생성형 AI의 저작권 침해 메커니즘과 책임귀속에 관한 연구 = A Study on the Copyright Infringement Mechanisms of Generative AI and the Attribution of Liability|RISS 상세보기

국문 초록 (Abstract)

본 연구는 생성형 인공지능으로 인한 저작권 침해를 개별 출력 물과 기존 저작물 간의 사후적 유사성 판단 문제가 아니라, 생성 시스템의 내부 기술 구조에 내재된 구조적 위험으로 재개념화한 다. 출력 중심 분석을 넘어, 침해를 생성 파이프라인 전반에서 발생하는 과정 기반 현상으로 파악한다.
본 연구는 생성형 인공지능의 작동을 학습, 표상 형성, 잠재 변환, 확률적 샘플링, 출력 및 유통으로 이어지는 연속적 단계로 분해한다. 저작권 침해 위험은 관찰 가능한 출력 이전에, 훈련 데이 터의 표현적 특성이 고차원 잠재공간에 압축·추상화·잔존하는 표상 단계에서 이미 구조적으로 형성될 수 있음을 논증한다.
본 연구는 내부 표상의 기술적 비가시성이 법적 귀속 실패의 주요 원인임을 규명한다. 표상은 외부에서 직접 관찰하거나 검증할 수 없기 때문에, 침해 위험은 특정 인간 행위자에게 추적되지 않은 채 실현될 수 있으며, 이는 기존 저작권법과 불법행위법이 전제해 온 행위자 식별 구조로는 충분히 대응할 수 없는 ‘행위자 부재(actio sine autore)’ 상태를 초래한다.
이에 본 연구는 기술 단계별 통제 가능성, 위험 예측 가능성, 기능적 기여를 기준으로 책임을 재배치하는 통제 기반 책임론을 제시한다. 또한 개발자, 플랫폼 운영자, 이용자의 책임이 중첩적으로 작동하는 다중 책임 구조를 제안하고, 생성형 인공지능을 단일 산출물이나 중개 행위가 아닌 통합된 서비스로 파악하는 ‘AI 서비스 결함책임’ 개념을 도입한다. 이 연구는 생성형 인공지능에 내재된 구조적 위험에 대응하기 위한 향후 입법·정책·사법 판단의 새로운 이론적 토대를 제공한다.

번역하기

본 연구는 생성형 인공지능으로 인한 저작권 침해를 개별 출력 물과 기존 저작물 간의 사후적 유사성 판단 문제가 아니라, 생성 시스템의 내부 기술 구조에 내재된 구조적 위험으로 재개념...

목차 (Table of Contents)

I. 서 론 ··················································································· 1
1. 연구의 배경 및 문제의식 ··································································· 1
2. 문제제기와 연구질문 ··········································································· 3
1) 기술구조적 측면 ···························································································· 3

I. 서 론 ··················································································· 1
1. 연구의 배경 및 문제의식 ··································································· 1
2. 문제제기와 연구질문 ··········································································· 3
1) 기술구조적 측면 ···························································································· 3
3) 책임법적 측면 ································································································ 5
4) 국제 규범 및 분쟁사례 측면 ······································································ 5
3. 연구의 목적 및 범위 ··········································································· 6
1) 연구목적 ·········································································································· 6
2) 연구범위 ·········································································································· 8
4. 연구방법 및 연구설계 ······································································· 11
1) 기술 분석(Technical Analysis) ······························································· 11
2) 법적 구조 분석(Legal Structural Analysis) ········································· 12
3) 정책 및 사례 분석(Policy & Case Analysis) ······································ 13
4) 책임구조 분석(Responsibility Analysis) ············································ 13
5) 연구설계의 통합(Integrated Modeling) ·················································· 14
5. 연구모델 개요 ····················································································· 14
1) 구조 모델(Structure Model) ····································································· 14
2) 귀속 모델(Attribution Model) ·································································· 15
3) 연구모형 도식화 ·························································································· 16
II. 국내외 선행연구 분석 ·················································· 18
1) 인간 저작자성 논쟁 중심 연구의 흐름 ·················································· 18
2) ‘실질적 유사성·의거성’ 판단 개혁 논의 ················································· 21
3) 인공지능 출력물의 저작물성 논의 ·························································· 23
4) 텍스트·데이터 마이닝(TDM) 연구와 데이터셋 적법성 분석 ··········· 29
2. 생성형 AI 기술구조를 다룬 공학 연구 ········································· 32
1) 표상학습(representation learning) 연구 ················································· 33
2) 잠재공간(Latent Space) 분석 연구 ························································· 37
3) 기억오염(Memorization)·데이터 추출 연구 ··········································· 41
4) 프롬프트 조향·모델 조작 연구 ································································ 43
3. 국제 정책·정부·규제기관 보고서 분석 ·········································· 46
1) 미국 "U.S. Copyright Office AI 보고서(Part 1–3)" ························· 46
2) EU "AI Act / AILD / PLD / EUIPO 보고서" ·································· 48
3) 일본 "문화청 AI·저작물성 기준 보고서" ··············································· 50
4) 중국 "Deep Synthesis Regulation" ························································· 51
5) 인도 "DPIIT 백서" ····················································································· 52
4. 주요 국제 분쟁사례에 대한 기존 연구의 분석 ··························· 52
1) NYT v. OpenAI ·························································································· 52
2) Getty v. Stability AI ················································································· 53
3) GitHub Copilot ···························································································· 54
4) Stable Diffusion·LAION 데이터 논쟁 ··················································· 54
5) 기존 학계·정책 비평의 한계 ···································································· 55
5. 선행연구의 구조적 한계 ··································································· 56
1) 출력 중심 연구의 편향 ·············································································· 56
2) 기술 파이프라인 분석의 결여 ·································································· 56
3) 행위자 부재 문제의 이론화 부족 ···························································· 57
4) 통제기반 책임의 불완전성 ········································································ 57
5) 국가 간 비교연구의 단편성 ······································································ 58
6) 데이터셋 투명성·감사체계 과소 평가 ···················································· 58
6. 소결 ····································································································· 59
1. 생성형 AI의 기술적 배경 ································································· 60
1) 서언 ················································································································ 60
3) 머신러닝의 기초 ························································································ 62
4) 생성형 언어모델과 멀티모달 모델 ·························································· 63
5) 훈련 데이터의 획득 및 특성 ·································································· 64
6) 훈련과정의 구조 및 메모리화(Memorization) ······································ 65
7) 출력 생성 및 배포 ······················································································ 65
8) 기술적 특성 및 법률적 함의 ···································································· 66
2. 학습 단계의 침해 위험 형성 ··························································· 67
1) 서언 ················································································································ 67
2) 트랜스포머의 단계적 구조와 침해 가능성 ············································ 67
3) 침해 발생 가능 지점의 3단계 구조 ························································ 68
4) 기술 파이프라인 분해와 귀속모델의 전제조건 ···································· 69
3. 표상 단계의 구조적 침해 위험 ······················································· 70
1) 표상 인코딩의 복제성 문제 ······································································ 70
2) 표상과 기억오염의 구조 ············································································ 71
3) 표상의 법적 객체화와 책임귀속의 전환 ················································ 74
4. 출력 단계와 위험의 가시화 ····························································· 75
5. 다층적 행위자 구조와 귀속 ········································································· 77
6. 소결 ··················································································································· 80
IV. 책임귀속의 구조화 ······················································· 82
1. 행위자 없는 침해와 법적 귀속 문제 ············································· 82
1) 행위자 부재의 법적 공백 ·········································································· 82
2) 법이론적 접근: 통제력 중심 귀속이론 ··················································· 83
3) 국가별 귀속구조와 단계기반 모델과의 충돌 ········································ 84
(1) 미국의 현행 귀속구조 ············································································ 84
(2) 유럽연합(EU)의 현행 귀속구조 ···························································· 86
(3) 한국의 현행 귀속구조 ············································································ 88
4) 결어 ················································································································ 91
2. 기존 법체계와의 충돌 및 조정 ······················································· 93
1) 법체계의 비대칭성과 AI 책임의 구조적 괴리 ······································ 93
2) 생성형 AI 단계별 침해구조와 법적 충돌 ·············································· 93
(1) 저작물성의 인간중심성 한계 ································································ 96
(2) 복제권·전송권과 표상 학습의 충돌 ····················································· 97
(3) 정책적 조정 가능성 ················································································ 98
4) 민법과의 충돌: 불법행위 책임 구조의 재편 ······································· 100
5) 형법과의 충돌: 행위주체 부재와 형사책임의 한계 ··························· 103
6) 법체계 조정의 방향성: 다층적 통합 모델 ··········································· 104
7) 행위자 중심 귀속의 한계와 전환 필요성 ············································ 108
8) 결어 ············································································································ 110
3. 개발자·플랫폼·이용자 다중책임 모델 ·········································· 111
1) 개발자·플랫폼·이용자의 책임집중도 ··················································· 111
2) 다중책임모델의 필요성 ············································································ 114
3) 개발자의 책임범위 : 구조적 통제력의 원칙 ······································· 116
4) 플랫폼의 책임범위: 경제적 이익과 관리의무 ····································· 117
5) 이용자의 책임범위: 예견가능성과 주의의무 ······································· 118
6) 결어 ·············································································································· 120
4. 다중 책임 모델의 정책적 확장 ····················································· 122
1) 논의 필요성 ································································································ 122
2) 책임 매트릭스와 다중책임의 구조화 ···················································· 125
3) 합의모형의 법적 함의 ·············································································· 125
4) 결어 : 책임 매트릭스의 이론적 의의 ··················································· 127
5. 소결 ····································································································· 128
V. 구조적 침해 위험에 대한 통합적 대응 ················· 130
1. 라이선스·보상 메커니즘 ·································································· 130
2. 투명성 의무 ······················································································· 132
3. 기술적 완화 조치와 책임 감경 구조 ··········································· 133
4. 소결 ····································································································· 134
VI. 결론 ·············································································· 136
1. 연구문제에 대한 종합적 응답 ······················································· 136
2. 연구결과의 이론적·학문적 함의 ···················································· 137
3. 정책적·제도적 시사점 ······································································ 138
4. 연구의 한계와 향후 연구과제 ······················································· 139
5. 전체소결 ····························································································· 140
참고문헌 ······························································································· 141
Abstract ································································································ 148

상세검색

RISS 보유자료

상세검색

해외전자자료

생성형 AI의 저작권 침해 메커니즘과 책임귀속에 관한 연구 = A Study on the Copyright Infringement Mechanisms of Generative AI and the Attribution of Liability

부가정보

분석정보

연관 공개강의(KOCW)

이 자료와 함께 이용한 RISS 자료

나만을 위한 추천자료