Games are one of the important "training grounds" for artificial intelligence, and the spillover effects of game AI technology applied in various fields such as character action generation and level creation are increasingly evident. The "Exploring the Way of AI Innovation - Game Technology and AI Innovation Development Report" jointly released by the Research Center of Virtual Economy and Data Science of the Chinese Academy of Sciences and the Key Laboratory of Big data Mining and Knowledge Management of the Chinese Academy of Sciences shows that the total operating income of AI listed enterprises in each province increases by an average of 142 million yuan per year for every 1% increase in the scale of China's game industry. With the boost of games to the development of artificial intelligence and its industry, cutting-edge technologies in the gaming field are constantly emerging from virtual scenes, driving the progress of more artificial intelligence technologies in the real world. The intelligent agent action generation technology is a type of artificial intelligence technology in games. We hope to enable every NPC (non player character) in the game to achieve autonomous learning in action generation through technological evolution. Therefore, starting from 2019, we have been conducting in-depth research with Tencent Robotics X Laboratory based on the ARNN model (Auto Recursive Neural Network algorithm), initiating and gradually implementing the development of intelligent agent action generation technology, Yu Dong, the development leader of Tencent Tianmei J3 Studio's CoDM and reverse war mobile game, said. The intelligent agent action generation technology is not only very useful in the gaming field, but also has made excellent performance and progress in its cross industry application in the field of robotics. Last June, Tencent announced the "Game Driven Robot Accelerated Intelligent Learning" project, which introduced the application progress of this technology on the quadruped robot Max. On the one hand, this technology can help robots improve their thinking ability, enabling them to achieve more natural and smooth motion trajectories, intelligent generation, control, and decision-making of finer particle actions under continuous training; On the other hand, with the help of real-time physical simulation technology, research teams can quickly establish realistic, complex, and diverse robot virtual training scenarios in games. Coupled with the acceleration ability of game engines, it can significantly shorten the robot training process and time. NPCs in games need to have a high degree of similarity to humans in order to maximize game authenticity and enhance players' gaming experience. Yu Dong stated that in FPS games such as Anti War, the training scenarios and application demands faced by artificial intelligence are more difficult than 2D and 2.5D games. However, due to the high similarity between the three-dimensional environment of FPS games and the real world, the artificial intelligence trained in this environment can also better reflect human reactions in the real environment. In other words, artificial intelligence in games needs to be constantly honed in order to progress, and the advancement of artificial intelligence technology in games will also drive the development of other artificial intelligence fields. Based on this, the spillover effect of artificial intelligence technology in games will also become increasingly apparent, "Yu Dong said. In the future, game AI technologies such as intelligent agent action generation technology may have the potential to create greater value in industrial production, elderly care, and other fields. (Reporter Ye Qing)（