中圖分類號：D922；TP399文獻標識碼：ADOI:10.19358/j.issn.2097-1788.2024.06.009
引用格式：劉學榮.從失范到規(guī)范：生成式人工智能的監(jiān)管框架革新［J］.網(wǎng)絡(luò)安全與數(shù)據(jù)治理，2024，43（6）：58-63，71.

From illegal to legal: evolving regulatory frameworks for generative artificial intelligence

Abstract： The risk of aberration caused by generative artificial intelligence under technological change challenges the existing artificial intelligence regulatory system. Starting from the underlying technical mechanism, it can be seen that the current generative artificial intelligence presents a hierarchical format of "basic model-professional model-service application", and faces regulatory challenges such as the failure of algorithm supervision tools, the intensified risk of training data infringement, the unclear legal positioning between different levels, and the unclear division of responsibility boundaries. Therefore, it is necessary to take layered regulation as the logical core and reform the existing artificial intelligence regulatory framework in China. In the way of supervision, we should make good use of technology supervision tools such as prompt engineering and machine forgetting. In the delineation of responsibilities, the main body should be disassembled and hierarchical backtracking should be carried out, so as to standardize the hierarchical regulatory framework of "basic model-professional model-service application", in order to achieve effective supervision and promote the healthy and high-quality development of generated artificial intelligence.

Key words : generative artificial intelligence; algorithm black box; technical supervision; legal responsibility