ABSTRACT

The data from 10 billion sentences, originally collected for a dating counselling service, were used to develop and operate an AI chatbot, Lee-Luda. However, the chatbot company was fined by the South Korean government for violating the Personal Information Protection Act (PIPA). The case of Lee-Luda is the first case in South Korea that raised the question as to whether the use of personal information for AI learning data falls outside the scope of the original purpose of collection. Although the Lee-Luda is a South Korean case, since the prohibition on using personal information for purposes other than the original purpose of collection is a globally accepted principle, it is expected that the Lee-Luda case will provide meaningful implications not only for South Korea but also for law enforcement in other countries. Similar ethical and legal issues will likely arise in other countries in the foreseeable future because using personal information as learning data for an AI program may conflict with the existing legal principle that requires using personal information only for the original purpose of collection. In this paper, we analyse why the Lee-Luda program’s use of personal information for AI learning data was ruled to violate the Personal Information Protection Act. In addition, we suggest alternative ways for AI services that use personal information as learning data to comply with the law. Therefore, we believe that this paper provides a useful case study for AI operators in other countries about AI programs and personal information protection.

摘要

最初为约会咨询服务收集的 100 亿个句子的数据被用于开发和操作 AI 聊天机器人 Lee-Luda。然而，这家聊天机器人公司因违反个人信息保护法（PIPA）而被韩国政府罚款。Lee-Luda案是韩国首例提出AI学习数据使用个人信息是否超出最初收集目的范围的案件。Lee-Luda案虽然是韩国案例，但由于禁止将个人信息用于原始收集目的以外的目的是全球公认的原则，因此预计Lee-Luda案将不仅对韩国产生有意义的影响韩国也适用于其他国家的执法。在可预见的未来，其他国家可能会出现类似的伦理和法律问题，因为将个人信息用作人工智能程序的学习数据可能与要求仅将个人信息用于原始收集目的的现有法律原则相冲突。在本文中，我们分析了为什么Lee-Luda程序将个人信息用于AI学习数据被裁定违反《个人信息保护法》。此外，我们建议人工智能服务使用个人信息作为学习数据的替代方式，以遵守法律。因此，我们认为本文为其他国家的人工智能运营商提供了一个关于人工智能程序和个人信息保护的有用案例研究。