高風險人工智能的法律界定及規制

人工智能技術的底層科學邏輯，決定了其風險可認知、可描述、可分析、可界定。與歐盟《人工智能法》確立的一般風險治理進路不同，我國可採取“風險+情境融合治理”的進路，以高質效應對高風險人工智能的系統風險。“高風險”的界定標準，可基於人工智能本體能力強弱、功能作用對象、潛在致害程度等三個維度進行確認。首先，風險級別與人工智能本體強弱呈正相關關係，強人工智能與超強人工智能均屬高風險系統;其次，直接作用對象為“重大安全”，涵蓋國家安全、社會公共安全、個人生命安全以及其他重要基本權利安全;再次，存在對“重大安全”造成實質性顯著減損的可能。鑒於此，高風險人工智能的治理與規制，主要是在特定情境中針對其安全性展開，應遵循合法性原則、科技倫理約束原則、技術治理原則，秉持包容審慎與合作規制理念。以情境作為治理單元，以安全維護為首要目標，採取“行為規制+個體賦權”立法模式，進行事前、事中、事後的全過程治理，立法上最終達致制定《人工智能安全法》的目標。

The underlying scientific logic of artificial intelligence technology determines that its risks are cognizable, describable, analyzable and definable. Unlike the general risk governance established by the European Union's Artificial Intelligence Act, China can adopt the approach of “risk + contextual integration governance” to respond to the systemic risks of High-risk artificial intelligence with high quality. The so-called “high-risk” definition standard can be recognized based on three dimensions, such as the strength of the AI ontology, the function of the object, and the degree of potential harm. Firstly, the risk level is positively correlated with the strength of the AI body, and both strong AI and superintelligent AI are high-risk systems. Secondly, the direct object of its function is “significant security”, covering national security, public security, personal life security and other important basic rights security. Thirdly, there is a possibility of substantial impairment of “significant security”. In view of the above standard, the governance and regulation of high-risk AI is mainly focused on its safety, and should follow the principle of legality, the principle of scientific and technological ethical constraints, the principle of technological governance, and uphold the concept of inclusive prudence and cooperative regulation. Taking the situation as the governance unit, with safety maintenance as the primary goal, and adopting the legislative model of behavioral regulation combined with individual empowerment, the whole process of governance is carried out before, during and after the event. Ultimately, the legislative goal is to enact the “Artificial Intelligence Security Law”.