality and
privacy. It encompasses policies, roles, responsibilities, processes and technology
aimed at improving data quality and creating an optimal data environment.

For further discussion of this subject, refer to the sub-guideline Data governance:
Data governance in a parliamentary context.

Data management for AI systems

Data management is a crucial factor in the implementation of AI systems in
parliaments. Staff must understand the key steps for establishing effective data
management practices, including creating governance policies, managing metadata,
ensuring data quality and protecting personal information.

For further discussion of this subject, refer to the sub-guideline Data governance:
Data management for AI systems.

Find out more

● Australian Government, Office of the Australian Information Commissioner:
Australian Privacy Principles guidelines
● Government of Brazil: Guia de Elaboração de Programa de Governança em
Privacidade (available in Portuguese only)
● Government of the United Kingdom: The Government Data Quality
Framework
● Government of the United Kingdom: Using personal data in your business or
other organisation
● Norwegian Data Protection Authority: Artificial intelligence and privacy

113
Guidelines for AI in parliaments

● Publications Office of the European Union: Data quality requirements for

inclusive, non-biased and trustworthy AI: Putting science into standards

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

114
Guidelines for AI in parliaments

Data governance: Data

quality
About this sub-guideline
This sub-guideline is part of the guideline Data governance. Refer to the main
guideline for context and an overview.

This sub-guideline explains why data quality matters, explores the dimensions of
data quality, and examines both the benefits of high-quality data and the risks
associated with low-quality data.

Why data quality matters

Data quality refers to certain features of data that make it accessible, useful and
reliable to support effective decision-making. Data must be demonstrably accurate,
complete, consistent, accessible, relevant and secure.

High-quality data is essential for parliamentary processes. It forms the foundation of

reliable, data-driven decisions that can improve operational efficiency, reduce both
operational and strategic risks, and help meet overall business needs. When AI
systems are integrated into decision-making processes, they can amplify the benefits
of digital solutions. However, it is important to note that this integration can also
magnify any associated risks. Therefore, ensuring data quality becomes even more
critical when AI is involved in parliamentary operations.

Dimensions of data quality

Data quality can be evaluated according to a number of criteria, such as the
following:

● Accessibility is the extent to which information is available, or easily and

quickly retrievable. Data availability requirements per business process
should be identified.
● Appropriate amount of information is the extent to which the volume of
information is appropriate for the task at hand. This can be managed by
defining, for each data element, how critical the amount of information
captured is for analysis purposes.

115
Guidelines for AI in parliaments

● Completeness is the extent to which information is not missing and is of

sufficient breadth and depth for the task at hand. This aspect is managed by
reporting on the completeness of data fields and devising plans to capture all
data as per business requirements.
● Consistency is the extent to which information is presented with the same
content across multiple systems and platforms. This aspect is managed by
understanding data standards and business rules and ensuring that systems
adhere to the defined rules.
● Freedom from errors is the extent to which information is correct and
reliable, and the degree of agreement between a data value (or set of values)
and a source assumed to be correct. Freedom from errors may be attained by
defining validation rules, conducting regular testing and reporting on samples
of data for compliance.
● Relevance is the extent to which information is pertinent to business
processes. This aspect is managed by determining the business use of each
data element and assessing its value and relevance through user feedback.

Benefits of high-quality data

High-quality data has numerous benefits for parliaments, including the following:

● Trustworthy AI systems: Accurate and reliable data enables better-trained

AI models.
● Improved analytics: High-quality data enhances the accuracy of analytics,
leading to more precise and actionable insights, conclusions and predictions.
● Improved decision-making: Accurate and reliable data allows parliaments to
make better-informed decisions with less risk.
● Reduced risk: Secure and protected data can help to prevent fraud, financial
losses and reputational damage.
● Cost savings: High-quality data reduces the costs associated with correcting
errors, and reduces the time needed to deal with data-related issues.
● Informed policymaking: Accurate and reliable data enables lawmakers to
draft bills based on solid evidence, leading to more effective and impactful
legislation.
● Support for legislative research: High-quality data is crucial for conducting
thorough legislative research, enabling MPs to understand complex issues
and make informed decisions.
● Support for innovation: High-quality data can be used to develop new
products and services, since the data can reveal not only hidden problems but
also potential ways to solve such problems.
● Enhanced public trust: Transparent and high-quality data fosters trust
among citizens, who can see that decisions are based on accurate
information.

Key data-quality issues

The list below outlines some of the primary obstacles to high-quality data:

● Data integrity issues: Errors in data entry or processing can compromise the
accuracy and reliability of data.

116
Guidelines for AI in parliaments

● Duplicate data: Having multiple records for the same entity can lead to
confusion and errors in reporting and analysis, making it difficult to keep
different versions of data in sync across operations.
● Inconsistent data: Variations in data formats or standards across different
systems can cause integration issues and inaccuracies.
● Outdated data: Using old or obsolete data can lead to decision-making based
on irrelevant information.
● Incomplete data: Missing data fields can lead to gaps in analysis and hinder
comprehensive decision-making.
● Ambiguous data: Data that is unclear or lacks context can be misinterpreted,
leading to incorrect conclusions.
● Misinformed decisions: Inaccurate or incomplete data can lead decision
makers to make the wrong choices.

Risks associated with low-quality data

In parliamentary contexts, low-quality data can cause a unique set of problems,
some of which are discussed below:

● Misguided legislation: When flawed data is used for evidence, research or

policy analysis, this can lead to legislation that is ineffective or has unintended
negative consequences. For example, an environmental protection bill based
on inaccurate pollution statistics might target the wrong industries or fail to
address the most pressing issues.
● Ineffective resource allocation: Budgetary decisions based on unreliable
data can lead to inefficient resource allocation. For example, allocating funds
to a social programme based on outdated poverty statistics might result in the
programme not reaching the communities most in need.
● Reduced public trust: The perception that parliamentary decisions are
based on questionable data can erode trust in the legislative process and
undermine the effectiveness of government institutions.
● Data privacy concerns: Mishandling sensitive data can lead to breaches and
loss of public trust. For example, a leak of citizens’ personal data can cause a
public outcry and lead to legal repercussions.
● Biases in data collection: Inherent biases in data-collection methods can
skew policy outcomes. For example, surveys that do not adequately represent
minority groups can lead to policies that overlook their needs.
● Overreliance on quantitative data: Ignoring qualitative data leads to
incomplete analysis. For example, focusing solely on statistical data without
considering public opinion or sentiment can result in unpopular and ineffective
policies and mean that key social trends or attitudes are overlooked.

For example: if a parliamentary committee uses flawed crime statistics to justify

increased police funding, the subsequent revelation of data errors could damage
public confidence in both the decision-making process and the resulting policy
changes.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

117
Guidelines for AI in parliaments

Data governance:
Personal data protection
About this sub-guideline
This sub-guideline is part of the guideline Data governance. Refer to the main
guideline for context and an overview.

This sub-guideline looks at data protection regulations, specific types of sensitive

data, and broader data privacy issues in the context of AI systems.

What is personal data protection?

Personal data protection refers to practices, policies and legislation designed to
safeguard individuals’ personal data from unauthorized access, misuse or exposure.
It encompasses various measures to ensure that personal data is collected, stored,
processed and shared in a way that respects individuals’ privacy and complies with
relevant laws and regulations.
Data protection regulations
Most countries have data protection regulations that govern how personal data is
used. In all cases, parliaments must comply with relevant local regulation(s) by
adopting or adapting the measures and processes required by law.

While the exact rules will vary in each case, such regulations generally impose the
following requirements:

● Personal data must be used fairly, lawfully and transparently.

● Personal data must be used for specified, explicit purposes.
● Personal data must be used in a way that is adequate, relevant and limited to
only what is necessary.
● Personal data must be accurate and, where necessary, kept up to date.
● Personal data must be retained for no longer than is necessary.
● Personal data must be handled in a way that ensures appropriate security,
including protection against unlawful or unauthorized processing, access,
loss, destruction or damage.

118
Guidelines for AI in parliaments

Sensitive data
There may be stronger legal protections for more sensitive information, such as the
following:

● Race and/or ethnic background

● Political opinions
● Religious beliefs
● Trade union membership
● Genetics
● Biometrics (where used for identification)
● Health
● Gender
● Sexual orientation

Data privacy issues

In today’s increasingly digitized society, there is a growing risk of data being wrongly
shared, stolen or leaked, and of inaccuracies perpetuating through multiple systems.
In the context of AI, some of the issues that must be addressed include the following:

● Exposure to privacy breaches and security incidents: Data breaches

might cause parliament to suffer long-lasting reputational damage and legal
consequences, including fines, lawsuits and other regulatory sanctions.
● Overcollection and mismanagement of data: Collecting more data than
necessary can heighten the risk of breaches and privacy violations, as well as
increase the complexity of data-management processes.
● Bias: The use of AI can introduce biases into decision-making processes,
leading to unfair treatment of, and discrimination against, individuals based on
their data.
● Intrusive surveillance: When data is used unethically for intrusive
surveillance of individuals’ personal life or for behaviour profiling, parliament
runs the risk of both legal action and reputational damage.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

119
Guidelines for AI in parliaments

Data governance:
Parliamentary context
About this sub-guideline
This sub-guideline is part of the guideline Data governance. Refer to the main
guideline for context and an overview.

This sub-guideline examines the importance of data governance and provides an

overview of related roles and responsibilities within parliament.

Why data governance matters

Data governance is crucial for coordinating efforts to enhance data quality and
privacy. It encompasses policies, roles, responsibilities, processes and technology
aimed at improving data quality and creating an optimal data environment.

Data governance provides guidance, assurance and support to transform

parliaments into data-driven organizations through the following measures:

● Clearly defining roles and responsibilities for data management to ensure

efficiency
● Introducing common rules, guidelines and practices for consistent data
management across the organization
● Ensuring data security and protecting privacy
● Promoting easy access to, and usability of, data to support informed decision-
making
● Maintaining the accuracy and integrity of data throughout its life cycle
● Ensuring compliance with relevant laws and regulations in data management
practices
● Providing high-quality data to inform policymaking and strategic planning
● Fostering a data-centric culture that emphasizes the importance of privacy,
data quality and data governance across all parliamentary units and
processes

120
Guidelines for AI in parliaments

Roles and responsibilities in data governance

Figure 1 provides an overview of roles and responsibilities in data governance.
Further discussion can be found in the remainder of this section.

Figure 1: Roles and responsibilities in data governance

Data owner
Every piece of data should have a data owner, who is in charge of decisions
regarding data protection, data storage, data classification, data access, data
formats, metadata, and all improvements necessary to make the data useful for
parliament’s needs.
Data steward
Data owners designate a data steward to support them with managing data quality,
ensuring compliance with policies, and facilitating communication between data
owners and users.
Establishing data ownership
Practices for establishing data ownership can include the following:

● Establish clear ownership roles, assigning specific individuals or teams as

data owners for each data set. These owners are responsible for the
accuracy, integrity and security of the data throughout its life cycle.
● Establish data stewardship roles to support data owners.

121
Guidelines for AI in parliaments

● Create ownership policies that outline the responsibilities and expectations of

data owners, including practices for data creation, maintenance, access and
disposal.
● Implement role-based access controls to ensure that only authorized
individuals can access and modify data.
● Conduct regular audits to verify that data ownership policies are being
followed and that data quality standards are maintained. Use audit findings to
make continuous improvements.
● Provide training programmes for data owners and stewards to ensure they
understand their roles and responsibilities.
● Encourage collaboration between data owners, data stewards and users to
ensure that data management practices are aligned with parliament’s goals.
Establish forums or committees for discussing data-related issues and making
collective decisions.
● Implement monitoring tools to track data quality and ownership compliance.
Generate regular reports to keep stakeholders informed of the status of data
governance initiatives.
● Define data life cycle management practices, covering the entire process from
creation to disposal. Ensure that data owners are responsible for overseeing
this process.
Responsibilities of data owners
The responsibilities of data owners are as follows:

● Appoint one or more data stewards to support them in their role and to
facilitate the implementation of the data ownership guidelines.
● Allocate resources to information ownership objectives.
● Apply operational guidelines and procedures for information ownership.
● Establish and measure data performance metrics and communicate about
actual data performance.
● Establish future data requirements based on strategies and business trends.
● Position and manage data as a corporate asset.
Responsibilities of data stewards
The responsibilities of data stewards – some from business units, others from the IT
unit – are as follows:

● Deploy and implement information ownership programmes, operational

guidelines and procedures.
● Support audits of information ownership processes.
● Upgrade and develop information ownership processes.
● Provide intensive data-quality training to parliamentary staff.
● Help parliamentary staff to identify and solve data-related issues and
problems.
● Determine and interpret trends in data quality.
● Support data-quality improvement efforts.
● Serve as first-line support to help users solve data accuracy issues, data
definition issues and data usage issues.

122
Guidelines for AI in parliaments

Responsibilities of the chief information officer (or equivalent)

Parliament’s chief information officer (or equivalent), acting as the data custodian,
has overall responsibility for the following aspects:

● Implement and maintain the infrastructure needed to deliver data from its
point of capture and storage to a point of need.
● Manage the availability of systems to access, retrieve and manipulate data.
● Ensure the integration and consistency of data across multiple applications
and sources.
● Ensure that backup and recovery procedures are in place to prevent data
loss.
● Secure data access and provide up-to-date solutions to protect against
malicious code.
● Implement an IT helpdesk function for the following purposes:
○ Logging data issues
○ Escalating data issues to the appropriate information owner
○ Connecting users with second-line support to help interpret data fields
and/or information within data fields
○ Granting controlled access to data to authorized users
○ Optimizing operational efficiency and effectiveness
○ Monitoring and reporting data issues
Responsibilities of parliamentary users
Parliamentary users of AI-based systems have the following responsibilities:

● Participate in data-improvement initiatives.

● Enhance their job skills related to improved data quality.
● Capture data and utilize it in parliament’s processes in order to optimize
operational efficiency and effectiveness.
● Monitor and report data issues to the data owner and/or the chief information
officer.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

123
Guidelines for AI in parliaments

Data governance: Data

management for AI
systems
About this sub-guideline
This sub-guideline is part of the guideline Data governance. Refer to the main
guideline for context and an overview.

This sub-guideline focuses on effective data management, which is a crucial aspect

of implementing AI systems in parliaments. It emphasizes that staff must understand
the key steps for establishing effective data management practices, including
creating governance policies, managing metadata, ensuring data quality and
protecting personal information. By following these recommendations, parliaments
can build a solid foundation for trustworthy AI systems, enhancing their ability to
make data-driven decisions and operate efficiently.

Introduction
Parliaments that are planning to implement data management for AI systems should
consider the following prerequisites, which are explained in more detail below:

● Establishing a corporate data governance programme

● Establishing a data governance policy
● Establishing a communication plan
● Implementing a metadata management process
● Implementing a data quality process
● Implementing a personal data protection process
Establishing a corporate data governance programme
In order to put in place an institutional data management programme to leverage AI
systems, parliaments must establish a corporate data governance programme with
clear roles and responsibilities, and adhere to general rules for executing data
management processes. A corporate approach, aligned with parliament’s AI
strategy, is the most effective way to ensure that data within future AI systems is
compliant and managed effectively.

124
Guidelines for AI in parliaments

Establishing a data governance policy

The first concern in any data governance implementation is to develop and publish a
data governance policy that outline roles and responsibilities and determines the
rules for the corporate data management processes.

In general, a data governance policy covers the following aspects:

● The organizational structure related to data governance

● The roles that will interact with data governance processes
● The competencies or job descriptions expected for each role
● The relevant data governance processes
Establishing a communication plan
Once the data governance policy in place, the next recommended step is to draw up
a communication matrix showing, in detail, the interactions determined by the policy,
encompassing the main roles and their responsibilities.
Implementing a metadata management process
A metadata management process allows parliament to become acquainted with its
own data assets, which is crucial for data governance.

Parliaments should identify relevant information to be captured as metadata – based

on the goals established by the organization’s corporate strategy – as well as the
specific information considered useful for data management and data description.

Below are some examples of metadata that could be captured for parliament’s
corporate data assets:

● Title (e.g. name of the bill or legislation + issue date)

● Description (e.g. date on which the bill or legislation was presented to the
parliamentary board)
● Data owner (e.g. “Secretary of the Parliamentary Board”)
● Data steward (e.g. “Protocol Registration Officer”)
● Date format (e.g. “dd/mm/yyyy”)
● Information systems (e.g. “Protocol Registration System”)
● Main source (e.g.: “LegislationBills_DB”)
● Personal data (“Yes”/“No”)
● Sensitive data (“Yes”/“No”)

As Figure 1 below shows, having a clearly identified metadata repository is crucial

for understanding aspects such as the following:

● What the correct meaning of each data item is

● Who the data owner is
● Who the data steward is
● Whether the data is sensitive
● Which processes depend on the data

125
Guidelines for AI in parliaments

Figure 1: Structure of a metadata repository

Parliaments are advised to undertake continuous maintenance activities – such as

frequent metadata review, validation and updating – in order to ensure that the
metadata is precise, consistent and up to date.
Implementing a data quality process
The purpose of a data quality process is to ensure that data is managed in
accordance with the rules laid down in the data governance policy. The main
activities in this process are as follows:

● Data profiling:
○ Analysing data structure and contents
○ Identifying patterns, inconsistencies and anomalies in data
● Data quality requirements definition:
○ Establishing quality metrics and criteria (precision, completeness,
consistency, uniqueness)
○ Defining business standards and rules to guarantee data compliance
● Data validation:
○ Applying business rules to validate data precision and consistency
○ Verifying whether the data meets the defined requirements
● Data cleansing/correction:
○ Fixing or removing incorrect, incomplete or duplicated data
○ Standardizing data formats
● Data integration:
○ Combining data from different sources and ensuring it remains
consistent and correct
○ Solving data conflicts and eliminating duplications
● Data enrichment:
○ Incorporating additional information in order to increase data
usefulness and completeness
● Data-quality monitoring:
○ Implementing continuous processes to monitor data quality
○ Using dashboards and reports to track data quality rates

126
Guidelines for AI in parliaments

Implementing a personal data protection process

The purpose of a personal data protection process is to ensure that parliament
complies with privacy and data protection regulations, giving data subjects the
necessary confidence to trust the institution with their personal data. The process
dictates and influences how personal data is handled throughout its entire life cycle,
encompassing the relevant strategies, skills, people, processes and tools.

The main steps in implementing a personal data protection process are as follows:

● Appointing a data protection officer

● Aligning the process with the expectations of senior parliamentary managers
● Assessing the maturity of parliament’s existing corporate data protection
arrangements
● Adopting data security measures to raise this level of maturity
● Establishing an organizational structure for the governance of personal data
protection
● Implementing a personal data inventory
● Reviewing contracts related to the processing of personal data
● Preparing a personal data protection impact report
● Establishing terms and conditions for personal data protection
● Implementing an incident management process
Formalizing existing governance processes
Parliaments that already use data will, to some degree, have existing data
governance and data management processes in place. Rather than creating a
burdensome list of new responsibilities for business stakeholders, it is advisable to
try to match AI-related tasks to existing job routines.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

127
Guidelines for AI in parliaments

Security management
Audience
This guideline is intended for senior technical staff involved in the development and
implementation of AI systems. Some of the material it contains may also be relevant
to senior parliamentary managers looking to gain a better understanding of technical
issues relating to security.

About this guideline

This guideline addresses the protection of AI systems from a wide range of threats
and risks, outlining a set of practices for ensuring the confidentiality, integrity and
availability of AI systems and data in parliaments.

AI systems are being deployed in many different ways, always with the aim of
helping professionals increase their productivity. Parliaments are undoubtedly going
to follow this trend, aiming for faster processes that accelerate democracy without
harming or hurrying debate. Processes are also expected to be also safer, since
parliaments are potentials target for national and international interest groups.

Why security management matters

As the use of AI increases, so does the risk to organizations using this technology.
AI systems are associated with a range of security issues, such the inference of data
– sometimes sensitive data – used in the training process, the alteration of such
data, and the use of a particular prompt – wilfully or not – that could lead the AI
system to reach a wrong or unexpected conclusion. All of these issues and more
must be addressed before the AI system can be handed over to users.

Moreover, some AI behaviours could have a significant negative impact on an

organization’s public reputation. This means that AI systems can only be deployed
after – at the very least – a basic risk assessment demonstrating that risks are low or
controlled, and that the benefits outweigh these risks.

The deeper the knowledge someone has about AI, the easier it will be for this person
to come up with a possible way of misleading the system and turning a breakthrough
technology into a personal weapon to threaten different actors.

Moreover, even organizations that do not use AI models and systems are at risk,
because criminals are already using AI in an attempt to increase the success rate of

128
Guidelines for AI in parliaments

their attacks. However, security considerations are especially important for

organizations that do use AI in their own systems, since these models are prone to
new types of attacks.

Considering the rise in cyberattacks, which surged after the COVID-19 pandemic,
and the increasing use of AI models, which are the new “holy grail” of technology,
overcoming AI threats is an important part of an organization’s cybersecurity plan.

Cybersecurity management in a parliamentary context

Cyberattacks are a growing concern as parliaments increase their reliance on
internet-enabled connectivity – whether cloud-based servers, external systems or for
users. Effective cybersecurity management is therefore critical for avoiding such
attacks or minimizing their impact.
For further discussion of this subject, refer to the sub-guideline Security
management: Parliamentary context.

Cybersecurity threats to AI systems

AI systems learn from the data they are fed and then apply models to help them
make decisions, generate new content or do anything else they are programmed to
do.
Just as parliaments must ensure that data is valid and of high quality, they must also
ensure that there are no opportunities for attackers to exploit inputs into AI systems
in order to corrupt and manipulate the data, modelling and outputs.

Attacks can occur in any phase, from data preparation through to AI system
development, deployment and operation (for further discussion of this subject, refer
to the guideline Systems development). As a result, the entire AI system life cycle
should be properly supervised in order to minimize unexpected behaviours.
For further discussion of this subject, refer to the sub-guideline Security
management: Threats.

Good practices for implementing AI-focused cybersecurity

Most types of attacks can be avoided or minimized by implementing good practices.
Nonetheless, some attacks specifically targeting AI systems require specific
countermeasures.

For further discussion of this subject, refer to the sub-guideline Security

management: Good practices.

Main considerations when implementing AI-focused

cybersecurity controls
Based on the main security frameworks, parliaments should gradually implement
controls in the following four areas, according to their specific structure, needs and
threat risks:

• Technical controls
• Organizational controls

129
Guidelines for AI in parliaments

• Human controls
• Physical controls

Together, measures across these four areas enable parliaments to enhance the
protection of their AI systems.

For further discussion of this subject, refer to the sub-guideline Security

management: Implementing cybersecurity controls.

Find out more

● Athalye A., N. Carlini and D. Wagner: Obfuscated Gradients Give a False
Sense of Security: Circumventing Defenses to Adversarial Examples
● Deloitte: Impact of COVID-19 on Cybersecurity
● Federal Bureau of Investigation (FBI): FBI Warns of Increasing Threat of
Cyber Criminals Utilizing Artificial Intelligence
● Hurst A.: NCSC releases guidelines for secure AI development
● International Monetary Fund (IMF): Rising Cyber Threats Pose Serious
Concerns for Financial Stability
● Kurakin A., I. Goodfellow and S. Bengio: Adversarial Examples in the Physical
World
● Kurakin A. and others: 9 Common Types of Attacks on AI Systems
● Norwegian National Security Authority (NSM): NSM ICT Security Principles
● Open Worldwide Application Security Project (OWASP): OWASP Machine
Learning Security Top Ten
● Oseni A. and others: Security and Privacy for Artificial Intelligence:
Opportunities and Challenges
● Saleous A. and others: COVID-19 pandemic and the cyberthreat landscape:
Research challenges and opportunities
● Schneider S. and others: Designing Secure AI-based Systems: a Multi-Vocal
Literature Review

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

130
Guidelines for AI in parliaments

Security management:
Parliamentary context
About this sub-guideline
This sub-guideline is part of the guideline Security management. Refer to the main
guideline for context and an overview.

Why cybersecurity is important for AI

Parliaments’ growing reliance on internet-connected systems, including AI, increases
existing risks and introduces new ones, ranging from the manipulation of legislative
processes to the disruption of administrative tasks. For instance, cyberattacks could
alter the content of bills, influence voting outcomes or compromise AI-driven
productivity tools. As parliaments adopt AI to manage public participation at scale,
these systems may become targets for groups seeking to manipulate democratic
processes.

Effective cybersecurity strategies and management practices help to maintain a

secure digital environment that protects the integrity of parliamentary operations –
especially as AI usage continues to grow. It is key to mitigating risks, ensuring that AI
systems enhance rather than compromise parliamentary functions.

AI-related security considerations for parliaments

For AI systems, risks extend beyond traditional concerns, requiring specific
consideration:

● Integrity of the legislative process: Cyberattacks could manipulate AI

systems to alter bills, skew voting results or introduce biased information,
potentially undermining the democratic process.
● Administrative efficiency: While AI can enhance productivity in
administrative tasks, it also introduces new vulnerabilities. Attacks on AI-
driven systems could disrupt resource allocation, budget management and
other critical operations.
● Public participation: Where parliaments implement AI to manage large-scale
public input, these systems can become targets for manipulation, potentially
distorting the representation of public opinion.

131
Guidelines for AI in parliaments

● Data protection: AI systems often require vast amounts of data, including

sensitive information. Ensuring the security and privacy of this data is crucial.
● AI model integrity: Attacks could target the AI models themselves, potentially
introducing biases or altering decision-making processes without detection.
● Disinformation campaigns: AI systems used for information analysis and
dissemination could be exploited to spread disinformation within parliamentary
networks or to bias research.
● Autonomous system vulnerabilities: As parliaments adopt more
autonomous AI systems, ensuring that such systems cannot be hijacked or
misused becomes critical.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

132
Guidelines for AI in parliaments

Security management:
Threats
About this sub-guideline
This sub-guideline is part of the guideline Security management. It should be read in
conjunction with the sub-guideline Security management: Good practices. Refer to
the main guideline for context and an overview.

Background
AI systems learn from the data they are fed and then apply models to help them
make decisions, generate new content or do anything else they are programmed to
do.

For this reason, it is essential that they are fed with correct, clean, unbiased data (for
further discussion of this subject, see the guideline Generic risks and biases). Any
change to that data, whether intentional or not, may lead to unexpected
consequences, results (e.g. a budget management system) or behaviour (e.g. an
autonomous car). The end result is akin to teaching improper behaviour or giving
wrong information to a child throughout their life.

It is also important to pay attention to the system that will receive user input, send
this input to the AI system and then return the result to users. In some cases,
attackers can exploit this chain of communication.

Attacks can occur in any phase, from data preparation through to AI system
development, deployment and operation (for further discussion of this subject, see
the guideline Systems development). As a result, the entire AI system life cycle
should be properly supervised in order to minimize unexpected behaviours.

Types of attacks
There are nine common types of attacks on AI systems:

● Adversarial attacks
● Evasion attacks
● Transfer attacks
● Data poisoning attacks

133
Guidelines for AI in parliaments

● Model inversion attacks

● Membership inference attacks
● Distributed denial of service (DDoS) attacks
● Data manipulation attacks
● Misuse of AI assistants attacks

Each of these is discussed in turn below.

Adversarial attacks
An adversarial attack involves an attacker manipulating the input data of an AI
system so that it produces inaccurate, unexpected or wrong responses. The more
information the attacker has about the AI system (especially about the AI model
being used), the easier the attack will be. This type of attack often targets AI image
recognition systems, making the system incorrectly recognize an image – such as an
image of a dog being recognized as a tiger or, worse still, a person being recognized
as an animal. The subtle changes in the image are not easily recognizable to the
human eye, which makes the issue even harder to solve in certain circumstances.
From a parliamentary perspective, an attacker could target a voting system that uses
facial recognition technology, causing it to incorrectly allow the attacker to vote as an
MP.

Figure 1: Examples of adversarial attacks

Evasion attacks
An evasion attack is considered to be a specific type of adversarial attack. In this
case, the attacker intentionally crafts the input data to evade AI detection or
classification. For example, the attacker may change the way an unsolicited email is
written to avoid being detected by the AI anti-spam system. This can lead to a
malicious message getting through to a regular user, who might click a fake link and
allow an attacker to gain access to an organization’s network.

134
Guidelines for AI in parliaments

Figure 2: Diagram of an evasion attack

Transfer attacks
A transfer attack occurs when an attacker uses adversarial-type attacks developed
for one model and to deceive other models. The consequences are the same as for
an adversarial attack.
Data poisoning attacks
In a data poisoning attack, the attacker adds data to the data set used to train an AI
model. The model will learn from incorrect information, leading it to make wrong
decisions. For instance, a system could wrongly diagnose a healthy patient as
having a deadly cancer – or, worse still, wrongly diagnose a patient with cancer as
being healthy, preventing the person from receiving proper treatment. In a
parliamentary context, a proposal could be forwarded to the wrong committee for
discussion.

Figure 3: Diagram of a data poisoning attack

Model inversion attacks

In normal circumstances, an AI model will learn from the input data and produce an
output on this basis. The aim of a model inversion attack is to use the output as a
way to infer the input data. By doing so, the attacker may gain access to confidential
or private information that was used to train the model. For instance, the attacker

135
Guidelines for AI in parliaments

could get the result of a specific patient’s blood test, or access other, more sensitive
data. In a parliamentary context, if an AI model is trained on secret voting data, an
attacker may be able to obtain information about how an MP voted.

Figure 4: Diagram of a model inversion attack

Membership inference attacks

With a membership inference attack, an attacker aims to find out if individual data
records were used to train the AI model. As with a model inversion attack, the
attacker may be able to infer sensitive information. For instance, an AI system
trained on financial information could leak an individual’s financial history.

Figure 5: Diagram of a membership inference attack

Distributed denial of service attacks

A DDoS attack involves an attacker flooding a system – including an AI system –
with an excessive number of requests. The aim is to cause the system to stop
working, preventing any response or, at least, making it so slow that users will not
engage. This often leads to financial losses and reputational harm for the
organization running the service. In a parliamentary context, an attacker could bring
down the AI chatbot designed to answer questions from citizens during a plenary
session discussing a theme with broad public support.

136
Guidelines for AI in parliaments

Figure 6: Diagram of a DDoS attack

Data manipulation attacks

In a data manipulation attack, the attacker changes the input data (often slightly) in
an attempt to generate inaccurate predictions. For instance, an AI system that would
normally easily identify a case of fraud may deem such a case to be a regular
transaction if the attacker makes minor changes to the input data.
Misuse of AI assistants attacks
The use of AI assistants (such as chatbots or applications built into everyday
communication devices) is increasing as these systems become more advanced. It
is therefore essential to ensure that a well-thought-out process is in place to select
and sanitize the training data set, avoid bias, select the right model for the
application, deal with security issues during development, and monitor the
application’s usage.

In a parliamentary context, a clerk might use an AI assistant (such as the one pre-
installed on their mobile phone) to assist them with daily tasks. However, the results
this AI assistant produces may be biased according to the clerk’s political or other
personal preferences.

The problem may become more serious if the AI assistant is attacked by a group
with particular preferences on any matter parliament may be discussing, especially if
this matter is sensitive. Likewise, if parliament decides to develop an AI assistant to
support citizens on legislative matters, it must be considered a target for
cyberattacks – not least because its audience is often unknown. In this case,
prompts need to be at least sanitized prior to their submission as an input to the AI
model, in the same way as inputs to any other AI-enabled system.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

137
Guidelines for AI in parliaments

Security management:
Good practices
About this sub-guideline
This sub-guideline is part of the guideline Security management. It should be read in
conjunction with the sub-guideline Security management: Threats. Refer to the main
guideline for context and an overview.

Good practices for countering threats to AI systems

Most types of attacks can be avoided or minimized by implementing good security
practices. Nonetheless, some attacks targeting AI systems require specific
measures. Recommended countermeasures are given below for the following types
of attacks:

• Adversarial attacks
• Evasion attacks
• Transfer attacks
• Data poisoning attacks
• Model inversion attacks
• Membership inference attacks
• Distributed denial of service (DDoS) attacks
• Data manipulation attacks
• Misuse of AI assistants attacks
Adversarial attacks
Countermeasures for adversarial attacks, especially those targeting image
recognition systems, include the following:

• Use trickier examples in the training phase. For instance, show the AI model
lots of slightly altered images so that it learns not to be fooled by them.
• Add a little randomness (technically known as “noise”) to the images used in
the training data set. That way, the model will learn to focus on the important
parts of the image, not just on the small details that can be easily changed.
• Use stronger models: design the model so that it looks at the big picture (like
a person’s overall shape) rather than just focusing on small details.

138
Guidelines for AI in parliaments

Evasion attacks
Countermeasures for evasion attacks include the following:

• Choose strong models that are less likely to be fooled by slightly altered
inputs.
• Check or validate the input data system to ensure that it is clean and as
expected. This can help to catch any abnormal or malicious inputs before they
cause harm.
• Train the model using examples of these tricky inputs so that it learns to
recognize and handle them properly.
• Regularly monitor how the model performs and update it to handle new types
of attacks as they are discovered.
• If possible, and provided that the benefits outweigh the increased cost, use
multiple models in combination so that if one model is fooled, the other
models can still catch the problem.

Transfer attacks
Countermeasures for transfer attacks include the following:

• Train the model on a wide variety of data. This reduces the chances that an
attack crafted on another model will work on the models that parliament is
using.
• As with evasion attacks, use multiple models to make decisions – provided
that the benefits outweigh the increased cost.
• During training, expose the model to adversarial examples (small, intentionally
crafted changes in input designed to fool the model). This helps the model
learn to recognize and defend against such attacks.
• If feasible, frequently update and retrain the model with new data. This can
help to close any vulnerabilities that might be exploited in transfer attacks.
• Use techniques designed to make the model more resistant to attacks, such
as smoothing or noise injection during training.

Data poisoning attacks

Countermeasures for data poisoning attacks include the following:

• Take care over who has access (physical or logical) to the training data set,
enforcing robust user permissions.
• Carefully check the data before using it to train the model, including ensuring
that the labels and data make sense. Proper sanitization is important for
getting rid of data that may negatively impact the learning process.
• Evaluate the machine learning algorithms and check if they are designed to
be less sensitive to corrupted data.
• Monitor the model’s performance after deployment in order to detect any
unusual behaviour that might suggest it was trained on poisoned data.

Model inversion attacks

Countermeasures for model inversion attacks include the following:

139
Guidelines for AI in parliaments

• Differential privacy: This technique adds a small amount of random noise to

the data or to the model’s outputs. The noise is carefully calibrated so that it
does not significantly affect the model’s performance but makes it much
harder for an attacker to extract precise information about individual data
points.
• Data minimization: Only collect and use the data that is absolutely necessary
for the model. The less sensitive data that is included in the training set, the
less risk there is of exposing private information.
• Regularization: This technique can make the model less sensitive to specific
data points, which reduces the risk of a successful inversion attack. It forces
the model to generalize better, making it harder for an attacker to reconstruct
specific inputs.
• Limitation of model access: Restrict who can query the model and how many
queries they can make. If an attacker can only make a limited number of
queries, it becomes more challenging for them to gather enough information
to perform a model inversion attack. A robust account management system
plays a key role in defending against this type of attack.
• Query auditing and anomaly detection: Monitor the queries made to the model
and look for unusual patterns that might indicate an attack. If suspicious
activity is detected, further queries from that source can be blocked.
• Adversarial training: Train the model with adversarial examples (inputs
designed to trick the model) to make it more robust against various types of
attacks, including model inversion attacks.

Membership inference attacks

Countermeasures for membership inference attacks include the following:

• Regularization: This approach makes the model less confident in its

predictions, which makes it harder for an attacker to tell if a specific data point
was included in the training data set.
• Differential privacy: This method involves adding noise to the data or to the
model’s predictions to make it difficult for an attacker to distinguish between
data that was included in the training set and data that was not.
• Model distillation: This method trains a simpler model to mimic the behaviour
of a more complex model. The simpler model is less likely to give away
specific information about the training data.

Distributed denial of service attacks

Countermeasures for DDoS attacks include the following:

• Use a content delivery network (CDN) to distribute the service across servers
in different locations.
• Install a web application firewall to detect and block malicious traffic, including
a DDoS attack, before it reaches the servers running the AI system.
• Increase server capacity. While this will not solve the problem itself, it will
make it more difficult for the attacker to crash the entire system.
• Use externally sourced DDoS protection services to detect and mitigate DDoS
attacks. These services can automatically identify and block malicious traffic,
keeping a system running smoothly.

140
Guidelines for AI in parliaments

• Limit the number of requests a single user can make within a given time
frame.

Data manipulation attacks

Countermeasures for data manipulation attacks include the following:

• Enforce the use of strong passwords and multi-factor authentication (MFA) to

minimize the risk of unauthorized access.
• Regularly update all software (after proper testing in a controlled environment,
as updates can sometimes introduce more problems into the system).
• Keep checking for the most recently discovered vulnerabilities and be ready to
fight against them.
• Encrypt all data, and especially data that can lead to privacy issues. Thus,
even if attackers are able to access the data, they will not be able to read it
without the decryption key (or, at least, it will take a very long time before they
can read it).
• Always back up all data properly to avoid data hijacking (which can make the
service suddenly stop, negatively impacting parliament’s image among
citizens and requiring the payment of a high ransom).
• Limit access to sensitive data, and monitor for unusual and suspicious activity
such as changes to data or unauthorized logins.

Misuse of AI assistants attacks

Countermeasures for misuse of AI assistants attacks include the following:

• Educate staff to exercise caution as to the information they provide to both

personal and enterprise AI assistants.
• Where possible, avoid using AI assistants for AI project purposes.
• If this is not possible, check the privacy settings, as there may be an option to
limit the data the AI assistant can access and store.
• Always review the AI assistant’s activity logs (if available) for any unusual
behaviour.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

141
Guidelines for AI in parliaments

Security management:
Implementing
cybersecurity controls
About this sub-guideline
This sub-guideline is part of the guideline Security management. Refer to the main
guideline for context and an overview.

Types of cybersecurity controls

Based on the main security frameworks, parliaments should gradually implement
controls in the following four areas, according to their specific structure, needs and
threat risks:

• Technical controls
• Organizational controls
• Human controls
• Physical controls

Together, measures across these four areas – which are discussed in turn below –
enable parliaments to enhance the protection of their AI systems.

Technical controls
Technical controls are measures and processes designed to protect AI systems,
data and algorithms from unauthorized access, tampering and exploitation.

Network security
• Use firewalls to segment the network into different zones based on security
requirements, implementing strict access controls between zones.
• Consider deploying intrusion prevention systems (IPS) to detect and block
malicious activities in real time as soon as possible.
• Ensure that all communication channels – including data transfers, model
updates and application programming interface (API) calls – are encrypted in
order to protect data in transit.

142
Guidelines for AI in parliaments

• Use robust encryption protocols such as transport layer security (TLS) and
secure sockets layer (SSL).
• Use virtual private networks (VPNs) to secure remote access to AI systems
and data.

System security
• Regularly update and patch all software, operating systems and AI algorithms
to protect against known vulnerabilities.
• Be careful to test patches in a controlled environment before deploying them
to production systems to ensure they do not introduce new vulnerabilities or
cause system instability.
• Deploy robust and regularly updated antivirus and anti-malware solutions on
all endpoints, including servers, workstations and mobile devices.
• Enable real-time protection features to detect and block malware and other
threats as they occur.

Data security
• Make sure training data sets are reliable and keep these data sets secure, as
they are one of the most important assets of the AI system.
• Use data from reliable and verified sources to ensure the authenticity and
accuracy of the information.
• If using third-party data, ensure that the data provider has undergone rigorous
security audits.
• Remove personally identifiable information from data sets to ensure privacy.
• If this is not possible, replace sensitive data with pseudonyms that can be
traced back to the original data only through secure means.
• Encrypt data stored in databases, and in cloud-storage and backup systems,
using strong encryption algorithms.
• Use encryption protocols such as TLS or SSL to protect data when it is
transmitted between systems or users.
• Pre-process the data to apply sanitization using a variety of methods such as
data anonymization, pseudonymization and data masking (for further
discussion of this subject, see the guideline Data management).
• Where necessary, establish data-sharing agreements and protocols with
trusted partners (such as other parliaments) to ensure the integrity and
security of shared data sets, and use secure communication channels when
sharing data or collaborating with external parties.

Application security
• Implement systems development best practices to resolve known
vulnerabilities and be ready for unknown ones (for further discussion of this
subject, see the guideline Systems development).

Organizational controls
Organizational controls focus on internal policies, procedures and practices.

143
Guidelines for AI in parliaments

Security policy development

• Develop and implement security policies covering data protection, user
behaviour, system access and incident response (for further discussion of this
subject, see the guideline Systems development).

Security risk management

• Assess the inherent risks involved in all projects in order to maximize the
chances of success (for further discussion of this subject, see the guideline
Risk management).

Incident response
• Establish well-defined procedures at a time when the system is not under any
real threat. That way, the team can think, discuss and come up with a
response plan that is not rushed by the imminent danger.

Human controls
Humans are one of the weakest links in the chain of an AI system, or indeed of any
system. Human controls focus on managing this risk through a range of different
measures and procedures.

Training and awareness

• Provide security training to the AI team, and ensure that security practices are
applied in all stages of the AI system development process within the
organization (for further discussion of this subject, refer to the guideline
Training for data literacy and AI literacy).

Access management
• Develop a strict role-based access model, implementing the principle of least
privilege (PoLP) in order to minimize the risk of unauthorized access and data
breaches.

Accountability
• Monitor security incidents and suspicious activities, and implement clear
channels for reporting such incidents and activities (for further discussion of
this subject, see the guidelines Ethical principles and Systems development).

Physical controls
Physical controls focus on protecting physical assets and infrastructure that support
AI systems from unauthorized access, damage or interference.

144
Guidelines for AI in parliaments

Facility security
• Ensure that only authorized people have physical access to the AI system.
• Implement at least two ways to allow access the computer room, and apply
proper visitor management to sensitive areas.

Environmental controls
• Ensure that facilities hosting IT hardware and staff are protected against fire.
• Install fire detection systems and have an evacuation plan in place.
• If possible, use a climate control system to keep all computers at appropriate
temperature and humidity levels.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

145
Guidelines for AI in parliaments

Risk management
Audience
This guideline is intended for senior parliamentary managers and parliamentary staff
involved in the development and implementation of AI-based systems. It also
provides a more detailed technical discussion for those involved in developing and
implementing AI-related projects.

About this guideline

This guideline provides guidance on AI risk management for parliaments. It
emphasizes the importance of continuous risk assessment throughout an AI
system’s life cycle, from project proposal to decommissioning, with a particular focus
on three stages: initial project authorization, the development phase and the
operational phase. The associated sub-guideline includes questionnaires to support
the identification, analysis and mitigation of AI-related risks.

Relevance of risk management to AI governance in

parliaments
There are certain risks associated with the use and development of AI systems in
parliamentary settings. Further discussion of these risks can be found in the
guidelines Generic risks and biases and Risks and challenges for parliaments.

Managing these risks is essential to ensure that AI systems are safe, ethical, fair,
private, trustworthy, transparent and compliant with regulations, as well as to ensure
respect for human autonomy and intellectual property rights. In summary, effective
AI risk management practices help parliaments to:

• protect their data and AI systems

• maintain business continuity
• safeguard their reputation
• prevent costly errors
• support responsible innovation

Planning to implement an AI risk management process

An AI risk management process needs to be aligned with parliament’s culture and
structure. As stakeholders are at different hierarchical levels, appropriate language

146
Guidelines for AI in parliaments

must be used in assessment questionnaires and periodic reports in order to prevent

delays or incorrect interpretations.

During the development phase, adjustments to traditional risk management

processes, embedded in project management, can simplify and speed up the
implementation of AI risk management. In this case, the process will involve not only
AI project managers, but also IT committees, corporate committees and senior
decision makers.

During the operational phase, a partnership between business and IT teams

facilitates the integration of oversight practices and the collection of user feedback as
input for AI risk management. In this case, it is crucial to involve a risk management
team composed of people with AI skill and business staff in charge of AI-enabled
digital services.

Applying risk management processes to the AI life cycle

Risk management is a continuous cycle that permeates all phases of an AI system’s
life cycle, ensuring that risks are consistently identified, assessed, and managed or
mitigated – from inception to deployment and beyond.

A typical AI risk management process includes the following phases:

• AI risk assessment
• AI risk analysis
• AI risk treatment
• AI risk communication
• AI risk monitoring and review

These phases are discussed in turn below.

AI risk assessment
The aim of this first step is to identify what risks exist, to understand the operational
and other implications of these risks, and to decide how they should be managed or
mitigated.

This assessment exercise will typically be carried out using questionnaires for key
stakeholders – which, in itself, can be a useful mitigation method because it exposes
potential risks and increases awareness of them.

These questionnaires can be used in the following phases:

● Initial project authorization

● The development phase (prior to commissioning)
● The operational phase (through to decommissioning)

147
Guidelines for AI in parliaments

Figure 1: AI risk assessments throughout the AI system life cycle

Risk assessment for initial project authorization

The AI system life cycles starts with the presentation of a proposal for an AI project
to the relevant governance body (council, committee or unit), along with a completed
AI risk assessment questionnaire (Q1), which is used to gather information about the
project’s purpose, stakeholders, compliance, data, agreements, potential biases and
other factors.

Governance staff use the responses in the questionnaire to estimate the project’s
risks and benefits, and to determine, on that basis, whether authorization should be
given to add the AI project to parliament’s portfolio.

Risk assessment in the development phase

During the development phase, the Q2 questionnaire will inform the risk
management process, with the aim of reducing and mitigating AI risks such that the
output AI system is considered trustworthy for deployment. Unacceptable AI risks
can, however, lead to the project being interrupted at this stage.

Figure 2: Q2 and Q3 assessment questionnaires used in the risk management

process

148
Guidelines for AI in parliaments

Risk assessment in the operational phase

After the AI system has been deployed, it is necessary to keep monitoring the
system’s behaviour as well as changes in the variables considered in the AI system
life cycle, such as data characteristics, business rules and social considerations.

The third risk assessment questionnaire (Q3) can be used during this phase.
Similarly to the Q2 questionnaire, the risk score resulting from this third
questionnaire will inform the risk management process, which at this point aims to
reduce and mitigate AI risks in order to ensure that the system remains trustworthy.
AI risk analysis
Once the risks have been identified and assessed, the next step is to analyse these
risks in light of parliament’s AI policy and its risk appetite – often based on its
regulatory requirements and strategic objectives – in order to determine which risk(s)
require(s) treatment. All identified risks should then be ranked in order to identify
which require immediate attention and which should be monitored over time. All such
decisions should be made with the close involvement of relevant stakeholders.

Trade-offs between different risk treatment options also need to be evaluated at this
stage. For instance, by eliminating one identified risk, parliament could be at risk of
not achieving other strategic goals by also eliminating the option of using the AI
system in another, important way.
AI risk treatment
Once the identified risks have been analysed and prioritized, parliament should
develop and implement plans to manage them, possibly using one or more of the
following strategies:

● Avoid: Eliminate the activity that gives rise to the risk. For example,
parliament may decide not to implement an AI system, or even to abandon an
AI project, if the associated risks are deemed too high.
● Reduce: Take steps to reduce the likelihood of the risk occurring, or to
mitigate its impact if it does occur.
● Transfer: Transfer the risk to a third party, such as through insurance or by
outsourcing certain services to a company better equipped to manage the
risk.
● Accept: Accept the risk without taking any action to alter its likelihood or
impact. This is typically done when the cost of mitigating the risk exceeds the
potential damage, and when the risk is considered low enough to be
acceptable.

149
Guidelines for AI in parliaments

Figure 3: Risk treatment strategies

AI risk communication
Identified AI risks and associated management measures should be communicated
to relevant stakeholders throughout the AI system’s life cycle. During the
development phase, project managers and project office staff will provide regular
updates on risk status and treatment effectiveness as part of their usual remit.
Communication is equally important in the operational phase.
AI risk monitoring and review
During the operational phase, it is essential to continuously monitor and review AI
risks.

Oversight and feedback mechanisms, coupled with training programmes, help to

build a risk-aware culture and keep stakeholders informed.

Periodic audits and reviews should also be conducted to ensure compliance with AI
policy and regulations. All identified incidents and near-misses should be analysed in
order to identify root causes and improve risk management practices, with lessons
learned documented and policies updated accordingly.

Should any unacceptable AI risks arise, it may be necessary to remove the AI

system from operation.

Find out more

• Cheatham, B., Javanmardian, K., and Saman, H. (Undated). Confronting the
risks of artificial intelligence, available at [Confronting AI risks | McKinsey]
• National Institute of Standards and Technology (NIST) Special Publication
(SP) 800-30, Guide for Conducting Risk Assessments, available at [NIST SP
800-30 | NIST]
• Online Browsing Platform (OBP), ISO/IEC 27005:2022(en) Information
security, cybersecurity and privacy protection — Guidance on managing
information security risks, available at [ISO/IEC 27005:2022(en), Information
security, cybersecurity and privacy protection — Guidance on managing
information security risks]
• Tucker, B.A. (undated). Carnegie Mellon University, Advancing Risk
Management Capability Using the OCTAVE FORTE Proces, available at
[Advancing Risk Management Capability Using the OCTAVE FORTE Process
(cmu.edu)]

150
Guidelines for AI in parliaments

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

151
Guidelines for AI in parliaments

Risk management: Risk

assessment
questionnaires
About this sub-guideline
This sub-guideline is a part of the guideline Risk management. Refer to the main
guideline for context and an overview.
This sub-guideline provides sample risk assessment questionnaires that can be used
to support parliament’s risk management process in three phases of the AI system
life cycle:

• Initial project authorization

• The development phase
• The operational phase

Question 1: AI risk assessment for initial project

authorization
When business managers want to submit a proposal to run an AI project, through a
partnership with the IT unit, they focus on the problem to be solved using AI. At this
stage, both business managers and the IT unit should complete an initial
questionnaire, which could include the following questions, among others:
Purpose and stakeholders
• What is the business case for, and the problem to be solved by, the proposed
AI system?
• Which stakeholders would this project benefit? How would they be impacted?
• Are there any stakeholders (internal and/or external) that could be negatively
impacted by this project? If so, could these negative effects be mitigated or
compensated for?
Compliance
• Will this project conflict with parliament’s policies, or with any law or other
compliance rule?

152
Guidelines for AI in parliaments

Data and privacy

• Have the owners of the data that the AI system will use been identified?
• Will the AI system use internal data? If so, has internal authorization been
obtained to use parliament’s data for this purpose?
• Will the AI system use external data? Has parliament signed a memorandum
of understanding or other agreement with the organization(s) that own(s) the
data?
• Are any groups potentially underrepresented in the data?
• Will the AI system use personal data? Is there an agreement or arrangement
with appropriate safeguards in place?
• Will the AI system’s output be available to external users?
Copyright
• Should any of parliament’s data be protected by copyright?
• Are there any copyrights or contractual conditions that need to be respected?
Capacity-building and outsourcing
• What expertise is missing within parliament (if any) to support the
procurement, development or implementation of the AI system?
• Will this AI system be developed internally, purchased as a commercial
product or developed through outsourcing?

Question 2: AI risk assessment in the development phase

This second questionnaire is to be completed by the AI development team and
business staff during the development phase. Where an agile development method
is employed, the responses should be reviewed for each new version of the system,
informing the decision as to whether to continue or suspend – or even cancel – the
project. This questionnaire could include the following questions, among others:
Data and privacy
• Have the data owners authorized all actions regarding data access and
treatment?
• Is enough data available for the project?
• Is the data quality adequate for this project? If not, what data quality issues
have been identified?
• How will parliament improve data quality in relation to the identified issues?
• Will the use of personal data be restricted to the purposes for which it was
planned/authorized?
• Is it possible to keep people’s privacy sufficiently protected? Will it be possible
to re-identify the data subjects?
Bias and discrimination
• Are there any underrepresented data categories or potential biases in the
data set? If so, what issues have been detected and how are they being
mitigated?
• If generative AI is used, which hallucinations should be avoided or minimized
in this project?
• Will the AI system generate any classification of people’s behaviour?

153
Guidelines for AI in parliaments

Transparency
• How are the planning, modelling, evaluation, testing and deployment phases
scrutinized?
• Does the documentation use appropriate language for the target audience(s)?
• How do the documents demonstrate that the model addresses business
requirements?
• How do the documents demonstrate that the AI system is sufficiently
accurate?
• Is there any direct interaction between human end users and the AI system?
Are users explicitly informed that they are interacting with an AI system?
Safety and robustness
• Are there any weaknesses in the defined model?
• Are there any weaknesses in the testing phase?
• Is the AI system robust to potential failures and security attacks?
• Is there a deployment plan?
• Is there a rollback or disaster recovery strategy in place?

Question 3: AI risk assessment in the operational phase

Once an AI system is deployed in a live operating environment, it should be
continuously monitored by business and IT staff. At this stage, these staff should
complete a third questionnaire to ensure that changes in data, business rules, social
trends and the operating environment have been taken into account. This
questionnaire could include the following questions, among others:
Human autonomy and oversight
• Is the AI system subject to continuous performance monitoring?
• Has parliament established clear criteria for classifying acceptable and/or
unacceptable AI system behaviours? If so, is the AI system’s current
behaviour acceptable according to these criteria?
• Has parliament implemented a continuous process for collecting user
feedback on the AI system’s behaviour? If so, is the AI system’s current
behaviour acceptable according to user feedback?
• Has parliament identified any new variables (changes) in the AI system that
were not considered in the development phase?
Transparency
• Is the performance monitoring and user feedback collection process
effectively scrutinized?

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

154
Guidelines for AI in parliaments

Systems development
Audience
This guideline is intended for IT managers and staff, software engineers, data
scientists and technical project managers involved in designing, deploying and
maintaining AI systems in parliaments.

About this guideline

This guideline focuses on AI systems development within parliamentary contexts,
addressing the crucial intersection of technology and governance. It covers essential
aspects such as the AI system life cycle, external development frameworks,
deployment strategies and planning considerations. By emphasizing ethical
principles, risk mitigation and best practices, this guideline aims to support
parliaments in implementing AI solutions that enhance efficiency, transparency and
decision-making while maintaining integrity and public trust in the legislative process.

Why AI systems development matters

In the context of AI governance, an AI systems development process is a set of
practices designed to ensure that all AI projects solve the problems for which they
were planned and adhere to AI ethical principles. As an inherently operational
process, AI systems development should adhere to parliament’s AI policy, as well as
follow the institution’s data management and security management procedures.

In a parliamentary context, an AI systems development process is relevant to AI

governance as a way of reducing ethical and operational risks. Details of how it does
this are given below.
Preserving the privacy of data subjects
Personal data should be protected not only in the development phases, but also in
the AI system’s outputs.
Ensuring transparency throughout the development and maintenance
phases
AI systems often imply complexity, making their internal decision-making processes
opaque. For this reason, practices should be scrutinized in order improve
transparency throughout the AI system life cycle – from initial project planning until
the AI system is withdrawn from operation. This approach should make it easier to

155
Guidelines for AI in parliaments

explain the AI system’s outcomes and ensure that the development phases respect
parliament’s rules and are compliant with regulations.
Reducing biases and discrimination
Techniques to identify groups that are to be protected from biases should be applied
throughout the process, from planning to deployment. While AI systems are in
operation, continuous monitoring helps to minimize new biases not seen in the
development phase.
Creating accountability
Through systematic steps for planning, implementing, testing and improving,
practices are delegated and approved by key stakeholders, with individual roles and
responsibilities clearly defined. The AI system’s functionality should be documented
in such a way that it can be audited.
Improving robustness and safety
The systems development process should focus on improving robustness and
safety, through a system architecture that prioritizes cybersecurity and through
extensive testing.
Maintaining human autonomy
Humans should play a continuous verification role in order to ensure that the AI
system’s outputs are reliable, both during development and following deployment in
a live environment. This human oversight will ensure that the system continues to
adhere to the ethical principles considered in the project phase, and allows for new
ethical risks to be identified.
Guaranteeing regulatory compliance
AI systems are required to comply with various legal and regulatory requirements,
established both internally and within parliament’s country or region.

Systems life cycle and development frameworks

The AI systems life cycle is a sequential list of steps, practices and decisions that
drive the development and deployment of AI-based solutions. Having a well-defined
life cycle is vital for parliaments that are developing their own AI-based systems and
tools, as it provides a structured and systematic approach to building, deploying and
maintaining ethical AI technologies.

Within this context, there are an increasing number of external AI development

frameworks that parliaments can use. These consist of building blocks and
integrated software libraries that make it easier to develop, train, validate and deploy
AI solutions through a high-level programming interface.

For further discussion of systems life cycle and development frameworks, refer to the
sub-guideline Systems development: Systems life cycle and development
frameworks.

156
Guidelines for AI in parliaments

Deployment and implementation

When deploying and implementing AI systems and tools, parliaments need to
understand key aspects such as deployment strategies, common deployment cases
and critical planning recommendations, including topics such as stakeholder
engagement, pilot project initiation and the use of agile methods. Context should
also be given consideration, including parliamentary workflows, internal expertise
and opportunities for leveraging responsible AI tools.

For further discussion of the deployment and implementation of AI systems, refer to

the sub-guideline Systems development: Deployment and implementation.

For further discussion of software deployment patterns, refer to the sub-guideline

Systems development: Deployment patterns.

Find out more

• Data Science PM: “The GenAI Life Cycle”
• Data Science PM: “What is CRISP DM?”
• Data Science PM: “What is the AI Life Cycle?”
• Google: “Domain-specific AI apps: A three-step design pattern for specializing
LLMs”
• Inter-American Development Bank (IDB): Responsible use of AI for public
policy: Data science toolkit
• Microsoft: “Empowering responsible AI practices”
• Microsoft: “What is the Team Data Science Process?”
• Wikipedia: “Cross-industry standard process for data mining”

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

157
Guidelines for AI in parliaments

Systems development:
Systems life cycle and
development frameworks
About this sub-guideline
This sub-guideline is part of the guideline Systems development. Refer to the main
guideline for context and an overview.

This sub-guideline focuses on the systems life cycle and development frameworks
for AI-based systems in parliamentary contexts. It provides an overview of the AI
systems life cycle, highlighting its importance in ensuring structured and responsible
AI development.

This sub-guideline outlines the benefits of adopting a systematic life cycle approach.
It also offers guidance on evaluating external AI development frameworks, covering
aspects such as ease of use, community support, performance, model support and
deployment readiness.

This sub-guideline is intended to help parliamentary IT professionals make informed

decisions about AI development processes and tool selection, ultimately supporting
the effective and responsible implementation of AI technologies in legislative
environments.

The AI systems life cycle

The life cycle of an AI system is a sequential list of steps, practices and decisions
that drive the development of AI-based solutions. Having a well-defined life cycle is
vital for parliaments that are developing their own AI-based systems and tools, as it
provides a structured and systematic approach to building, deploying and
maintaining ethical AI technologies.

Specifically, adopting an AI systems life cycle approach offers the following benefits:

• Increased success rate: Following each essential step in the development of

an AI system improves the chances of project success.

158
Guidelines for AI in parliaments

• Risk reduction: Identifying potential issues early in the process helps to

mitigate risks and prevent costly setbacks.
• Improved efficiency and productivity: An organized project timeline makes
work smoother, ensuring that all team members understand their roles and
responsibilities at each stage.
• Enhanced quality: Completeness and rigour at each stage of the life cycle
lead to higher-quality AI systems.
• Better resource allocation: AI projects require significant resources, including
time, human expertise and computational power. Properly identifying and
balancing these resources ensures that they are used effectively throughout
the project.

External development frameworks

There are an increasing number of external AI development frameworks that
parliaments can use. These consist of building blocks and integrated software
libraries that make it easier to develop, train, validate and deploy AI solutions
through a high-level programming interface.

The natively pre-configured blocks and functions provided by these frameworks

speed up implementation time. By allowing developers to solve tasks by customizing
existing blocks without having to start from scratch, these frameworks also improve
productivity and algorithm quality. Moreover, using standard frameworks makes it
easier to integrate AI features with a great variety of application platforms and
domains.

In order to compare and evaluate different external AI development frameworks,

parliament must understand their characteristics and determine their suitability for its
workflows and business needs. It is advisable to review specific frameworks for
specific use cases, and to compare options through experimentation.

Key considerations for this decision-making process are detailed below.

Ease of use
• Documentation: Quality, clarity and comprehensiveness of the documentation
• Learning curve: How easy it is to start using the framework, including the
availability of tutorials and community support
• API design: Simplicity and intuitiveness of the API
Community and support
• Community size: The number of users and developers contributing to the
framework
• Support: The availability of forums, user groups and other support channels
• Updates: The frequency of updates, how many known issues and
vulnerabilities exist, and how actively the framework is maintained and
improved

159
Guidelines for AI in parliaments

Performance
• Speed: How quickly models can be updated for training and inference
• Scalability: The ability to manage large data sets and complex models, and
support for distributed training
• Optimization: Built-in features for optimizing and tuning model performance
and resource usage
Model support
• Model variety: The range of supported model types (neural networks, decision
trees, etc.)
• Pre-trained models: The availability and variety of pre-trained models that can
be fine-tuned or used out of the box
• Customization: Flexibility in defining and experimenting with custom models
and architectures
Tooling and integration
• Ecosystem: The availability of complementary tools for data preprocessing,
visualization and deployment
• Compatibility: Integration with data-handling libraries, visualization tools,
deployment platforms, etc.
• Interoperability: Support for importing/exporting models between different
frameworks
Deployment and production readiness
• Deployment options: Ease of deploying models to different environments
(cloud, edge, mobile)
• Serving: Support for model serving and inference in production settings
• Monitoring: Tools for monitoring model performance and detecting issues in
production
• Data protection regulations: Assurance that data classification, retention and
residency rules are followed
Licensing and cost
• Open-source versus proprietary: Whether the framework is open-source or
commercial
• Licensing terms: Any restrictions or requirements imposed by the licence
• Cost: The potential costs associated with using the framework, especially for
proprietary options
Hardware support
• GPU/TPU support: Compatibility with various hardware accelerators
• Distributed computing: Support for running on multiple GPUs or across a
cluster of machines
Extensibility
• Plugins: The availability of plugins or extensions for added functionality
• APIs for custom extensions: The ability to write custom extensions or
integrate third-party tools

160
Guidelines for AI in parliaments

Scalability
• Scaling up and out: Support for horizontal and vertical scaling (manual or
automatic)
• Performance: Load testing and simulations to measure the performance of the
framework
• Cost: The ability to set costing limits in the event that the system needs to
scale
Reproducibility
• Versioning: Tools for model and data versioning to ensure the reproducibility
of outcomes
• Experiment management: Support for tracking experiments and managing
their results
Security
• Security features: Built-in security features for safe deployment and model
usage
• Compliance: Compliance with industry standards and regulations

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

161
Guidelines for AI in parliaments

Systems development:
Deployment and
implementation
About this sub-guideline
This sub-guideline is part of the guideline Systems development. It can be read in
conjunction with the sub-guideline Systems development: Deployment patterns.
Refer to the main guideline for context and an overview.

This sub-guideline focuses on the deployment and implementation of AI systems and

tools within parliament. It provides essential insights for IT professionals and
decision makers involved in integrating AI solutions into parliamentary operations.

This sub-guideline covers key aspects such as deployment strategies, common

deployment cases and critical planning recommendations. By addressing topics like
stakeholder engagement, pilot project initiation and the use of agile methods, this
guideline aims to support parliaments in effectively and responsibly implementing AI
technologies to enhance legislative processes, improve efficiency and maintain
transparency. It emphasizes the importance of understanding parliamentary
workflows, building internal expertise and leveraging responsible AI tools throughout
the implementation process.

Deployment strategy
A structured and coordinated approach to AI systems deployment is essential to the
effective integration of these systems into various parliamentary applications and
workflows. These patterns of deployment reinforce good practices, helping to ensure
that AI deployments are scalable, robust and maintainable. For further discussion of
software deployment patterns, refer to the sub-guideline Systems development:
Deployment patterns.

Parliament’s deployment strategy will depend on the degree of task automation. The
various options are discussed below:

• Human-only: In this case, there is no automation. The task is carried out

manually by users without AI support.

162
Guidelines for AI in parliaments

• AI assistance: The task is performed mainly by users, possibly with

assistance and support from the AI system.
• Partial automation: The task is performed mainly by the AI system, which
produces suggestions for users.
• Full automation: The task is performed entirely by the AI system, without
human intervention.

Both AI assistance and partial automation are examples of “human-in-the-loop”

deployments (for further discussion of this topic, refer to the sub-guideline Ethical
principles: Human autonomy and oversight).
Deployment cases
The most common deployment cases and their characteristics are detailed below:

• New product or feature: In this case, a new AI-based product or feature is

introduced.
• Partial task automation: In this case, a task was previously done manually and
an AI algorithm is introduced to either automate this task or assist the user.
• Replacement of a previous AI system: In this case, a task was carried out via
a previous implementation of an AI system, and another AI system is
introduced to replace the previous one with a view to improving quality and/or
execution time.

In the above deployment cases, parliament should consider the following two basic
aspects:

• Gradually increase traffic with monitoring. It is advisable to avoid sending a lot

of production traffic to an algorithm that is still learning and is not yet fully
proven. It may be better to send it only a small amount of traffic, monitor it and
then progressively ramp up the amount of traffic.
• Ensure there is a possibility to roll back. It is advisable to have a contingency
plan in place to revert back to the previous, stable configuration in case the
new algorithm does not work as expected.

Planning the implementation of AI systems

Key recommendations for planning the implementation of AI systems within a
parliamentary context are outlined below.
Understand parliamentary processes
• Conduct a comprehensive analysis of existing parliamentary workflows
and challenges.
• Examine key processes such as legislative drafting, committee meetings,
voting procedures and public consultations.
• Map out the flow of documents, communications and decisions within
parliament.
• Understand how information is processed and identify things such as pain
points, redundancies and inefficiencies in current processes.
• Understand the current state of digital systems, databases and
communication tools used for parliamentary operations.

163
Guidelines for AI in parliaments

• Identify where AI can be integrated to enhance current systems or replace

outdated technologies.
• Identify and develop potential use cases to improve understanding of the
potential for AI solutions at both technical and business levels.
• Ensure that any AI implementation complies with existing laws and with
parliament’s rules (or identify where changes to these rules are needed).
Engage stakeholders
• Ensure that the following stakeholders are represented and involved
throughout the entire AI system life cycle:
o Business experts
o MPs, clerks and administrative staff, who will use the AI system for
legislative activities
o Legal experts, who will ensure that the AI implementation adheres
to relevant laws
o Citizens, who need to be informed and consulted about AI initiatives
in parliament
o External experts such as AI researchers or data scientists, who can
provide insights into AI implementation
• Regularly solicit feedback from stakeholders at all stages of the AI project
and incorporate this feedback into the AI development process.
• Keep stakeholders informed about the progress of AI projects, milestones
achieved, and any changes or updates, through regular reports and
presentations.

Start with a pilot project

• Initiate small-scale pilot projects to test AI solutions in real parliamentary
settings.
• Choose pilot projects with a clear and manageable scope.
• Look for areas within parliamentary processes that can benefit from AI,
such as document analysis, constituent communication or data-driven
decision support.
• Assess the feasibility and potential impact of proposed pilot projects.
• Establish clear, specific objectives for each pilot project. These should be
aligned with the overall goals of enhancing parliamentary efficiency,
transparency and decision-making.
• Develop success criteria to measure the effectiveness of the pilot projects.
These criteria could include performance improvements, AI output
accuracy, user satisfaction and cost savings.
• Create a detailed project plan that outlines the steps, timelines, resources
and responsibilities for the pilot project.
• Identify potential risks and develop mitigation strategies.

Use agile methods

• Plan the activities in sprints, each lasting a couple of weeks.
• Involve business experts in each phase of the project. Doing so offers
numerous benefits including a faster response to legislative changes,

164
Guidelines for AI in parliaments

better cross-team collaboration, higher user satisfaction, reduced project

risks and increased transparency for stakeholders.
Build internal AI expertise
• Invest in developing AI expertise among parliamentary staff.
• Provide training and resources to help staff understand, evaluate and
effectively use AI technologies.
Collaborate with AI experts and share with other parliaments
• Collaborate with AI experts, researchers and technology providers to build
knowledge and experience.
• Collaborate with academic institutions, research organizations and other
parliaments to share and learn, and to access cutting-edge AI research
and innovations.
Leverage responsible AI tools
Consider using the recommended tools detailed in the Inter-American Development
Bank publication Responsible use of AI for public policy: Data science toolkit:

• Robust and Responsible AI Checklist: This tool consolidates the main

concerns by stage of the AI life cycle. The checklist must be reviewed
continuously by technical teams and decision makers.
• Data Profile: This tool is an initial exploratory analysis conducted during
the data-collection and processing stage of the AI life cycle. It provides
information to reassess the quality, completeness, temporality and
consistency of the training data set, possible biases within this data set,
and the implications of the use of an AI system, including potential harm.
• Model Card: This tool summarizes the main features of the AI system,
highlighting the main assumptions, the most important characteristics of
the system, and the risk mitigation measures implemented.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

165
Guidelines for AI in parliaments

Systems development:
Deployment patterns
About this sub-guideline
This sub-guideline is part of the guideline Systems development. It can be read in
conjunction with the sub-guideline Systems development: Deployment and
implementation. Refer to the main guideline for context and an overview.

Background
The choice of technology and infrastructure significantly influences the robustness,
security, scalability, performance and ease of supervision of AI systems. Together,
these factors contribute to the overall reliability of such systems.

Different deployment patterns come with their own set of trade-offs. Some may incur
higher costs, while others might require more extensive management resources. As
a result, there is no one-size-fits-all “best approach” to AI system deployment.
Instead, the key to successful deployment lies in carefully assessing parliament’s
needs, resources and goals, and then selecting an approach that offers the best
balance of features and practicality for parliament’s particular situation.

Characteristics of deployment patterns

The deployment of AI systems involves following several patterns and practices to
ensure that models perform effectively and reliably in production environments.
When designing the AI system architecture, parliament should therefore consider the
deployment pattern characteristics discussed below.
Deployment architecture
The deployment architecture of an AI system is determined by two key factors: how
the AI algorithm responds to requests, and where the AI model is hosted.

Request-handling patterns:

• Batch processing: Data is processed in large batches at scheduled

intervals, making this pattern suitable for non-time-sensitive tasks.
• Online serving: Requests are handled in real time as they come in, making
this pattern ideal for applications requiring immediate responses.

166
Guidelines for AI in parliaments

• Streaming: Under this pattern, data streams are continuously processed,

enabling near-real-time analysis and predictions.

Hosting location types:

• On-premises: Models are deployed on local servers, often for the purpose
of enhanced security or to meet specific compliance requirements.
• Cloud: Models are hosted on cloud platforms, offering benefits such as
scalability, flexibility and reduced infrastructure management.
• Edge: Models are deployed on edge devices, providing low-latency
predictions and offline capabilities, making this approach suitable for
Internet of Things (IoT) and mobile applications.
• Hybrid: This approach combines on-premises, cloud and edge
deployments to optimize performance and resource usage based on
specific needs.

The choice of deployment architecture depends on factors such as data sensitivity,

response-time requirements, available resources and the specific use case of the AI
system.

Scalability
It is important to understand the average number of requests the AI system will
receive, along with its life cycle. These factors will determine the deployment
scalability characteristics:

• Horizontal scaling: Adding more instances of the model server to handle

increased load
• Vertical scaling: Enhancing the capacity of existing servers (e.g. by adding
more memory or faster central processing units (CPUs))
• Auto-scaling: Automatically adjusting the number of model instances
based on demand
Latency and throughput
When deploying AI systems, two critical performance metrics to consider are latency
and throughput:

• Latency refers to the time it takes for the AI model to respond to a request,
which is particularly crucial for real-time applications.
• Throughput measures the number of requests the AI model can process
per unit of time, which is essential for high-volume applications.

It is important to establish acceptable values for both latency and throughput to

ensure that the system meets the specific needs of the application for which it is
intended, and that it can handle the expected workload efficiently.
Model management
Effective AI model management is crucial throughout the entire life cycle of an AI
system. However, it becomes particularly important once the AI system is put into

167
Guidelines for AI in parliaments

operation. A well-designed model management strategy should address several key

aspects:

• Versioning: This involves keeping track of different versions of the model,

ensuring traceability and the ability to roll back if needed. Proper
versioning allows teams to manage changes, compare performance
across iterations and maintain a clear history of the model’s changes over
time.
• Life cycle management: This approach encompasses the tools and
processes for deploying, monitoring, updating and, eventually, retiring
models. The aim is to ensure that models are properly maintained
throughout their operational life, from initial deployment through to
eventual replacement.
• A/B testing: This practice involves running multiple versions of a model
simultaneously to compare their performance. A/B testing allows teams to
make data-driven decisions about which model version performs best in
real-world conditions before full deployment.
Monitoring and observability
• Performance metrics: Monitoring metrics such as response time,
throughput and resource utilization
• Drift detection: Identifying when the model’s performance degrades owing
to changes in data distribution
• Alerting: Setting up alerts for anomalies or performance degradation
Security
• Access control: Ensuring that only authorized users and applications can
interact with the model
• Data privacy: Protecting sensitive data and adhering to regulations (e.g.
GDPR)
• Model security: Safeguarding models against adversarial attacks and data
poisoning
Continuous integration/continuous deployment (CI/CD)
• Automation: Automating the deployment process to reduce errors and
deployment time
• Testing: Including automated testing (unit, integration, regression) in the
deployment pipeline
• Rollbacks: Providing mechanisms for quickly reverting to previous versions
in case of issues
Resource management
• Hardware acceleration: Utilizing graphics processing units (GPUs), tensor
processing units (TPUs) or other accelerators for improved performance
• Resource allocation: Managing resources to optimize cost and
performance
• Integration with existing systems: Providing APIs for integration with other
systems and services

168
Guidelines for AI in parliaments

• Data pipelines: Integrating with data ingestion and pre-processing

pipelines
• Feedback loops: Implementing systems to collect feedback from model
predictions to improve future performance
Resilience and fault tolerance
• Redundancy: Having multiple instances or backups to ensure availability
• Failover: Automatically switching to backup systems in case of failure
• Retry logic: Implementing mechanisms to handle transient failures
Auditability and explainability
In most cases, audit logs are mandatory for predictions, inputs and system
interactions.

In addition to auditing, explainability tools can be used to interpret AI model

decisions, thus improving trust and compliance.

Combinations of deployment patterns

Various combinations of characteristics are often seen in AI use cases. These are
detailed below:
Model-as-a-service (MaaS)
• Characteristics: Exposing models via web APIs for easy integration
• Use cases: Real-time predictions, microservices architecture
Model embedded in applications
• Characteristics: Embedding models directly in applications, either locally or
via a microservice
• Use cases: Edge computing, offline capabilities, low-latency requirements
Containerized deployment
• Characteristics: Packaging models in containers (e.g. Docker) for
consistent deployment across environments
• Use cases: Cloud deployments, microservices, scalable architectures
Serverless deployment
• Characteristics: Using serverless computing platforms to deploy models
• Use cases: Event-driven applications, cost optimization for intermittent
workloads
On-demand/batch processing
• Characteristics: Deploying models that run on demand or process large
batches of data periodically
• Use cases: Data-processing pipelines, periodic analytics
Streaming analytics
• Characteristics: Deploying models to analyse and predict data from
streaming sources

169
Guidelines for AI in parliaments

• Use cases: Real-time analytics, IoT applications

A/B testing and canary releases
• Characteristics: Testing new models on a subset of traffic before full
deployment
• Use cases: Incremental updates, risk minimization
Federated learning
• Characteristics: Training models across multiple decentralized devices or
servers while keeping data local
• Use cases: Privacy-sensitive applications, distributed data sources

Recommended practices when deploying AI systems

Below are some suggestions and characteristics for parliaments to consider when
planning and executing AI deployments that are scalable, reliable, safe and efficient:

• Ensure best fit: Select a deployment pattern according to the specific use
case, performance requirements and domain constraints.
• Monitor and iterate: Continuously monitor deployed models and iterate
based on user feedback and performance metrics.
• Maintain security: Implement robust security practices to protect models
and data in production environments.
• Optimize resources: Efficiently manage resources to balance
performance and cost, leveraging approaches such as containerization
and serverless architectures where appropriate.

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the
IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more
information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].

170
Guidelines for AI in parliaments

Appendices

171
Guidelines for AI in parliaments

Glossary of terms
Accountability The principle that ensures clear responsibility can be
assigned for all decisions and actions throughout an AI
system's lifecycle, from planning to decommissioning
Affinity bias When someone prefers individuals who are similar to
them in terms of ideology, attitudes, appearance, or
religion
Agile A project management and development approach that
emphasizes flexibility, iterative progress, and
collaboration
AI governance The framework of policies, structures, and processes
created to maximize the benefits of AI while minimizing
its risks
AI literacy The ability to understand, critically evaluate, and
effectively interact with AI technologies, including
knowledge of AI's capabilities, limitations, and potential
impacts
AI PPM (AI Project Portfolio The centralized management of an organization's AI
Management) initiatives to meet strategic objectives by optimizing
resource allocation, balancing risks, and maximizing
value
Algorithm A set of rules or instructions given to an AI system to
help it learn, make decisions, and solve problems
Amplification bias Occurs when several AI systems, each with separate
biases, interact and mutually reinforce each other's
biases
API (Application A set of rules and protocols that allows different
Programming Interface) software applications to communicate with each other
Automation bias When conclusions drawn from algorithms are valued
more highly than human analyses
Cloud storage The practice of storing data and applications on remote
servers accessed via the internet, rather than on local
computers
Coverage bias A form of sampling bias that occurs when a selected
population does not match the intended population

172
Guidelines for AI in parliaments

Data architecture The overall structure of an organization's data assets

and data management resources
Data bias A type of error where certain elements of a data set are
more heavily weighted or represented than others
Data cleaning The process of detecting and correcting errors,
inconsistencies, and inaccuracies in data sets
Data governance The framework of policies, processes, and standards
that ensure the effective management of data assets
Data literacy The ability to read, understand, create, and
communicate data as information, including
understanding data collection, analysis, interpretation,
and presentation
Data migration The process of moving data from one system or
storage type to another
Data mining The process of discovering patterns and relationships
in large data sets
Data set A collection of related data points or information used
to train AI systems
Data steward A person responsible for managing and overseeing an
organization's data assets
Data visualization The graphical representation of data and information
using charts, graphs, and other visual elements
Database An organized collection of structured information or
data
Deepfake Synthetic media where a person's likeness is replaced
with someone else's using AI
Deployment The process of making an AI system or application
available for use
Deployment bias When a system that works well in a test environment
performs poorly when deployed in the real world
Ethical principles Guidelines that ensure AI systems respect privacy,
transparency, accountability, fairness, and human
autonomy while promoting societal well-being
Explainability The ability for humans to understand and trust
decisions, recommendations, or predictions made by
an AI systems
Feedback loop bias When the output of an AI system influences future
inputs, potentially reinforcing existing biases
Generative AI AI systems capable of creating new content (text,
images, code, etc.) based on patterns learned from
training data
HIC (Human-in-Command) A comprehensive oversight approach that considers
broader economic, social, legal, and ethical impacts of
AI systems
HITL (Human-in-the-Loop) An approach where a human mediates all decisions
made by the AI system

173
Guidelines for AI in parliaments

HOTL (Human-on-the-Loop) An approach where humans monitor AI system

operations and can intervene when necessary
Infrastructure The hardware, software, networks, and facilities that
support an organization's IT operations
Intellectual property rights Rights that protect the investment of rights-holders in
original content, including copyrights and accessory
rights
KPI (Key Performance Measurable values that demonstrate how effectively an
Indicator) organization is achieving key objectives
LLM (Large Language AI models trained on vast amounts of text data that can
Model) understand and generate human-like text
Linguistic bias When an AI algorithm favours certain linguistic styles,
vocabularies, or cultural references over others
Machine learning A subset of AI that enables systems to learn and
improve from experience without explicit programming
Natural Language The ability of computers to understand, interpret, and
Processing (NLP) generate human language
Neural network A computer system modelled on the human brain,
designed to recognize patterns
Open source Software whose source code is freely available for
anyone to inspect, modify, and enhance
Participation bias A form of sampling bias that occurs when certain
groups choose not to participate in data collection
Pilot project A small-scale preliminary study to evaluate feasibility,
cost, and potential issues
Privacy The principle that AI systems should respect and
protect personal data and information
Prompt engineering The practice of designing and optimizing inputs to AI
systems to generate desired outputs
Proxy bias When variables used as proxies for protected attributes
introduce bias into the model
RPA (Robotic Process Technology that automates repetitive tasks through
Automation) software robots
Robustness The ability of AI systems to maintain reliable operation
and withstand adverse conditions or attacks
Sampling bias When data is not randomly selected, resulting in a
sample that is not representative of the population
Shadow AI The unsupervised or unsanctioned use of AI tools
within an organization outside of its IT and
cybersecurity framework
Shadow IT The use of IT systems, devices, software, or services
without explicit organizational approval
Stakeholder Any person, group, or organization that has an interest
in or is affected by an AI project

174
Guidelines for AI in parliaments

SVG (Scalable Vector A web-friendly vector image format that can scale
Graphics) without losing quality
Temporal bias When training data becomes outdated and no longer
represents current realities
Traceability The ability to follow and monitor the entire lifecycle of
an AI system
Training data The initial data set used to teach an AI system to
perform its intended function
Transparency The communication of appropriate information about AI
systems in an understandable and accessible format
Use case A specific situation or scenario where an AI system or
application could be used
XAI (eXplainable AI) AI systems designed to be interpretable and
understandable by humans

175
Guidelines for AI in parliaments

This publication has been produced with the financial support of the European Union (EU), in
partnership with the International Institute for Democracy and Electoral Assistance (International
IDEA), as part of INTER PARES–Parliaments in Partnership, the EU’s Global Project to
Strengthen the Capacity of Parliaments.

The designations employed and the presentation of material in this information product do not
imply the expression of any opinion whatsoever on the part of the Inter-Parliamentary Union
(IPU) or the EU concerning the legal or development status of any country, territory, city or area,
or of its authorities, or concerning the delimitation of its frontiers or boundaries.
The mention of specific companies or products of manufacturers, whether or not these have
been patented, does not imply that these have been endorsed or recommended by IPU or the
EU in preference to others of a similar nature that are not mentioned.

All reasonable precautions have been taken by the IPU to verify the information contained in this
publication. However, the published material is distributed without warranty of any kind, either
expressed or implied. Responsibility for the interpretation and use of the material lies with the
reader. In no event shall the IPU or the EU be liable for damages arising from its use.

176