텍스트･데이터 마이닝 과정의 저작물 이용 면책 규정 신설안에 대한 소고 ‒적대적 생성신경망에의 적용‒

Provisions allowing for duplication and transmission of information analysis were established in Korea. It was because the current fair use clause lacks legal stability regarding whether companies using AI are exempted from copyright infringement when using information containing copyrighted works. However, the provisions are controversial whether the amendment can be resolved with the current fair use clause. The requirement to avoid responsibility will not be “used for people to feel or enjoy thoughts and feelings.” The words “feel” and “enjoy” have unclear interpretations, so this study examines how Japan interprets Article 30-4 of the Japanese Copyright Act. Through such, it was possible to check how the information analysis tools can be used when the provisions were applied, and to prepare examples for business operators and rights holders to interpret provisions. In addition, the article presents seven examples of the use of generative adversarial networks when training an AI model using data containing copyrighted works and evaluate whether they correspond to “feel” and “enjoy,”. It is difficult to say that the provisions have overcome the issue of legal stability of the fair use clause, given that each model has various purposes, and the type and use method of the work cannot be specified.