하이데거의 존재론을 통한 AI 애니메이션의 4S 구조 분석 = An Analysis of the 4S Structure of AI Animation through Heidegger's Ontology

다국어 초록 (Multilingual Abstract)

This paper interprets the generative artificial intelligence animation production pipeline based on Martin Heidegger's ontology, aiming to philosophically elucidate the conditions for its establishment as a work and its structure of meaning disclosure, moving beyond discussions centered on technical efficiency. Since 2022, the emergence of generative AI and its subsequent gradual integration into the entire animation production process has rapidly expanded an environment where 3D characters and animations are automatically generated through text-to-image input. However, existing research and industry reports have primarily focused on efficiency—tool performance, reduced production time, cost savings, and workflow optimization—failing to sufficiently clarify theoretically when and on what grounds the output of AI-assisted production, i.e., the work itself, is recognized as a ‘work,’ and to whom and through what channels its validity and responsibility are attributed. This paper seeks to answer fundamental questions: When and under what conditions can AI-generated outputs be justified as works? Where do the subject and point of disclosure (Alētheia) and recognition, in the Heideggerian sense, reside? How do revelation/concealment and bias operate as automation levels increase, and how can these be controlled and managed? The research methodology adopts a qualitative approach based on conceptual analysis. Six entities derived from Heidegger's ontology are established as the units of analysis: Human(H), Artificial Intelligence(AI), Work(W), Data(D), Client(C), and Standard(S). The AI-based production structure is reconstructed around their interrelationships. Key analytical axes include Heidegger's concepts of ‘being-in-the-world’, ‘being-with’, ‘Gestell’, and ‘intermediary world-building’. To complement Heidegger's concepts, Don Idhe's post-phenomenology(embodiment, interpretation, alterity, background relations), Andrew Feenberg's philosophy of technology (technical design and social code), and Luciano Floridi's information ontology and information ethics. Furthermore, it reinterprets Parasuraman's 10-level automation scale for decision-making and action selection, establishing four relational models: ‘H-H(Human-Human)’, 'H-AI-H(Human Initiation-AI Assistance-Human Approval)‘, 'AI-AI-H(AI Chain Generation-Human Final Approval)’, and ‘AI-AI(AI Fully Automated Chain)’. It compares how the position of meaning initiation, the approving entity and timing, the disclosure/concealment mechanism, and the attribution of justification differ across each model. The analysis proceeds by tracing the work's formation path through stages: ‘from meaning initiation to interpretation, and from final approval to documentation’. This is systematized into a structural diagram intersecting the six ontological axes with the four relational model axes. The research findings reveal: First, within the AI-based animation production pipeline(AI-Pipeline), distinct from the traditional 3D CGI animation production pipeline(Traditional-Pipeline, hereafter T-Pipeline), the establishment of the existence of the work(W) is not completed solely by AI's generative act. It is only when it undergoes human(H) interpretation, modification, and final approval that it is recognized as a ‘work(W)’. What AI presents is merely a potentiality based on statistical patterns; it only transforms into a work through human meaning-assignment, editing, contextualization, and approval. At this point, source, transformation, and decision-making metadata were logged alongside the work(W) and data(D), ensuring reproducibility and accountability. Second, from Heidegger's perspective of Being-in-the-world, the AI-based animation production pipeline(AI-Pipeline) revealed a structure where humans(H), AI, work(W), data(D), clients(C), and standards(S) are interdependently entangled within the field of the working world, initiating meaning. Therefore, pipeline design should be understood not as a simple sequence of tasks, but as organizing the structure of meaning formation and responsibility pathways among beings. Third, from Heidegger's perspective of Mitsein(being-with), the boundaries of meaning and responsibility within the shared world were preserved when AI was positioned not as an other replacing humans, but as an assistant mediating and amplifying the human interpretive world. Thus, in the H-AI-H and AI-AI-H models, a desirable coexistence structure emerged where AI handles mass variation and exploration, while humans(H) and clients(C) are responsible for identity maintenance, meaning judgment, and final approval. Fourth, from Heidegger's Gestell perspective, automation simultaneously possesses the ambivalence of expanding possibilities and strengthening domination. What is revealed and what is concealed was predetermined by the rules governing data(D) and standards(S). Fifth, the suitability of the four relationship models was summarized as follows: The H-H and H-AI-H models were suitable for tasks sensitive to identity and rights, while the AI-AI-H model could be efficiently applied to mass, repetitive tasks where speed, scale, and standardization are key. The AI-AI model must be limited to internal experimental and synthetic data generation domains equipped with isolation, non-distribution, watermarking, and kill switches, assuming a fully automated chain. It was confirmed that as automation levels increase, the technological lead required for meaning initiation grows, and validity is proportional to the quality of approval procedures, source management, distribution design, and auditability. The scope of this research was limited to the 3D CGI animation production pipeline, based on qualitative and conceptual analysis. Experimental verification in actual production environments, quantitative data analysis, and subsequent stages such as audience reception, distribution, and economic value are beyond the scope of this study. Generalization to other creative domains like 2D animation, live-action video, games, and interactive media requires additional adjustments. Furthermore, as the theoretical framework centers on Heidegger, Don Eide, Andrew Pinberg, and Luciano Floridi, comparative examination with other techno-philosophies and media theories—such as those of Bernard Stiegler, Gilbert Simondon, and Bruno Latour—remains a future task. Nevertheless, this paper has presented ontological criteria and conceptual guidelines necessary for designing and operating AI pipelines by reframing AI animation not merely as an issue of efficiency, but as a matter of the work's ontological constitution and its structure and system of responsibility. Future research should apply the analytical framework and relational model proposed here to actual studio and project cases for quantitative and empirical evaluation. It should also extend this work to other visual, gaming, and interactive media, and conduct follow-up studies linked to policy, governance, and copyright norm design.

국문 초록 (Abstract)

본고는 생성형 인공지능(Generative Artificial Intelligence) 애니메이션 제작 파이프라인(Pipeline)을 하이데거(Martin Heidegger)의 존재론을 기 반으로 해석하여, 기술적 효율성 중심 논의를 넘어 작품으로서의 성립 조건과 의미 개시 구조를 철학적으로 규명하고자 한다. 2022년 이후 생 성형 인공지능의 출현과 그후에 점차 애니메이션 제작 전 과정에 도입되 면서, 텍스트·이미지(Text to Image) 입력을 통해 3D 캐릭터와 애니메이 션이 자동 생성되는 환경이 빠르게 확산되었다. 그러나 기존 연구와 실 무 보고는 주로 도구 성능, 제작 시간 단축, 비용 절감, 작업 흐름의 최 적화 등 효율성에 초점을 두어, AI가 개입한 제작 과정의 생산물, 즉 작 품이 언제, 어떤 근거로 ‘작품’으로 인정되는지, 그리고 타당성과 책임이 누구에게 어떤 경로로 귀속되는지를 이론적으로 충분히 해명하지 못하고 있는 실정이다. 이에 본고는 AI가 만든 생산물이 언제·어떤 조건 아래 작품으로 정당화되는가, 하이데거적 의미의 개시(Alētheia)와 승인의 주 체·시점은 어디에 위치하는가, 자동화 수준이 높아질수록 드러냄/은폐·편 향은 어떻게 작동하며, 이를 어떻게 통제·관리할 수 있는가라는 근본적 인 의문에 답하고자 한다. 연구 방법론은 질적방법으로 개념적 분석에 기반한 이론적 연구를 채택하였다. 분석 단위로 하이데거의 존재론에 따른 여섯 존재자, 즉 인간 (Human, H), 인공지능(AI), 생산물(Work, W), 자료(Data, D), 클라이언 트(Client, C), 기준(Standard, S)을 설정했다. 이들의 상호관계를 중심으 로 AI 기반 제작 구조를 재구성한다. 이때 하이데거의 ‘세계-내-존재’, ‘더불어-있음’, ‘게슈텔’과 ‘상호매개적 세계구성’ 개념을 주요 분석 축으 로 삼았다. 하이데거의 개념을 보완하기 위해 돈 이데(Don Idhe)의 포스 트 현상학(체현, 해석, 타자, 배경 관계), 앤드류 핀버그(Andrew Feenberg)의 기술철학(기술 설계와 사회적 코드), 루치아노 플로리디 (Luciano Floridi)의 정보 존재론·정보윤리학을 아울러 ‘AI 기반 애니메이 션 제작 파이프라인(Artificial Intelligence-파이프라인, 이하 AI-파이프 라인)’을 해석하는 개념적 모델을 제시한다. 또한 파라슈라만 (Parasuraman)의 의사결정 및 행동 선택의 자동화 수준 10단계를 재해 석하여 네 가지 관계모델, 즉 ‘H(Human)-H(Human)(인간-인간)’, ‘H-AI-H(인간 개시-AI 보조-인간 승인)’, ‘AI-AI-H(AI의 연쇄 생성-인 간 최종 승인)’, ‘AI-AI(AI의 완전 자동 연쇄)’를 설정하고, 각 모델에서 의미 개시의 위치, 승인 주체와 시점, 드러냄/은폐 메커니즘, 근거의 귀 속이 어떻게 달라지는지를 비교한다. 분석 절차는 작품 성립 경로를 ‘의 미 개시에서 해석으로 그리고 최종 승인에서 기록’이라는 단계로 진행한 다. 그리고 이를 여섯 존재자 축과 네 가지의 관계모델 축을 교차한 구 조도로 체계화한다. 연구 결과는 첫째, 전통적인 3D CGI 애니메이션 제작 파이프라인 (Traditional-파이프라인, 이하 T-파이프라인)과는 다른 AI 기반 애니메 이션 제작 파이프라인(AI-파이프라인)에서 생산물(W)의 존재 성립은 AI 의 생성 행위 자체로는 완결되지 않으며, 인간(H)의 해석·수정·최종 승 인을 거칠 때 비로소 ‘작품적 생산물(W)’로 인정되었다. AI가 제시하는 것은 통계적 패턴 기반의 가능태에 불과하며, 인간의 의미 부여·편집·맥 락화·승인을 통해서만 작품으로 전환된다. 이때 출처·변환·의사결정 메타 자료는 생산물(W)과 자료(D)에 로그로 기록되어 재현성과 책임 귀속을 보증하였다.
둘째, 하이데거의 세계-내-존재(Being-in-the-world) 관점에서 AI 기반 애니메이션 제작 파이프라인(AI-파이프라인)은 인간(H)·AI·생산물(W)· 자료(D)·클라이언트(C)·기준(S)이 작업세계라는 장(Field) 안에서 상호의 존적으로 얽혀 의미를 개시하는 구조로 드러났다. 따라서 파이프라인 설 계는 단순한 작업 순서가 아니라, 존재자들 간의 의미 형성 구조와 책임 경로를 조직하는 일로 이해되어야 한다. 셋째, 하이데거의 더불어-있음(Mitsein) 관점에서 AI는 인간을 대체하 는 타자가 아니라, 인간의 해석 세계를 매개·증폭하는 조력자로 배치될 때 공동세계의 의미와 책임의 경계가 보존되었다. 그러므로 H-AI-H와 AI-AI-H 모델에서 AI는 대량 변주·탐색을 담당하고, 인간(H)과 클라이 언트(C)는 정체성 유지·의미 판단·최종 승인을 담당하는 구조가 바람직 한 공존 형태로 도출되었다. 넷째, 하이데거의 게슈텔(Gestell) 관점에서 자동화는 가능성의 확장과 지배의 강화라는 양가성을 동시에 지니며, 무엇이 드러나고 무엇이 은폐 되는지는 자료(D)와 기준(S)의 규칙에 의해 선제적으로 결정되었다. 다섯째, 네 가지 관계모델의 적합성은 다음과 같이 정리되었다. H-H, H-AI-H 모델은 정체성·권리 민감 과제에 적합하였으며, AI-AI-H 모델 은 속도·규모·표준화가 핵심인 대량·반복 작업에 효율적으로 적용될 수 있었다. AI-AI 모델은 완전 자동연쇄를 전제로 하되, 격리·비배포·워터 마킹·킬스위치를 갖춘 내부 실험·합성 자료 생성 영역에 한정되어야 한 다. 자동화 수준이 높아질수록 의미 개시의 기술 선행 비중이 커지고, 타 당성은 승인 절차·출처 관리·분포 설계·감사 가능성의 품질에 비례한다 는 점을 확인하였다. 연구의 범위는 질적·개념적 분석에 기반해 3D CGI 애니메이션 제작 파 이프라인으로 한정하였다. 실제 제작 현장에서의 실험적 검증이나 정량 적 데이터 분석, 관객 수용·유통·경제적 가치 등 후속 단계는 본 연구의 범위 외에 있으며, 2D 애니메이션·실사 영상·게임·인터랙티브 미디어 등 다른 창작 영역으로의 일반화에는 추가적인 조정이 필요하다. 또한 하이 데거와 돈 이데·앤드류 핀버그·루치아노 플로리디를 중심으로 이론 구도를 구성함에 따라 스티글레르(Bernard Stiegler), 시몽동(Gilbert Simondo), 라투르(Bruno Latour) 등 다른 기술철학·매체이론과의 비교 검토는 향후 과제로 남는다. 그럼에도 본고는 AI 애니메이션을 단순한 효율성의 문제가 아니라 작품의 존재 성립과 책임 구조와 체계의 문제로 재구성함으로써, AI-파이프라인을 설계·운영하는 데 필요한 존재론적 기 준과 개념적 지침을 제시하였다. 향후 연구에서는 본 연구에서 제안된 분석틀과 관계모델을 실제 스튜디 오·프로젝트 사례에 적용하여 정량적·실증적으로 평가하고, 다른 영상·게 임·인터랙티브 매체로 확장하는 작업, 그리고 정책·거버넌스·저작권 규범 설계와 연계한 후속 연구가 요구된다.

목차 (Table of Contents)

• I. 서 론 ····································································································1
• 1.1 연구 배경과 목적 ·····················································································1
• 1.2 연구 내용과 방법 ·····················································································2
• II. 이론적 배경 ····································································································7
• 2.1 애니메이션의 정의와 내용 ·····································································7
• 2.2 AI와 AI 애니메이션 ·············································································10
• 2.2.1 AI의 개념과 역사 ·············································································10
• 2.2.2 AI 애니메이션의 분류 ·····································································12
• 2.3 3D CGI 애니메이션 제작의 T-파이프라인과 AI 기반 제작의
• AI-파이프라인 ···················································································15
• 2.3.1 파이프라인의 의미와 기능 ······························································15
• 2.3.2 T-파이프라인과 AI-파이프라인의 비교 ·····································17
• 2.4 인간과 기계의 상호작용 이론 ·····························································20
• 2.5 AI 애니메이션 분석을 위한 하이데거의 존재론 ····························22
• 2.5.1 하이데거의 존재론과 작업세계의 구조 ·······································23
• 2.5.2 더불어-있음과 관계 구조 ·······························································25
• 2.5.3 게슈텔과 기술적 드러냄/은폐 구조 ·············································26
• 2.6 상호매개 구조와 존재 생성 ·································································27
• III. 하이데거의 존재자 기반 분석 설계 ·················································30
• 3.1 존재자 분류와 역할 ···············································································30
• 3.1.1 존재자 개념 및 설정 근거 ······························································30
• 3.1.2 존재자 유형별 개념과 역할 ····························································32
• 3.2 AI 애니메이션의 존재론적 접근 ························································33
• 3.2.1 AI 애니메이션의 존재구조 ·····························································35
• 3.2.2 AI 애니메이션의 관계구조(더불어-있음) ····································36
• 3.2.3 AI 애니메이션의 기술적 드러냄 구조(게슈텔) ··························37
• 3.2.4 AI 애니메이션의 상호매개 구조와 존재 생성 ···························39
• 3.3 관계모델의 유형과 네 가지 구조 설정 ···············································39
• 3.3.1 H-H 모델 ···························································································40
• 3.3.2 H-AI-H 모델 ····················································································40
• 3.3.3 AI-AI-H 모델 ···················································································42
• 3.3.4 AI-AI 모델 ························································································42
• 3.4 소결 ············································································································43
• IV. AI 기반 제작 파이프라인의 존재론적 구성과 해석 ························ 45
• 4.1 분석 대상과 방법론적 틀 ····································································45
• 4.2 AI-파이프라인의 존재론적 분석 ························································46
• 4.2.1 세계-내-존재 관점에서 본 AI-파이프라인 ································46
• 4.2.2 더불어-있음 기반의 인간과 AI-파이프라인의 공존 구조 ······ 48
• 4.2.3 게슈텔과 기술적 드러냄/은폐 문제 ··············································50
• 4.2.4 상호매개적 존재 생성과 창작 구조 ··············································52
• 4.2.5 소결 ······································································································55
• 4.3 관계모델 기반 제작 방식 분석 ···························································56
• 4.3.1 T-파이프라인과 관계모델 적용(H-H모델) ································ 56
• 4.3.2 AI-파이프라인의 관계구조 분석 ···················································58
• 4.3.2.1 H-AI-H 모델(ANI 개념) ··························································58
• 4.3.2.2 AI-AI-H 모델(AGI 개념) ·························································59
• 4.3.2.3 AI-AI 모델(ASI 개념) ·······························································62
• 4.4 소결 ··········································································································64
• Ⅴ. 결 론 ··································································································66
• 참고문헌 ·········································································································71
• 부 록 ··········································································································75
• Abstract ··········································································································79