人工智能治理原则的实践导向:可靠性、问责制与社会协同

Practical Orientation of AI Governance Principles:Reliability,Accountability and Collaboration

人工智能的迅速发展与应用带来了难以避免的治理缺口。为了寻求敏捷治理,全球范围内陆续出台了一批具有"软法"性质的治理原则。通过对88份原则文本的整合与分析,聚焦其中的实践准则,研究发现在激增的文本和多元化的主张下,存在着一个公共的问题域:人工智能系统应该满足怎样的技术特性?应该如何应对技术的社会后果?应该以怎样的社会组织形式开展治理实践?尽管围绕这些问题不同原则的主张之间仍然存在张力,但不可否认,当下人工智能治理领域已经形成了一个具备共识基础的行动框架。基于文本分析,众多原则条目可以归纳为三项实践准则:可靠性原则、问责制原则和社会协同原则,它们分别回应了上述问题。其中,可靠性原则和问责制原则设定了构建符合期望的技术系统、应对技术的社会后果这两组相互补充的治理目标,而社会协同原则探讨了怎样的社会行动结构能够支持和保障这些目标的实现,三者构成了人工智能治理的行动框架。

Artificial intelligence has created an unavoidable governance gap as it is rapidly deployed and applied.In the quest for agile governance,many principles of a ‘soft law' nature have emerged around the world.This study analyzed the governance principles announced in 88 texts and focused on their practical guidelines.We found that underneath their diverse claims,there was a common domain of questions.What technical characteristics should AI systems meet? How should we deal with the social consequences of this technology? What forms of social organization should be used for governance practices? Although tension remains among the claims of different principles around these questions,it is undeniable that a consensus-based framework for action has been developed in the field of AI governance.Based on our textual analysis,the principles can be grouped into three codes of practice:reliability,accountability and collaboration,which respond to each of these questions.The principles of reliability and accountability set two sets of complementary governance goals:constructing technology systems that meet expectations and address the social consequences of technology.The principle of collaboration explores what social action structures can support and guarantee the achievement of these goals.These goals form the framework for action on AI governance.