Abstract: Human-machine collaboration will always be the basic form of diagnosis and treatment. According to different functional orientations, human-machine collaboration can be divided into two forms and two stages: medical AI-assisted stage, and medical AI-dominated stage. Different forms of human-machine collaboration determine different logic for identifying fault and ways of coexistence between humans and machines. In the stage of assisted diagnosis and treatment, medical AI only provides referential information for doctors. The legal status of medical AI is not fundamentally different from traditional medical devices. Human doctors should not rely too much on medical AI. In addition to referring to AI's suggestions, they should also comprehensively analyze other materials and not neglect the work that should be carried out without the help of medical AI. Form of Medical AI-domination should meet two legal conditions. Firstly, medical AI has significant and comprehensive advantages over human doctors in diagnosis and treatment. Secondly, medical AI has sufficient clinical applicability. When the legal conditions are met, human doctors should adjust their patterns of behavior in time. In principle, human doctors should faithfully convey and implement the decisions of medical AI, and should not veto or change them without careful procedures and sufficient reasons. At the same time, when there are obvious errors, misconduct, or other significant risks in the decisions of medical AI, human doctors should promptly request expert consultation and make final decisions prudently.