生成式人工智能训练数据的风险与治理研究  

Research on Risk and Governance of Generative Artificial Intelligence Training Data

作　　者：陈宏光[1] 安诗凤 CHEN Hong-guang;AN Shi-feng(School of Law,Anhui University,Hefei 230039,China)

机构地区：[1]安徽大学法学院,安徽合肥230039

出　　处：《海南开放大学学报》2024年第1期110-118,共9页Journal of Hainan Open University

基　　金：2022年安徽省哲学社会科学规划青年项目“人脸识别技术在地方政府治理中的应用风险及其法律规制研究”阶段性成果(项目编号:AHSKQ2022D118)。

摘　　要：生成式人工智能技术目前正处于蓬勃发展时期,但其在训练数据收集和处理时,潜在的合规、泄露和偏见等风险时刻威胁着生成式人工智能技术的健康发展、扰乱社会秩序,成为生成式人工智能治理必须解决的问题。我国目前对生成式人工智能训练数据风险治理既有政策支持,又有法律保障。在这些基础上,行政机关可以建立生成式人工智能基础模型信息库,预防训练数据合规风险;也可以开发、使用合成数据,扩大训练数据来源,规避训练数据泄露风险;还可以通过适当细化训练数据审核制度和逐步实施生成式人工智能企业数据信用制度的方式,规范生成式人工智能训练数据使用行为,降低风险发生概率。Generative artificial intelligence(AI)technology is currently in a period of vigorous development,but its potential risks such as compliance,leakage and bias in training data collection and processing always threaten the healthy development of generative AI technology and disrupt social order,and become problems that must be solved by generative AI governance.At present,China has both policy support and legal protection for the risk governance of generative AI training data.On this basis,administrative organs can establish the generative AI basic model information database to prevert the compliance risk of training data,develop and use synthetic data to expard training data sources and avoid the risk of training data leakage,it is also possible to standardize the use behavior of generative AI training data and reduce the probability of risk occurrence by appropriately refining the training data review system and gradually implementing the generative AI enterprise data credit system.

关 键 词：生成式人工智能 训练数据 风险治理 

分 类 号：DF3[政治法律—宪法学与行政法学]