Building trustworthy artificial intelligence : frameworks, applications, and self-assessment for readiness

Title page 1

Contents 4

ACKNOWLEDGMENTS 3

ACRONYMS 6

1. INTRODUCTION 7

2. AI IN PUBLIC GOVERNANCE 8

2.1. AI'S POTENTIAL TO DRIVE INNOVATION AND EFFICIENCY IN PUBLIC SECTOR SERVICES 8

2.1.1. Efficiency of policy-making process 8

2.1.2. Better public services 9

2.2. ETHICAL CONSIDERATIONS IN AI AND GOVERNMENT 9

2.3. KEY CONCEPTS AND TERMINOLOGIES 9

2.3.1. Main concepts of trustworthy AI governance 10

2.3.2. Privacy protection techniques 12

3. BIG DATA-DRIVEN DECISION-MAKING 13

3.1. DATA ANALYTICS AND PATTERN RECOGNITION: AN INTEGRATED APPROACH 13

3.1.1. The symbiotic relationship: How pattern recognition powers data analytics 14

3.1.2. Key techniques and applications 16

3.2. AI-BASED SIMULATION 17

4. ETHICAL CHALLENGES AND RISKS IN AI APPLICATION 18

4.1. EXPLAINABILITY 18

4.2. ILLUSION OF OBJECTIVITY IN AI 18

4.3. DATA PRIVACY 21

4.4. ACCOUNTABILITY 21

4.5. COMPREHENSIVE ETHICS 22

5. TECHNOLOGICAL SOLUTIONS TO ADDRESS AI CHALLENGES 24

5.1. TECHNOLOGICAL APPROACHES TO EXPLAINABILITY 24

5.1.1. Explainability by model design 24

5.1.2. Model-specific explanation methods 25

5.2. TOOLS AND TECHNIQUES FOR ADDRESSING AI BIAS AND THE ILLUSION OF OBJECTIVITY 25

5.2.1. Bias detection and fairness audits 25

5.2.2. Data preprocessing techniques 26

5.2.3. Algorithmic adjustments for fairness 26

5.2.4. Post hoc bias mitigation 26

5.3. PRIVACY-ENHANCING TECHNOLOGIES (PETs) 27

6. AI GOVERNANCE 30

6.1. GLOBAL FRAMEWORKS 30

6.1.1. United Nations 30

6.1.2. World Bank 31

6.1.3. Organisation for Economic Co-operation and Development 31

6.2. GOVERNMENTS 32

7. AI AND SOCIETY 35

7.1. DIGITAL DIVIDE OR JOB DISPLACEMENT 35

7.2. RESPONSIVENESS VIA AI 35

7.3. MEASURES TO SUPPORT MARGINALIZED GROUPS 36

8. SELF-ASSESSMENT FOR AI USE 37

8.1. BACKGROUND 37

8.2. AI READINESS CHECKLISTS 38

8.2.1. Decision flow for AI data collection and processing 41

8.3. CHECKLIST FOR DATA PRIVACY IN AI 41

8.4. CONSIDERATIONS IN USING CHECKLISTS 43

9. IMPLICATIONS 44

REFERENCES 45

Tables 16

Table 1. Key characteristics of data analytics and pattern recognition in AI 16

Table 2. Trade-offs between AI principles 23

Table 3. OECD AI principles and recommendations 32

Table 4. Regulatory frameworks for AI development and utilization 33

Table 5. Data protection measures and privacy regulations 34

Table 6. Checklist for data privacy in AI 42

Figures 40

Figure 1. Checklist for AI data collection and processing 40

Boxes 11

Box 1. Autonomy in humans vs. AI 11

Box 2. Types of bias 19

Box 3. The case of Korea: How Korea leveraged trustworthy AI to combat COVID-19 while ensuring privacy protection 28