정신건강 영역에서 생성형 AI 규제에 관한 연구 : 미국 주법과의 비교를 중심으로

With the development of AI technology, AI is being utilized in many fields, and the mental-health domain is no exception. AI that allows users to share emotional rapport anytime and anywhere at low cost has rapidly acquired the status of a psychological companion, but it does not always lead to positive results. In the United States, there was a case in which a boy, while engaging in counseling with ChatGPT, made an extreme choice, and many U.S. states are enacting bills to regulate the use of generative AI in the mental-health field. In Korea as well, psychological counseling through ChatGPT is on the rise, yet related discussion remains limited. Focusing on the need for public-law regulation of generative AI in the mental-health area, this paper conducts a comparative-law analysis of legislative and regulatory trends in five U.S. states and sets out how Korea should design the contents and methods of the relevant regulatory system. First, the paper specifically presents the risk factors that generative AI may bring about in the mental-health domain and examines that the current legal framework cannot sufficiently prevent the concerned risks. Next, it reviews the main contents of the relevant laws of five U.S. states (New York, Utah, California, Illinois, and Nevada), derives implications for us from a comparative perspective, and, for its part, proposes regulatory principles suited to Korea. Considering the reality that a particular generative AI, ChatGPT, shows an overwhelming market share, the paper recommends that, rather than limiting the regulatory target to generative AI intended for use in the mental-health domain, general-purpose generative AI that can interact and commune with users so as to build emotional bonds be included within the scope of regulation; it also argues that risk-response protocols should be in place, including providing users at set intervals with notice that the conversational counterpart is not human and guidance to professional institutions when risk-indicating expressions are detected. In addition, it is necessary that an expert group verify whether generative AI can interact properly with users from a mental-health perspective. The use of generative AI in the mental-health domain is approaching as a threat to citizens’ lives and mental health, and the state must protect the people from such risks. Hereafter, continued research and the prompt establishment of an appropriate legal system capable of protecting the public from newly emerging risks to life and mental health will be required.