Artificial intelligence is pushing society into the era of intelligence. Social experiment is considered to be an important way to systematically understand the social impact of artificial intelligence, but it also superimposes multiple potential risks. It urgently needs research on the formation of consensus experimental norms. Thus, the present paper first analyzes the potential risks of artificial intelligence social experiments from the perspective of technology, society, and their interactions. Then, by combining representative experiments and ethical norms in life medicine, psychology, social science, and anthropology fields, we put forward 9 experimental specifications including the norms of protection of object rights, the conditionality of experimental factors, the controllability of the experimental process, and the ethics of experimental benefits. Finally, we try to raise an artificial intelligence social experiment operating mechanism that emphasized on the diverse subject participation, operational feedback, and procedural justice that aiming to support a healthy development of artificial intelligence technology in China.