"Xiao AI, Xiao AI, what's the weather like in Beijing tomorrow?" At present, AI dialogue system has been deeply integrated into all aspects of people's daily life, and the technical architecture and route are blooming. However, as a cutting-edge technology, AI dialogue system lacks standards, resulting in different evaluation systems and uneven levels in application. In response to this situation, Professor Huang minlie, deputy director of the intelligent technology and Systems Laboratory of Tsinghua University, jointly formulated the world's first "classification definition of AI dialogue system" (hereinafter referred to as "classification definition") and officially released it recently. The classification definition aims to better evaluate the ability level of AI dialogue system and promote the application of AI dialogue system in the fields of virtual personal assistant, smart home, smart car (car voice), emotional escort and mental health. AI dialog system, which originated from Turing test, is one of the most important research directions in the field of artificial intelligence. "If natural language processing is the 'pearl on the crown' of artificial intelligence, AI dialogue system is one of the most difficult and core tasks in natural language processing, and it is the 'brightest Pearl'." Huang minlie said. Huang minlie said that in recent years, with the continuous development of deep learning technology, AI dialogue system has developed from the first generation based on rules and the second generation with traditional machine learning as the core to the third generation with big data and big models as the prominent characteristics. The dialogue ability has undergone revolutionary changes, showing amazing dialogue ability on open topics, which is of great significance and value to further promote the development of artificial intelligence industry. "Considering the wide range of tasks, diverse evaluation dimensions and rich technical routes of AI dialogue system, the writing team only focused on the dialogue system completely dominated by machines when formulating the hierarchical definition, and the man-machine hybrid dialogue system was not considered." Huang minlie said. According to Huang minlie, in order to play its value in practical application, the definition of classification is formulated from the perspective of user perception, observation, measurement and measurement, without considering the specific technical implementation of the system, and without distinguishing assistant tasks, chat, knowledge dialogue, etc., which are expressed in "scenarios". Based on the above principles, the "hierarchical definition" divides the AI dialogue system into six levels l0-l5 from the perspectives of automatic dialogue ability, dialogue quality, single / multiple scenes, cross scene context dependence and natural switching ability, personification degree, active and continuous learning ability, multimodal perception and expression ability. The higher the level, the higher the level of AI dialogue system. "From the above standards, the current AI dialogue system has developed to the highest level between L2 and L3, such as the common AI intelligent assistants such as' Xiaoai classmate '." Huang minlie said. Compared with foreign countries, what is the level of AI dialogue system in China? In this regard, Wang Bin, chairman of Xiaomi Technical Committee and director of AI laboratory, said that from the perspective of current industrial applications, China's AI dialogue system is not significantly different from foreign AI dialogue systems, and is generally at the same level. "Now the AI dialogue system as a whole is on the way to L3 and L4. It is still a certain distance from the ideal, and it still needs one to two years or even more of continuous efforts." Huang minlie said that to move forward to L4 and L5, we need to crack many key technologies in memory, association, reasoning, self-learning ability and so on. To make high-performance speech synthesis, if applied to the metauniverse, we also need to make fine-grained expression of actions and expressions, which are extremely challenging. In the opinion of Liu Qun, chief scientist of speech semantics at Huawei Noah's Ark laboratory, the highest level application of AI dialogue system is complex emotional tasks. The release of hierarchical definition will promote AI dialogue system to reflect a higher level of intelligence in emotional tasks, so as to promote the future application of artificial intelligence in emotional companionship, mental health, virtual human, metauniverse and other aspects, and greatly reduce human and material costs, Promote cutting-edge technology into the daily life of the public. (Xinhua News Agency)