인공지능과 저작권 제한의 헌법적 논의 — 학습데이터와 창작적 기여 및 변형물의 관계를 중심으로 —

This paper develops a constitutional discussion on the use of training data and the limitations of copyright—an aspect that has not been prominently addressed in previous debates on AI training data and copyright. To this end, this paper examines related topics such as the relationship between training data and creative contribution, as well as the relationship between training data and AI-generated derivative works. In conclusion, this paper confirmed that training data and AI-generated outputs are not situated in separate contexts from the perspective of creative contribution, but instead exist in a relationship of close interaction. In addition, stimulating creative incentives through copyright can protect the potential sources of training data. Thus, paradoxically, copyright may also function as an engine of AI development. Therefore, it may not be constitutionally desirable to codify an overly broad and general TDM exception into the Copyright Act. Meanwhile, the scope of protection for AI-generated derivative works is limited to those elements where additional human creative contributions are recognized. In particular, although it may be possible to identify certain additional human creative contributions in AI-generated derivative works, one can imagine scenarios where it is difficult to clearly distinguish and isolate those contributions from the original work’s expressive elements. In those scenarios, even if certain human creative contributions can be recognized in the derivative work, when it comes to the scope of protection one adopts the thin copyright protection theory.