Abstract: In recent years, the breakthrough development of artificial intelligence technology has brought a lot of convenience to human society, but at the same time, it has also brought risks and challenges that can not be underestimated. Therefore, a series of related research on the ethical governance of artificial intelligence have emerged. At present, the main governance strategies include the construction of ethical governance model and the advocacy of ethical service concept. This paper attempts to propose a third strategy as a parallel scheme, that is, to carry out ethical governance in artificial intelligence based on experimental methodology. The importance of carrying out ethical governance in artificial intelligence by experimental method lies in its ability to effectively improve the efficiency of identifying ethical risks, evaluating ethical governance framework and resolving ethical governance conflicts. Experimental approaches to ethical governance in artificial intelligence include computational experiments and social experiments. The former aims to simulate and test scenes with computer simulation technology, while the latter aims to study problems in real scenes through social investigation, social interview and other methods. These two experimental methods have their own implementation process and application scenarios, but there are also difficulties and problems in the actual development process. In general, carrying out ethical governance in artificial intelligence with experimental methods and exploring the risk impact mechanism through pre-designed controllable experiments are conducive to fundamentally improving the effectiveness of ethical governance in artificial intelligence and providing solutions for narrowing the gap from what it is to how to do in ethical governance in artificial intelligence.

摘要： 近年来，人工智能技术的突破性发展为人类社会带来了诸多便利，但与此同时也给人类社会带来了不容小觑的风险和挑战，由此涌现了一系列关于人工智能伦理治理的研究成果。目前，主要的治理策略包括建构伦理治理模型和倡导伦理服务理念，本文试图提出第三种策略作为平行方案，即以实验方法论为基础开展人工智能伦理治理。以实验方法开展人工智能伦理治理的重要性就在于能够有效提升识别伦理风险、评估伦理治理框架、化解伦理治理冲突的效能。以实验方法开展人工智能伦理治理包括计算实验和社会实验，前者旨在利用计算机仿真技术对场景进行模拟和测试，而后者则旨在通过社会调查、社会访谈等方法对真实场景中的问题展开研究。这两种实验方法有其各自的实施过程和适用场景，但也存在着在实际开展过程中的困难和问题。总的来说，以实验方法开展人工智能伦理治理，通过预先设计可控实验探索风险影响机制，有助于从根本上提升人工智能伦理治理的效能，为缩小人工智能伦理治理中“是什么”到“如何做”的鸿沟提供解决思路。