The biggest difference between artificial intelligence products and ordinary products is their autonomy.On the one hand,the operation mode of artificial intelligence products is more complicated.They can make independent decisions and act independently,and are in the mode of "perception-analysis-action".Their operation process is a "black box",which is difficult to be explored by human beings.On the other hand,the subject of product infringement is more diversified.In addition to the producers and sellers who are the subject of traditional product infringement,they also include the researchers and even the artificial intelligence products themselves.Based on this,the current judicial identification of product defects is difficult.First,the existing judgment criteria lack adaptability in the field of artificial intelligence products.Due to the abstractness and complexity of the "unreasonable standards" in the Product Quality Law and the urgency and uncertainty of the "technical standards" in the White Paper on Artificial Intelligence Standardization (2018 Edition),the use of existing standards is difficult to solve the problem of judicial identification of artificial intelligence product defects.Second,the existing legislation in China does not categorize product defects,but only makes general judgment on product defects.In the field of artificial intelligence,which is a subject of multi-subject responsibility and a variety of unknown risks,the judicial identification mode of product defects is more inadequate in application.In order to standardize the judicial practice of artificial intelligence product defects,defects should be divided into design defects,manufacturing defects,warning defects and tracking observation defects under the framework of product liability,which is not only conducive to applying the attribution principle of artificial intelligence products,but also corresponding to the elements of product tort liability.The judicial identification standard of artificial intelligence product defects should be combined with the obligations of relevant subjects.Specifically,the developers shall be responsible for the design defects of artificial intelligence products,and they shall fulfill the obligations of safety guarantee,performance balance and comprehensive testing.The judicial identification rule of "risk-utility" shall be adopted for the defects.The liability subject of manufacturing defect is the manufacturer,who should fulfill the obligation of cooperating with the expected design effect and manufacturing according to the design scheme.The judicial identification standard of "deviation from the expected design" shall be adopted for this defect.The subject of responsibility for warning defect is the manufacturer and seller,and the warning obligation should meet the requirements of the adequacy of warning content,the change of warning time and the conciseness and striking of warning language,etc.,and the judicial identification standard of "reasonable and sufficient" should be adopted for the defect.The subject of responsibility for tracking and observing defects is the developer,manufacturer and seller.Tracking and observing not only requires unimpeded product feedback channels,but also requires active and regular supervision of products,and more importantly,timely response to the problem products.The judicial identification principle of "case identification and comprehensive judgment" should be adopted for this defect.