Adopting Artificial Intelligence for Knowledge Management in Multilateral Development Organizations: Enablers, Knowledge Dynamics, and Early Implications = 다자개발조직에서의 지식관리를 위한 인공지능 도입: 촉진요인, 지식 역학 및 초기 함의

국문 초록 (Abstract)

지식은 개발협력 분야에서 가장 큰 잠재력을 지닌 자원 중 하나이나, 동시에 가장 오랫동안 실현되지 못한 과제이기도 하다. 다자개발조직들은 그간 자신들의 가장 큰 비교우위가 재정 지원뿐만 아니라 수년 간의 경험을 통해 축적한 통찰력, 교훈, 분석적 깊이에 있다고 선언해왔다. 그러나 최근 개발조직들이 당면하고 있는 문제의 복합성은 심화되었으며, 재정적 여력 또한 감소함에 따라 더 적은 재원으로 더 많은 성과를 내야 한다는 삼중고(triple squeeze)를 마주하고 있다. 이에 따라 지식 집약적 업무와 조직의 운영을 지원하기 위해 인공지능(AI)을 포함한 디지털 도구에 점점 더 의존하고 있는 추세이다. 이러한 배경 속에서 본 논문은 다자개발조직이 지식관리(KM)을 위해 AI 기반 도구를 왜 도입하고, 어떻게 사용하는지, 그리고 이러한 도구들이 기존의 지식 프로세스와 어떻게 상호작용하는지를 탐색한다. 대부분 조직에서의 AI 도구 도입이 초기 단계에 있음을 고려하여, 본 연구는 지식관리를 위한 AI 도구 도입 촉진요인, 이로 인한 KM 프로세스의 변화와 초기 함의를 중점적으로 이해하고자 탐색적 접근법을 채택하였다.
본 연구는 질적 다중사례연구 방법을 사용한다. 아시아개발은행, 세계은행, 미주개발은행, 유엔개발계획 등 4개 기관의 사례를 분석하였으며, 위 기관들의 AI 도입 방식(중앙집중형 또는 분산형)과 AI 기술이 중점적으로 지원하는 지식관리 기능(검색 또는 지식 생성/종합)에 따라 4개 기관의 유형을 분류하였다. 자료수집을 위해 연구대상 기관에 소속된 13명의 직원들과 반구조화된 인터뷰를 실시하였으며, 전략문서, 평가보고서, 기술보고서 등의 문서를 추가 검토하였다. 인터뷰 자료는 Gioia 기법을 활용하여 분석하였다.
가장 먼저 본 연구는 기술(Technology), 조직(Organization), 환경(Environment) 요인을 통합적으로 분석하는 모델인 TOE 프레임워크를 활용하여 다자개발조직들이 AI 기반 KM 도구를 도입하는 이유와 조건, 즉 도입과정을 살펴보았다. 분석 결과, AI의 기술적 이점 외에도 조직적, 환경적 요인이 도입과정에 영향을 미치는 것으로 나타났다.
두번째로 본 연구는 Evans et al. (2014)가 제안한 지식관리주기에 기반하여 AI 기반 도구가 각각의 KM 프로세스에 미치는 영향을 탐색하였다. 본 연구는 AI 도구가 지식의 식별, 생성, 저장, 공유 등 지식관리 전 주기의 각 단계를 조직이 수행하는 방식을 재구성한다는 것을 발견하였다. 특히, 본 연구는 앞서 제시한 2x2 유형론을 사용하여 다자개발조직들이 조직적 특성과 운영 문제를 정의하는 방식의 차이에 따라 조직학습, 증거활용 및 제도적 일관성 측면에서 뚜렷한 패턴을 나타내며, 이에 따라 KM 프로세스가 달라진다고 분석하였다. 이러한 역학 관계를 개념화하기 위해 본 연구는 AI 기반 KM 프로세스와 증거 기반 운영 과 제도적 일관성 및 입장을 포함하는 상위 조직적 성과를 연결하는 AI 기반 지식역학의 다중 루프 모델(Multi-loop model of AI-supported KM)을 제안한다.
종합하자면 본 연구는 AI 기반 KM이 증거 기반 운영과 제도적 일관성을 뒷받침할 수 있는 조성적 인프라(enabling infrastructure)로서 기능하며 이를 통해 파급력 있는 개발 성과를 거둘 수 있는 여건을 조성한다는 결론을 내린다. 마지막으로 본 논문은 증거활용, 제도적 일관성 및 입장에 대한 함의와 함께 AI 시대에 개발 실무자들이 사고하고 결정하며 행동하는 방식에 미치는 함의를 논의하며 연구를 끝맺는다.

다국어 초록 (Multilingual Abstract)

Knowledge has been the development sector’s greatest promise and its most persistent failure. For decades, MO have declared that their greatest comparative advantage lies not only in financing but in the insights, lessons, and analytical depth accumulated across years of experience. At the same time, development organizations face a “triple squeeze”: surging demands for knowledge as the nature of development challenges becomes increasingly complex and interlinked, shrinking fiscal space, and heightened accountability demands to deliver more with less. Against this backdrop, this dissertation examines how and why multilateral development organizations (MO) adopt and use AI-powered tools for knowledge management (KM), and how these tools interact with existing knowledge processes. Recognizing that most tools are still at an early diffusion stage, the study adopts an exploratory approach to understand how and why organizations adopt AI technology for KM, how they integrate it into their work, and how early changes in KM processes are interpreted institutionally.
To investigate these questions, the dissertation employs a qualitative multiple case study design. Four organizations, the Asian Development Bank, the World Bank Group, the Inter-American Development Bank, and the United Nations Development Programme, were selected to capture variation in approaches to AI adoption (centralized or decentralized) and in the primary KM functions supported by AI technologies (retrieval or synthesis). Sixteen semi-structured interviews were conducted with staff working in KM, digital innovation, evaluations, and other operational roles, supplemented by document review of strategies, evaluation reports, and technical publications. Interview data were analyzed using Gioia’s methods.
First, the study draws on the Technology-Organization-Environment (TOE) framework to examine AI adoption processes for KM. The findings suggest that organizational and environmental factors shape the adoption process alongside AI’s technological affordances. Second, this study employs Evans et al.’s (2014) Knowledge Management Cycle to explore how AI-enabled tools reshape knowledge dynamics within these organizations. Through a grounded cross-case analysis of four MO, the study finds that AI tools reconfigure how organizations enact each phase of the KM cycle, including identifying, creating, storing, and sharing knowledge. Building on the 2x2 typology of AI-enabled KM configurations, the study shows that MO exhibit four distinct patterns of organizational learning, evidence use, and institutional coherence, shaped by differences in organizational characteristics and the ways organizations frame operational problems. To conceptualize these dynamics, the research proposes a multi-loop model of AI-supported knowledge dynamics that links AI-enabled KM processes to organizational-level outcomes, including evidence-informed operations and institutional coherence and voice.
Taken together, this research concludes that AI-enabled KM functions as an enabling infrastructure that can support evidence-informed operations and institutional coherence, creating conditions for development results at scale. The dissertation further discusses implications for evidence use, institutional coherence and voice, and the ways development practitioners think, decide, and act in the AI era.

목차 (Table of Contents)

• Abstract i
• List of Figures ix
• List of Tables x
• List of Abbreviation xi
• Part 1. Introduction and Research Design 1
• Chapter 1. Introduction 2
• 1.1. Setting the Scene: Triple Squeeze 2
• 1.2. AI for Knowledge Management in International Development 4
• 1.3. Purpose, Significance, and Structure of the Study 5
• Chapter 2. Literature Review and Theoretical Framework 9
• 2.1. Knowledge in International Development 9
• 2.1.1. What is Knowledge for Development Organizations? 9
• 2.1.2. Persistent Challenges in Managing Development Knowledge 10
• 2.2. Artificial Intelligence as a New Opportunity for KM 13
• 2.2.1. Potential Roles and Risks of Using AI for Knowledge Management 13
• 2.2.2. Empirical Understanding of the AI-KM Interplay 18
• 2.3. Artificial Intelligence for Development (AI4D) 21
• 2.3.1. From ICT4D to AI4D 21
• 2.3.2. Thematic Landscape of AI and Sustainable Development Research 23
• 2.3.3. Operationalizing AI for Development 27
• 2.4. Research Gap 30
• 2.5. Theoretical Backgrounds 32
• 2.5.1. Knowledge Management Life Cycle Models 32
• 2.5.2. Technology-Organization-Environment (TOE) Framework 39
• Chapter 3. Research Design and Methods 45
• 3.1. Overview of Research Design 45
• 3.2. Case Selection 47
• 3.3. Data Collection 50
• 3.3.1. Semi-structured Interview 51
• 3.3.2. Document Review 57
• 3.4. Analytical Framework 59
• 3.5. Analytical Procedures 60
• 3.6. Validity and Reliability 62
• 3.7. Ethical Considerations 62
• Part 2. Case Study Findings 64
• Chapter 4. Asian Development Bank 66
• 4.1. Knowledge Integration under the ‘One ADB’ Approach 66
• 4.1.1. Early AI Experiments under the One ADB Vision 68
• 4.2. AI-Powered Tools for Knowledge Management at Asian Development Bank 70
• 4.3. Factors Driving AI Adoption for Knowledge Management at Asian Development Bank 74
• 4.3.1. Technological Factors 74
• 4.3.2. Organizational Factors 76
• 4.3.3. Environmental Factors 81
• 4.4. Changes in Knowledge Dynamics at Asian Development Bank 82
• 4.4.1. Enhanced Knowledge Identification and Retrieval 82
• 4.4.2. Increased Visibility of Institutional Knowledge and Early Signs of Reuse 84
• 4.5. Summary 85
• Chapter 5. World Bank Group 88
• 5.1. Reclaiming the Knowledge Bank 88
• 5.2. AI-Powered Tools for Knowledge Management at the World Bank Group 91
• 5.2.1. AI-Powered Solutions to Core Knowledge Challenges 92
• 5.2.2. ‘Knowledge 360’: Integrating Knowledge Flows 95
• 5.2.3. Extending Thematic Knowledge Access 96
• 5.3. Factors Driving AI Adoption for Knowledge Management at the World Bank Group 98
• 5.3.1. Technological Factors 98
• 5.3.2. Organizational Factors 100
• 5.3.3. Environmental Factors 102
• 5.4. Changes in Knowledge Dynamics at the World Bank Group 104
• 5.4.1. Enhancing Knowledge Retrieval and Reuse through Knowledge 360 105
• 5.4.2. Codifying Tacit Knowledge for Systematic Knowledge Sharing 107
• 5.4.3. Connecting People to Experience at Speed 109
• 5.5. Summary 110
• Chapter 6. Inter-American Development Bank 113
• 6.1. Advancing Knowledge and Learning through Institutional Reform 113
• 6.1.1. Redefining Knowledge Effectiveness and Regional Public Goods 114
• 6.2. AI-Powered Tools for Knowledge Management at Inter-American Development Bank 116
• 6.2.1. Unified Knowledge Platform for Loan Operations 117
• 6.3. Factors Driving AI Adoption for Knowledge Management at Inter-American Development Bank 121
• 6.3.1. Technological Factors 121
• 6.3.2. Organizational Factors 125
• 6.3.3. Environmental Factors 129
• 6.4. Changes in Knowledge Dynamics at Inter-American Development Bank 132
• 6.4.1. Locating and Accessing Knowledge Faster 132
• 6.4.2. Strengthening Learning Loops through Reuse and Application of Knowledge 135
• 6.5. Summary 137
• Chapter 7. United Nations Development Programme 140
• 7.1. Reimagining Knowledge for Development in UNDP 140
• 7.1.1. Organizational Structure and Knowledge Architecture 141
• 7.2. AI-Powered Tools for Knowledge Management at UNDP 143
• 7.2.1. Decentralized and Multi-layered AI Adoption 144
• 7.2.2. Innovation and Functional Plurality 151
• 7.3. Factors Driving AI Adoption for Knowledge Management at UNDP 152
• 7.3.1. Technological Factors 153
• 7.3.2. Organizational Factors 156
• 7.3.3. Environmental Factors 160
• 7.4. Changes in Knowledge Dynamics at UNDP 163
• 7.4.1. Toward Evidence-Informed Practice 163
• 7.4.2. Accelerating Knowledge Search 165
• 7.5. Summary 166
• Chapter 8. Cross-Case Reporting 168
• 8.1. Overview of Cross-Case Findings 168
• 8.2. Cross-Case Summary of Enablers and Constraints using the TOE Framework 169
• 8.3. Cross-Case Summary of Changes in KM Processes 175
• 8.3.1. Knowledge Identification 175
• 8.3.2. Knowledge Creation 176
• 8.3.3. Knowledge Storage 178
• 8.3.4. Knowledge Sharing 180
• 8.3.5. Knowledge Use 181
• 8.3.6. Learn and Improve 183
• 8.4. Revisiting the Typology for Cross-Case Analysis 186
• 8.4.1. Centralized x Retrieval/Standardized Pattern 187
• 8.4.2. Centralized x Synthesis/Codification Pattern 190
• 8.4.3. Decentralized x Synthesis/Codification Pattern 191
• 8.5. Explaining Cross-Case Variation 192
• Part 3. Discussions and Conclusion 196
• Chapter 9. Proposal for Multi-Loop Model 197
• 9.1. Overview of the Multi-Loop Model 197
• 9.2. Transformation in KM Processes with AI Support 201
• 9.3. From KM Transformation to Organizational Outcomes 203
• 9.4. Mapping the Pathways by Type 205
• 9.5. Summary 213
• Chapter 10. Discussion 216
• 10.1. From AI-enabled KM to Evidence-Informed Operations 217
• 10.1.1. Lowering Coordination Costs 218
• 10.1.2. Increasing the Visibility of Shared Evidence 219
• 10.1.3. Embedding in Operational Workflows 221
• 10.1.4. Sharing Evidence Across Organizations 222
• 10.2. Institutional Coherence and Voice 224
• 10.2.1. Why Coherence Matters in Development Organizations 224
• 10.2.2. Coherence through the Lens of Common Knowledge 226
• 10.3. Implications for Development Results 227
• 10.4. Evolving Role of Development Practitioners in the AI Era 229
• Chapter 11. Conclusion 231
• 11.1. What Was Asked and What Was Answered 231
• 11.2. Contribution of This Study 233
• 11.3. Limitations 235
• 11.4. Suggestions for Future Studies 236
• Bibliography 240
• Abstract in Korean 265
• Acknowledgements 268
• Appendix: Interview Guide and Questions 269