Abstract: Through the analysis of the judicial decisions of 220 cases involving artificial intelligence, the paper finds that there are following prominent problems in the application of law in the cases involving artificial intelligence at present. Direct reference to the differentiation criteria of artificial intelligence in the field of computer science can not meet the needs of judicial practice. The efficiency of traditional courts in hearing cases is relatively low. There is an unclear understanding of the duty of care of artificial intelligence developers and users and there is a lack of in-depth investigation into the facts of cases in which employers dismiss employees in disguised form on the grounds of job transfer. Furthermore, there is a lack of proper understanding and application of new methods of gathering evidence without people involvement such as self-inspection of artificial intelligence, and there are difficulties in identifying the ownership of artificial intelligence works. To solve these problems, we should change our way of thinking. Artificial intelligence can be divided into instrumental and intelligent artificial intelligence based on whether the concrete (intellectual) achievements are created or whether their risks can be predicted in advance. Cases involving artificial intelligence should be under the jurisdiction of professional courts. The standards for determining the duty of care of artificial intelligence developers and users should be flexible. Employers are required to delimit a reasonable buffer period for the employees who are forced to be transferred before their salaries are substantially reduced. In addition, the qualified self-inspection reports of artificial intelligence are allowed to be used as evidence, artificial intelligence works can be temporarily classified as virtual property for protection, and they are required to be distinguished from human works during display.