Abstract: While Artificial Intelligence (AI) technology is reorganizing human society disruptively, it also brings severe challenges to all aspects of society. In order to meet these challenges, Chinese scholars took the lead in proposing the concept and framework of Artificial Intelligence Social Experiments (AISE) in 2019, exploring the Chinese road of intelligent governance. AISE means introducing an experimental approach to society, where new technologies are placed in society to experiment to see their actual social impact. This is because that the disruptive nature of AI technology will induce a series of risks, but these risks are often difficult to predict and judge, therefore, artificial intelligence technology can be partially introduced in social scenarios, and its feasibility can be tested closely by monitoring its social impact. As a social experiment, artificial intelligence has inherent advantages in preventing and controlling risks, but its implementation also has ethical issues that cannot be ignored. This is mainly due to the fact that AISE takes humans as the research object and examine the internal relationship between human and AI. Taking human as research objects requires that AI social experiments be conducted prudently, respectfully, and rationally, with comprehensive ethical review and evaluation to ensure minimal risks and optimal methods. Therefore, its ethical concerns are not only important but necessary. The reason why AISE needs ethical concerns urgently stems from their ethical characteristics, especially the ethical characteristic of second-order and the unpredictability and uncertainty closely related to it. The fundamental feature of AISE is its ethical characteristic second-order. It is shown that one of the purposes of social experiments on AI is to address the ethical issues of AI through social experiments. However, at the same time, such experiments are controversial inherently in ethics which means there are ethical problems in these experiments when they are tried to solve ethical problems. The following characteristics are unpredictability and uncertainty. The former refers to the potential ethical hidden dangers of AISE that are sometimes difficult to judge in advance, and the latter indicates the fact that the ethical consequences of AISE are uncertain, and their solutions are difficult to predetermine. Another reason for ethical concerns about AISE originates from specific ethical issues raised by these characteristics, including privacy issues, issues of dirty hands and deception issues. Based on this, our paper put forward an ethical framework for AISE, that is, the ethical norms that should be followed to carry out artificial intelligence social experiments, in order to deal with the potential ethical risks of AI social experiments. The basic principles of AISE are respect for autonomy, beneficence, non-maleficence and justice, which is in the light of the principles of biomedical ethics explained by Beauchamp and Childress. Based on the demonstration of the criterion principle and generative logic of AISE ethics, this paper puts forward relevant suggestions for its ethical governance. In conclusion, it is necessary to understand the ethical necessity of AISE correctly, and construct an ethical framework for AISE to perceive, evaluate and reduce the potential risks in the introduction of AI technology into society. This will provide solutions to bridge the gap between what and how in scientific and technological ethics governance, so as to explore the Chinese path of ethical governance of science and technology.