重构人工智能风险治理:从内部视角到外部视角

Refactoring ai risk governance:from inside perspective to outside perspective

学界对人工智能风险治理的研究集中在基于内部视角的人工智能合规制度,无法全面揭示人工智能风险治理的制度内核。从风险治理的外部视角出发,将人工智能合规置于政企关系的框架下,人工智能合规的基本理念将不再是纯粹的自我规制,而是自我规制与政府规制的互动。在这种视角转变下,平台企业与政府之间形成关于人工智能治理任务分配的“发包”关系,体现为一种平台发包制,可以具体展开为四个子制度:授权机制是平台发包制的制度前提,即政府基于信任授予平台企业规制权力,这种授权是通过隐私政策证成的;评估机制用于衡量平台企业的人工智能风险治理能力;激励机制与考核机制都旨在提高平台企业的人工智能风险治理能力,前者通过锦标赛理论推动平台企业参与依据人工智能风险治理能力的竞争,后者通过金字塔理论推动平台企业形成体现人工智能风险治理能力的指标。

The academic research on artificial intelligence risk governance focuses on the artificial intelligence compliance system based on the internal perspective,which is unable to fully reveal the institutional core of artificial intelligence risk governance.From the external perspective of risk governance,if AI compliance is placed under the framework of government-enterprise relationship,the basic concept of AI compliance will no longer be pure self-regulation,but the interaction between self-regulation and government regulation.Under such a change of perspective,a"outsourcing"relationship is formed between platform enterprises and the government on AI governance task assignment,which is reflected as a platform outsourcing system and can be specifically expanded into four sub-systems:authorization mechanism is the institutional premise of platform outsourcing system,that is,the government grants regulatory power to platform enterprises based on trust,and such authorization is certified through privacy policies;The evaluation mechanism is used to measure the artificial intelligence risk management ability of the platform enterprises;Both incentive mechanism and assessment mechanism aim to improve the AI risk governance ability of platform enterprises.The former promotes platform enterprises to participate in the competition based on AI risk governance ability through tournament theory,while the latter promotes platform enterprises to form indicators reflecting AI risk governance ability through pyramid theory.