Ireland: Guidelines for the Responsible Use of Artificial Intelligence in the Public Service

1. 1. Executive Summary 7
2. 2. Introduction 10
3. 2.1 What is AI? 10
4. 2.2 Overview of the Guidelines 10
5. 2.3 Audience 11
6. 3. AI in the Public Service 14
7. 3.1 Types of tasks AI can be used for in public services 14
8. 3.2 Benefits of AI for Better Public Services 16
9. 3.3 Risks associated with using AI in Public Services 18
10. 4. Principles for Responsible Use of AI in the Public Service 21
11. 4.1 Human Agency and Oversight 23
12. 4.2 Technical Robustness and Safety 25
13. 4.3 Privacy and Data Governance 27
14. 4.4 Transparency 30
15. 4.5 Diversity, Non-discrimination, and Fairness 32
16. 4.6 Societal and Environmental Well-being 34
17. 4.7 Accountability 36
18. 5. Decision Framework when working with AI 38
19. 5.1 Is AI the best solution? 38
20. 5.2 What type of AI solution is the best fit? 39
21. 5.3 Using ‘free-of-charge’ or enterprise versus licensed AI offerings 39
22. 5.4 Inclusion and diversity in AI from the start 41
23. 5.5 Next steps after the Decision Framework 41
24. 6. The Responsible AI Canvas 42
25. 6.1 The Responsible AI Canvas Overview 42
26. 6.2 How to use the Responsible AI Canvas 44
27. 6.3 Next steps after the Responsible AI Canvas 44
28. 7. AI lifecycle for responsible adoption 45
29. 7.1 AI lifecycle Introduction 45
30. 7.2 Implementing the seven principes across an AI lifecycle 46
31. 8. Guidance for End-Users of GenAI 60
32. 9. AI Use Cases in the Public Service 63
33. 9.1 Internal Operations - Efficiency improvements of internal public sector operations 63
34. 9.3 Internal and External Oversight - Enhancing oversight and risk detection to facilitate
35. 9.4 Policy-making – Improve the decision-making process 68
36. Appendix 1: Related national and EU Strategies 69
37. Appendix 2: The EU AI Act – Key concepts and definition 72
38. Appendix 3: Other Regulations relevant to these Guidelines 75
39. Chapter 1:: Executive Summary
40. 1. The Seven Principles for Responsible AI
41. 2. A Decision Framework
42. 3. A Responsible AI Canvas Tool
43. 4. AI Lifecycle Guidance
44. Chapter 2:: Introduction
45. 2.1 What is AI?
46. 2.2 Overview of the Guidelines
47. 2.3 Purpose
48. 2.4 Audience
49. Chapter 3:: AI in the Public Service
50. 3.2 Benefits of AI for Better Public Services
51. 3.3 Risks associated with using AI in Public Services
52. Chapter 4:: Principles for Responsible
53. 1. Human agency and oversight
54. 2. Technical robustness and safety
55. 3. Privacy and data governance
56. 4. Transparency
57. 5. Diversity, non-discrimination, and fairness
58. 6. Societal and environmental well-being
59. 7. Accountability
60. 1 Technical
61. 4.1 Human Agency and Oversight
62. 4.2 Technical Robustness and Safety
63. 4.3 Privacy and Data Governance
64. 1. Does the intended dataset, to be used for AI processing, contain any personal data? The
65. 4.4 Transparency
66. 4.5 Diversity, Non-discrimination, and Fairness
67. 4.6 Societal and Environmental Well-being
68. 4.7 Accountability
69. Chapter 5:: Decision Framework
70. 5.1 Is AI the best solution?
71. 5.2 What type of AI solution is the best fit?
72. 5.3 Using ‘free-of-charge’ or enterprise versus licensed AI offerings
73. 5.4 Inclusion and diversity in AI from the start
74. 5.5 Next steps after the Decision Framework
75. Chapter 6:: The Responsible
76. 6.1 The Responsible AI Canvas Overview
77. 6.2 How to use the Responsible AI Canvas
78. 3. Risk management: The canvas encourages proactive risk management to assess and mitigate
79. 6.3 Next steps after the Responsible AI Canvas
80. Chapter 7:: AI lifecycle for
81. 7.1 AI lifecycle Introduction
82. 7.2 Implementing the seven principes across an AI lifecycle
83. 2. Technical ● Ensure the prospective AI system is in line with data security
84. 3. Privacy & Data ● Ensure processes around privacy and data protection are set
85. 4. Transparency ● Establish what level of transparency / explainability is required
86. 5. Diversity, Non- ● Where applicable, establish a plan for engaging with
87. 6. Societal & ● Conduct social impact assessments to ensure AI systems
88. 7. Accountability ● Establish accountability frameworks early in design for each
89. 1. Human Agency ● Validate data processing to allow for human intervention and
90. 2. Technical ● Ensure data used during this phase is kept secure and
91. 3. Privacy & Data ● Ensure processes around privacy and data protection are
92. 4. Transparency ● Document to the best possible standard the datasets and the
93. 5. Diversity, Non- ● Ensure the dataset is representative and any discriminatory
94. 6. Societal & ● Ensure that data collection and processing is done responsibly
95. 7. Accountability ● Ensure the accountability framework established is being
96. 1. Human Agency ● Build models while considering their ability to include human
97. 2. Technical ● Ensure the model is kept secure and procedures are followed in
98. 4. Transparency ● Ensure the model can be explained to the level defined in the
99. 5. Diversity, Non- ● Continuously test for bias in model outputs during
100. 6. Societal & ● Consider the environmental impact of AI models and if any
101. 1. Human Agency ● Models should be tested for their ability to include human
102. 2. Technical ● Consideration should be given to possible unintended
103. 6. Societal & ● Assess how the model could negatively impact people’s
104. 1. Human Agency ● The appropriate level of human oversight established in the
105. 2. Technical ● Monitor real-world AI performance to ensure system
106. 4. Transparency ● Decisions made by the AI system should be documented to the
107. 5. Diversity, Non- ● Monitor AI decision-making during deployment, to ensure
108. 6. Societal & ● Before deployment, assess if the AI system contributes
109. 1. Human Agency ● Set up continuous monitoring systems ensuring AI systems
110. 4. Transparency ● Monitor feedback to see if users are struggling to comprehend
111. 5. Diversity, Non- ● Conduct post-deployment testing to ensure the system
112. 6. Societal & ● Monitor the societal impact of the model to gauge ongoing
113. 3. Privacy & Data ● Ensure safe handling of personal data post-retirement. Follow
114. 5. Diversity, Non- ● Consider fairness to all stakeholder groups with decisions
115. 6. Societal & ● Decommissioning should be done in a means that is as
116. 7. Accountability ● An accountable stakeholder should be appointed for the
117. Chapter 8:: Guidance for End-Users
118. Chapter 9:: AI Use Cases in the
119. 9.1 Internal Operations - Efficiency improvements of internal
120. 9.2 Service Delivery - Enhancing the delivery of public services
121. 9.3 Internal and External Oversight - Enhancing oversight and
122. 9.4 Policy-making – Improve the decision-making
123. Appendix 1: Related national and EU Strategies
124. 2030 Strategy’ is a strategy with the ambition “to collaboratively deliver
125. Appendix 2: The EU AI Act – Key concepts and
126. Appendix 3: Other Regulations relevant to these
127. Section 3.3, data privacy and protection are a key component in responsible AI systems and GDPR is an
128. Article 4 of the AI Act requires providers and deployers of AI systems to ensure a sufficient level of AI
129. 05 November 2024. [Online]. Available: https://www.edps.europa.eu/data-protection/data-
130. 12 October 2022. [Online]. Available: https://commission.europa.eu/strategy-and-policy/
131. 21 December 2023. [Online]. Available: https://www.gov.ie/en/collection/c72a9-value-

The outline comes from headings detected in the text, so it might miss some or include extras.