인공지능 학습용 데이터세트에 대한 저작권법과 부정경쟁방지법상 보호와 그 한계

The competitiveness of AI technology is dependant upon good quality of trainable data sets. In other words, the quality of AI heavily relies on the acuteness of AI program and the contents of data sets to train AI. In AI-related industry, data sets to train AI plays a very important role in several stages such as creation of data sets to train AI, creation of complete model trained by AI and usage of AI-trained model. A producer of data sets to train AI should not infringe another’s copyright when he/she extracts and/or reproduce raw data. In this stage, text and data mining exceptions to copyright protection will be in favor of the producer of the data sets. In this regard, the right of the producer of the data sets needs to be protected under copyright and/or unfair competition laws. In this context, this Paper explores copyright protection and its limitations in terms of text and/or data mining in connection with data sets to train AI. Then, the Paper discuss unfair competition law issues, in particular, in terms of data offered only for specified persons under the Japanese unfair competition law. In concluding, this Article proposes how to efficiently protect the producer of data sets to train AI and to harmonize the interests and rights between the producer of date sets and the right holders of raw data.