算法应当被解释吗?——人工智能“可控制”的治理向度  

Should Algorithms be Interpreted?——The governance dimension of‘controllable’AI

作　　者：袁曾 YUAN Zeng(Law School,Shanghai University,Shanghai 200444,China)

机构地区：[1]上海大学法学院,上海200444

出　　处：《法学论坛》2025年第1期130-142,共13页Legal Forum

基　　金：国家社科基金后期资助项目“人工智能的法律人格与未来发展研究”(22FFXB026)的阶段性成果。

摘　　要：人工智能的治理需要明确的法律指引。为应对算法等新技术引致的“算法黑箱”“算法歧视”等问题,学界提出了以“算法解释论”为代表的治理进路,若算法主体无法对算法决策做出合理解释则应承担相应责任。“算法解释论”在传统人工智能时代具备一定的合理性,但其对相关主体苛以过重责任义务,而单一环节的治理无法应对技术治理的整体要求。算法可以被解释,但以算法解释为核心的规制体系无法应对生成式人工智能问世后的治理难题,从成本与成效上无法涵盖发展“负责任的人工智能”的现实需要。结合欧盟、美国等域外立法经验与中国人工智能应用的现实优势,应将以“算法解释”为核心的立法进路修正为以“可控制”为核心的立体治理标准,以期用规则优势引领发展优势。The governance of AI needs clear legal guidance. In order to deal with the problems of ‘algorithm black box' and ‘algorithm discrimination' caused by new technologies, the academic community has proposed a governance approach represented by ‘algorithm interpretation theory'. If the algorithm subject cannot make a reasonable explanation for the algorithm decision, it should bear the corresponding responsibility. The ‘algorithm interpretation theory' has a certain rationality in the era of traditional artificial intelligence, but it imposes excessive responsibilities and obligations on relevant subjects, and the governance of a single link cannot meet the overall requirements of technical governance. Algorithms can be explained, but the regulatory system with algorithm interpretation as the core cannot cope with the governance problems after the advent of generative AI and cannot cover the realistic needs of developing ‘responsible AI' in terms of cost and effectiveness. Combined with the legislative experience of the EU, the United States and the practical advantages of China's AI application, the legislative approach with ‘algorithm interpretation' as the core should be amended to a three-dimensional governance standard with ‘controllable' as the core, in order to lead the development advantage with the advantages of rules.

关 键 词：算法解释 算法透明 生成式人工智能 数据确权 GDPR 

分 类 号：D922.17[政治法律—宪法学与行政法学]