Stage-wise Efficiency and Cognitive Collaboration Mechanisms in AI-mediated 3D Animation Production = AI 개입 3D 애니메이션 제작의 단계별 효율과 인지 협업 메커니즘 연구

다국어 초록 (Multilingual Abstract)

With the rapid advancement of generative artificial intelligence (AI) technologies, 3D animation production is undergoing a profound paradigm shift; however, em- pirical evidence on how AI reshapes production efficiency and cognitive collab- oration mechanisms remains fragmented, limiting a holistic understanding of its systemic impact. This doctoral dissertation bridges the gap between macro-level production performance and micro-level human–machine interaction dynamics by adopting a socio-technical systems perspective and constructing an integrated framework to explain how AI reconfigures the technical and social subsystems of the production environment. To this end, the research employs a dual empirical design: Study 1 applies Network Data Envelopment Analysis (NDEA) to full-pro- cess data from ten animation projects to rigorously quantify stage-wise effi- ciency contributions of AI across the production pipeline, while Study 2 adopts an exploratory sequential mixed-methods approach—combining formative focus groups with a 2×2 mixed-factorial experiment involving 90 participants—to ex- amine how task attributes moderate the effects of AI tools as collaborative part- ners on designers’ cognitive load, self-efficacy, and output quality in divergent ideation and convergent execution tasks. Findings demonstrate that AI’s contri- bution to production efficacy is highly context-dependent: macro-level efficiency gains are not universal but vary significantly by stage, with clear advantages in highly standardized post-production and potential bottlenecks in animation pro- duction stages that demand complex creative performance; at the micro level, AI consistently enhances output quality across task types, while its effects on task duration, cognitive load reduction, and self-efficacy are significantly moderated by task characteristics, showing particularly strong advantages in convergent tasks. Synthesizing evidence across both studies, the dissertation proposes an Empirically-Grounded Dynamic Feedback Model that articulates dual feedback mechanisms triggered by AI intervention: a reinforcing loop in which cognitive offloading enables higher-order strategic iteration and efficacy enhancement, and a balancing loop in which over-reliance drives skill degradation and in- creased integration costs. The study contributes a systematic lens for under- standing the complex dynamics of human–AI collaboration and offers actionable guidance for educational and industry stakeholders to develop task-differentiated integration strategies that balance immediate performance gains with long-term capability development in digital transformation.

국문 초록 (Abstract)

생성형 인공지능(AI) 기술의 급속한 발전과 함께 3D 애니메이션 제작 분야는 심대한 패러다임 전환을 겪고 있으나, AI가 제작 효율성과 인지적 협업 메커니즘을 어떻게 재구성하는지에 대한 실증 연구는 여전히 파편화되어 있어 그 체계적 영향에 대한 총체적 이해가 제한적이다. 본 박사학위 논문은 사회기술시스템(socio-technical systems) 관점을 토대로, 거시적 제작 성과와 미시적 인간–기계 상호작용 역학 간의 연구 공백을 메우고 AI 기술이 제작 환경의 기술 하위시스템과 사회 하위시스템을 어떻게 재구성하는지를 설명하는 통합 프레임워크를 구축한다. 이를 위해 이중 실증 설계를 적용하였다. 연구 1은 10개 애니메이션 프로젝트의 전 공정 데이터를 활용하여 네트워크 데이터 포락분석(Network Data Envelopment Analy-sis, NDEA)으로 제작 단계별 AI의 효율 기여도를 엄밀하게 정량화하였고, 연구 2는 탐색적 순차 혼합방법(exploratory sequential mixed-methods) 설계를 채택하여 형성적 포커스 그룹과 90명을 대상으로 한 2×2 혼합 요인 실험을 결합함으로써, 발산적 발상(divergent ideation) 과제와 수렴적 실행(convergent execution) 과제에서 협업 파트너로서의 AI 도구가 디자이너의 인지부하, 자기효능감, 산출물 품질에 미치는 영향이 과제 속성에 의해 어떻게 조절되는지를 규명하였다. 분석 결과, AI의 제작 효율 기여는 높은 수준으로 맥락 의존적임이 확인되었으며, 거시 수준에서 AI 효율 증대는 보편적 현상이 아니라 단계별로 유의미한 차이를 보여 고도로 표준화된 후반작업(post-production) 단계에서는 뚜렷한 효율 우위를 보이는 반면, 복잡한 창의적 퍼포먼스를 요구하는 애니메이션 제작 단계에서는 잠재적 병목이 될 수 있음을 시사하였다. 또한 미시 수준에서는 AI가 과제 유형과 무관하게 산출물 품질을 일관되게 향상시키는 동시에, 수행시간 최적화, 인지부하 감소, 자기효능감 증진과 관련된 효과는 과제 특성에 의해 유의하게 조절되며 특히 수렴적 과제에서 결정적 우위를 보였다. 두 연구의 증거를 종합하여, 본 논문은 ‘실증 기반 동적 피드백 모델(Empirically-Grounded Dynamic Feedback Model)’을 제안하고, AI 개입이 촉발하는 이중 피드백 메커니즘—조작적 인지 자원을 방출해 고차원적 전략적 반복·개선을 촉진하는 강화 루프(reinforcing loop)와, 과도한 의존으로 인한 기술 저하 및 통합 비용 증가로 특징지어지는 균형 루프(balancing loop)—를 변증법적으로 설명한다. 본 연구는 인간–AI 협업의 복합 역학을 이해하기 위한 체계적 관점을 제공하는 동시에, 교육기관과 산업 조직이 과제 특성에 기반한 기술 통합 전략을 수립하도록 지원하는 실행 가능한 시사점을 제시하며, 디지털 전환에서 단기 성과 향상과 장기 역량 개발 간 균형의 필요성을 강조한다

목차 (Table of Contents)

• 1. Introduction 1
• 1.1 Research Background: The Generative AI Paradigm Shift in 3D Animation Production 1
• 1.2 Research Gaps and Objectives: Beyond Technological Determinism Towards a Systemic Analysis 4
• 1.3 Research Questions: Deconstructing the Production Landscape Transformation 7
• 1.4 Research Methods: A Dual Empirical Framework 9
• 2. Literature Review and Theoretical Framework 12
• 2.1 AI and the Socio-Technical Landscape of 3D Design 12
• 2.1.1 Technology Dimension 13
• 2.1.2 Process Dimension 15
• 2.1.3 Collaboration Dimension 18
• 2.1.4 Market Dimension 20
• 2.2 Theoretical Perspectives 21
• 2.2.1 Activity Theory and Distributed Cognition 22
• 2.2.2 Technology Acceptance and Experience Economy 25
• 2.3 Conceptual Framework for the AI-Mediated 3D Animation Production System 28
• 3. Study 1 — AI and Production Efficiency in 3D Animation 34
• 3.1 Research Questions & Scope 34
• 3.1.1 Objectives and RQ Mapping 34
• 3.1.2 Scope and Boundary Conditions 35
• 3.1.3 Expected Contribution to the Integrated Socio-Technical Model 36
• 3.2 Data and Variables 36
• 3.2.1 Stages of the Production Pipeline 36
• 3.2.2 Efficiency Indicators and Inputs/Outputs 40
• 3.3 Research Method: Network Data Envelopment Analysis 43
• 3.4 Results and Robustness Tests 46
• 3.4.1 Overall Efficiency Distribution 47
• 3.4.2 Stage-Wise Efficiency Analysis 49
• 3.4.3 AI Integration and Efficiency Correlation 51
• 3.5 Discussion of Findings 55
• 4. Study 2 — Human–AI Collaboration in Creative Workflows 58
• 4.1 Research Questions & Scope 58
• 4.1.1 Objectives and RQ Mapping 58
• 4.1.2 Scope and Boundary Conditions 60
• 4.1.3 Expected Contribution to the Integrated Socio-Technical Model 61
• 4.2 Methodology 63
• 4.2.1 Focus Group Study 63
• 4.2.2 Task-Based Experimental Study 65
• 4.3 Quantitative Experimental Results 72
• 4.3.1 Descriptive Statistics and Rater Reliability 73
• 4.3.2 Hypothesis Testing: Quality Indicator 74
• 4.3.3 Efficiency and Subjective Experience 76
• 4.4 Discussion of Findings 79
• 5. Integrated Discussion and Implications 82
• 5.1 Synthesizing Evidence Across Studies 82
• 5.2 An Empirically-Grounded Dynamic Feedback Model of Human–AI Collaboration in 3D Design 85
• 5.2.1 The Reinforcing Loop (R1): Cognitive Offloading and Efficacy Enhancement. 86
• 5.2.2 The Balancing Loop (B1): Dependency Risk and Skill Erosion 87
• 5.2.3 Synthesizing Study 1 and Study 2: The Model's Empirical Grounding 88
• 6. Conclusion 89
• 6.1 Research Findings and Contributions 89
• 6.2 Practical Implications 93
• 6.3 Limitations and Future Research 97
• Reference 102