이루다 사건을 통해서 보는 개인정보의 인공지능 학습데이터 활용 가능성

Abstract

‘Iruda’ was a artificial intelligence chat bot whose learning data were Kakao Talk conversation sentences uploaded by users for psychological counseling purposes. However, the Personal Information Protection Committee (hereinafter referred to as the “PIPC”) imposed a fine on the grounds that Iruda used the personal information outside the scope of the purpose for which it was collected. Eventually, Iruda closed its service. This paper starts by analyzing in which passages Iruda was judged to be illegal, and intends to suggest a ‘legitimate alternative action’ should be taken in order for AI services that overcome Iruda’s illegality to appear in the future. A common solution would be using AI only within the scope of the purpose for which personal information was originally collected, or to de-identify the learning data if it exceeds the scope. However, in order to maintain the quality of AI, there are some cases that the use of the original data is inevitable. For this case, this paper propose an alternative solution; blocking the combination of AI learning data and original personal information, and filtering personal information from AI service operation data (instead of learning data). Since the ‘legitimate alternative action’ suggested in this paper was not actually Iruda’s action, it was not subject of PIPC’s judgement. Therefore, it cannot be concluded that this alternative will necessarily become legal. However, according to this paper, we can get hints about what safeguards should be put in place to properly control the risk of illegality when using user data for machine learning.