生成式人工智能虚假信息风险与治理研究

Research on Risk and Governance of the Disinformation by Generative Artificial Intelligence

以ChatGPT为代表的生成式人工智能改变了人机交互、信息获取、信息传播的方式,但大语言模型的“幻视”问题也给虚假信息治理带来了新的挑战。训练数据瑕疵、算法黑箱以及大语言模型的“涌现能力”等因素致使虚假信息风险呈扩大趋势,对个人权益保护、社会稳定以及国家安全构成了严重威胁。《生成式人工智能服务管理暂行办法》明确要求不得利用生成式人工智能生成虚假信息,但在虚假信息的监管主体、监管职责及法律责任等方面缺乏具体规定,不足以有效应对生成式人工智能下的虚假信息风险。为破解当下困局,应当秉持包容审慎的治理态度,立足我国法治实践现状,从规范体系、治理主体、责任承担和制度环境等维度构建生成式人工智能虚假信息的治理框架。

Generative AI,represented by ChatGPT,has revolutionized the ways of human-computer interaction,information acquisition,and information dissemination,but the“hallucination”issue of large language models has also brought new challenges to the governance of false information.The“Emergent Ability”of large language models,combined with training data flaws and algorithm black boxes,has resulted in a remarkable growth of false information risks,posing a critical risk to personal rights,social stability,and national security.The“Interim Measures for the Management of Generative Artificial Intelligence Services”strictly prohibits generative artificial intelligence from being used to create false information;nonetheless,there are no precise regulations regarding the oversight body,oversight duties,and legal ramifications of false information,which are inadequate to efficiently manage the risk of disinformation by generative artificial intelligence.To solve the current predicament,we ought to keep an inclusive and prudent attitude towards governance,take into account the current state rule of law practice,and construct a framework to manage false data by generative AI based on aspects such as normative system,governance subject,responsibility and institutional environment.