All instruments Guideline Hong Kong: Ethical Artificial Intelligence Framework No summary is recorded for this instrument. Adopted Not yet reviewed English Open the source (opens in a new tab) Suggest a correction Details Full text Details Identity Title Ethical Artificial Intelligence Framework Original title Not recorded Instrument ID hk-na-etharin-2021 Type Guideline Status Adopted Jurisdiction and publisher Jurisdiction Hong Kong Publisher Office of the Government Chief Information Officer, Hong Kong Language English Dates and versions Date Not recorded Versions Enacted · Not recorded Classification Policy areas Not recorded Source Source URL https://www.digitalpolicy.gov.hk/en/our_work/data_governance/policies_standards/ethical_ai_framework/doc/Ethical_AI_Framework_en.pdf Retrieved August 29, 2026 License Unknown. Follow the publisher’s terms. Access note Not recorded Full text 306,583 characters Original text, English · 306,583 characters in 2 parts Load the full text Download as .txt Jump to part Jump to part Part 1 of 2Part 2 of 2 Outline (136 headings) 2. Added “Personal Information Protection 3. Updated Appendix B 4. Supplemented Appendix E 2 Updated the references in Appendix E Appendix E 1.2 July 2023 3 Updated the references in Appendix E Appendix E 1.3 August 4 Change “Office of the Government Chief 1.4 July 2024 2. Updated Appendix A-D Appendix A-D 2025 3. Removed Appendix E 1. EXECUTIVE SUMMARY 1.1 INTRODUCTION 1.2 WHAT IS THE ETHICAL AI FRAMEWORK? 1. Project Strategy 3. Project Ecosystem 5. System Deployment 6. System Operation 2. Project Planning 4. Project Development Section 4 “AI Practice Guide” contains detailed practices to be followed for a number of practice 1. Risk Gating 2. AI Application 3. AI Practice Guide 4. Impact 1.3 HOW CAN THE ETHICAL AI FRAMEWORK BE USED? Section 3.5 “Key Components and Relationships”. 2. PURPOSE Section 5 - AI Assessment Section 4.1.6 - System Operation and 3. OVERVIEW OF THE ETHICAL AI FRAMEWORK 3.1 VISION FOR THE ETHICAL AI FRAMEWORK 3.2 OBJECTIVES 3.3 BENEFITS 3.4 INTRODUCTION TO AI AND DATA ETHICS 3.5 KEY COMPONENTS AND RELATIONSHIPS Section 5 “AI Assessment”). The Ethical AI Framework should be read in conjunction with 3.5.1 Ethical AI Principles 3.5.1.1 Transparency and Explainability 3.5.1.2 Reliability, Robustness and Security 3.5.1.3 Fairness 3.5.1.4 Diversity and Inclusion 3.5.1.5 Human Oversight 3.5.1.6 Lawfulness and Compliance 3.5.1.7 Data Privacy 3.5.1.8 Safety 3.5.1.9 Accountability 3.5.1.10 Beneficial AI 3.5.1.11 Cooperation and Openness 3.5.1.12 Sustainability and Just Transition 3.5.2 AI Governance Section 5 “AI Assessment” for details. 3.5.2.1 AI Governance Structure 3.5.3 AI Lifecycle 1. Project Strategy 2. Project Planning 3. Project Ecosystem 4. Project Development 5. System Deployment 6. System Operation and Monitoring 4. AI PRACTICE GUIDE 4.1 AI LIFECYCLE AND PRACTICES 4.1.1 Project Strategy 4.1.1.1 Organisation Strategy, Internal Policies and Practices Section 5 “AI Assessment” for details on the governance 4.1.1.2 Industry Standards and Regulations 1. Accountability – Standards and regulations should necessitate organisations and 3. Lawfulness and Compliance – Ensuring industry standards and regulations are considered 4.1.2 Project Planning 4.1.2.1 Portfolio Management 3. Diversity and Inclusion – Stakeholders impacted can be mapped across the range of AI 4.1.2.2 Project Oversight and Delivery Approach Section 4.1.4.5.6 “Adversarial Attacks” and Section 4.1.5.1 2. Human Oversight – Human intervention can be reviewed as part of the project and 4. Cooperation and Openness – Organisations should proactively engage both internal and 4.1.3 Project Ecosystem 4.1.3.1 Technology Roadmap for AI and Data Usage 2. Sustainability and Just Transition – Technology used should be scalable and have 4.1.3.2 Procuring AI Services (Sourcing) 3. Reliability, Robustness and Security – AI Applications from third parties should be 4.1.4 Project Development 4.1.4.1 Business and Data Understanding 1. Cooperation and Openness – organisations should engage both internal and external 4.1.4.2 Solution Design 1. Consider a human-out-of-the-loop approach in decision- 2. Consider a human-in-the-loop approach when there is limited 3. Consider a human-in-command when the human-in-the-loop 1. Human Oversight – Ensure an appropriate level of human intervention based on multiple 4.1.4.3 Data Extraction 3. Safety, Reliability, Robustness and Security – ETL tools must be secure and robust when 4.1.4.4 Pre-processing 1. Statistical (Equal) Parity – Positive classification rates 2. Equal Opportunity – Mathematically, it implies True 3. Overall Accuracy Equality – Both groups should give 4. Gini Equality – Gini is defined as the ratio area between 1. Data Privacy - Ensure minimal collection and processing of personal data and retention 2. Lawfulness and Compliance - Ensure consideration and/or compliance with local and 4.1.4.5 Model Building 1. Reliability, Robustness and Security – Rewards functions in AI can reinforce learning 1. Reliability, Robustness and Security – Overfitting should be avoided to ensure outcomes 1. Reliability, Robustness and Security – Third-party’s compliance with relevant 1. Reliability, Robustness and Security – AI applications have to remain robust and secure 4.1.5 System Deployment 4.1.5.1 Model Integration & Impact Section 4.1.6.1 “Data and Model Performance Monitoring” for 3. Cooperation and Openness – Project Managers can work together with the Project Team 4.1.5.2 Transition & Execution 1. Reliability, Robustness and Security – Following security practices and implementing 2. Human Oversight/Safety – Controls should be implemented that allow for human 4. Beneficial AI – Provide disclosure statements on the benefits of the AI application to 4.1.5.3 Ongoing Monitoring 4.1.5.4 Evaluation & Check-in Section 4.1.4.5.1 “Model Assumptions” for contents related to 1. Lawfulness and Compliance – Relevant measures are taken to ensure industry standards 4.1.6 System Operation and Monitoring 4.1.6.1 Data and Model Performance Monitoring 2. Accountability – Ensure there are accountable roles to perform quality checks and 4.1.6.2 Operational Support 2. Accountability – IT support should be made accountable to fix the AI application within 4.1.6.3 Continuous Review/Compliance Section 5 “AI Assessment” of this document. 1. Lawfulness and Compliance – Ensure compliance as new industry standards and 5. AI ASSESSMENT 5.1 AI APPLICATION IMPACT ASSESSMENT 1. Risk Gating Criteria – A set of questions that are used to distinguish high-risk AI 2. AI Application Impact Assessment Questions – The questions are provided to ensure that 3. AI Practice Guide Reference – Assessment questions have included a reference (where 4. Impact Considerations – A set of questions about beneficial and negative impacts on 5.1.1 Process for Completing AI Application Impact Assessment 1. The Project Team within an organisation should complete an AI Application Impact 2. The AI Application Impact Assessment can be used as a ‘live’ document throughout the AI 4. Results of the AI Application Impact Assessment should be reviewed and approved by the 5.2 FREQUENCY OF AI ASSESSMENT 5.3 RECOMMENDATION APPENDIX 6. APPENDIX APPENDIX A – GLOSSARY APPENDIX B – REFERENCE Part 1: General requirements Part 2: Testing and evaluation for metrics Part 3: Service capability maturity APPENDIX C – AI APPLICATION IMPACT ASSESSMENT TEMPLATE Section 4.1.3.2:: • Define requirements Section 4.1.6.1: APPENDIX D – AI STRATEGY TEMPLATE The outline comes from headings detected in the text, so it might miss some or include extras. JavaScript is turned off. To read the text, open each of its 2 parts: Part 1Part 2 1,083 of 1,262, newest first Previous Ethical AI Toolkit / AI Principles and Guidelines for the Emirate of Dubai (Smart Dubai) Next Example Applications of Digital Health Technology for Epidemic Prevention and Control (Third Phase) Show a random instrument All instruments