How did game AI enter 2.0?

Original Source: Game Value Theory

Author: Li Yajie

Image source: Generated by Unbounded AI‌

Although it has been stated many times before that technology will become one of the most important thrusts for the phased development of the game industry, the rapid iteration of game AI applications is still staggering.

If the use of AI tools to improve production efficiency is stage 1.0 (volume), then Tencent NetEase, as the industry leader, has already stepped forward with more complex content diversification, whether it is the behavior setting of NPC or the application of intelligent actions , these are the qualitative 2.0 stages.

The application of technology is more directly linked to content competitiveness.

Academic and Applied

At the 2023 World Artificial Intelligence Conference (WAIC), which just ended this year, Xiamen University, in conjunction with Communication University of China, Central Academy of Fine Arts, Beijing Institute of Technology, Shanghai Jiaotong University and other university departments, is preparing to establish the “University Joint Research Center for Game Artificial Intelligence”. , continue to deepen the research on game artificial intelligence.

According to Zhang Junsong, deputy dean of the School of Information, Xiamen University, the combination of games and artificial intelligence technology has given birth to the frontier research field of game artificial intelligence. Game artificial intelligence is committed to creating human-like intelligent agents, helping to promote the innovation and development of general artificial intelligence, and has become an important part of the international technological competition. Emphasizing and promoting the key role of game artificial intelligence in the development of cutting-edge science and technology is an important goal and task for the academic and industry circles in the future.

Training AI in the game to help the innovation and development of general AI is the core keyword.

For example, the Enlightenment platform, which Tencent wants to open to colleges and universities from 2020, is a typical open desensitization game data set and game core cluster for students, and provides evaluation tools, computing platforms, and other research support services to test and improve AI in the game. Multi-dimensional decision-making levels in complex battle game scenarios. At the same time, it will cooperate with 19 top universities such as Peking University, Tsinghua University and University of Science and Technology of China to develop courses, so that students can use more interesting ways to understand knowledge points such as machine learning, reinforcement learning, and multi-agent decision-making to cultivate reserve talents.

This is the key significance of AI in academics, but from a business perspective, the outside world pays more attention to the practical application of game AI.

In the stage of game AI 1.0, using it to achieve a substantial increase in production efficiency is an important indicator to measure the development of enterprises in this area.

As an entrepreneurial industry, games always maintain a thirst for content. After the outbreak of mobile games, in order to improve product quality, the art arms race became the mainstream; domestic development solidified and version number rectification forced companies to go overseas for gold on a large scale; without normalization, cross-border exploration with game technology as the core became the key to raising the ceiling of the industry. new future.

As we all know, science and technology are the primary productive forces. The progress of science and technology has improved the quality of laborers, and their status and role in the production process have undergone profound changes; the penetration of science and technology into the labor data system has resulted in the emergence of many advanced production tools, which have fundamentally changed the mode of modern production. The appearance has greatly improved advanced productivity; the rapid development of science and technology has expanded the scope of labor objects and improved the quality, and the development and utilization have become more in-depth and effective.

Now, the more advanced production tools represented by AIGC are in front of all game companies. Through the proficient use of AI tools, it is possible to help small and medium-sized teams control costs and improve efficiency, while trying to produce more and larger projects to compete on the same stage.

In addition, creativity and ideas are of course the core competitiveness of the development team, but now the use of emerging technologies and advanced tools has also become a decisive force in the planning from the imagination to the actual product landing.

Taking YOOZOO as an example, it announced the official opening of the AI Innovation Institute a few days ago. The person in charge of the AI Innovation Institute said in an interview with the media: At present, YOOZOO’s planning and layout in the field of AI technology are divided into two aspects. Excellent technical talents, set up an AI engineering team, introduced mature commercial AI solutions, and also adopted some open-source AI technologies to cooperate with external technology companies in art asset generation, dubbing, and jointly set up a joint research institute for Specific application scenarios combining games and AI, customized solutions.

And at the “2023 Shanghai Game Elite Summit and Game Publishing Industry Report Conference” last month, Perfect World shared the comprehensive application of AI technology in all aspects of the internal game development pipeline, aiming at the needs of different types of AIGC tools and different projects, Targeted training of models for different AIGC tools. The content production team can directly output various more suitable pictures, audio and other materials to speed up the production process of UI, animation, material and other art assets. Specifically, AI generates material nodes with specified functions for the game engine, AI-assisted scene construction, AI plot assistance, virtual humans, etc.

The other is for the first time to demonstrate the complete form of the AI fusion drive that Perfect is trying - the composite application AI in GamePlay, including scene information, character information, plot development, player behavior, dialogue, etc. are all calculated by AI.

If production pipeline automation is the standard 1.0 model to reduce costs and increase efficiency, then intelligent NPCs can perform intelligently according to the current environment and situations, and provide different players with different senses and game experiences. In fact, they have already touched the core of game AI 2.0. Do it with peace of mind" to “how to do it better”.

Deep intervention in content quality

The theory of game value mentioned earlier that the current competition in the mobile game industry has entered the stage of mass control, and popular gameplay has become a consumable to some extent and a member of “content”. Most games need to provide content to maintain the popularity of the game. The difference is that the definition of content includes both plot-related and gameplay resources, not only the intuitive performance of game updates, but also the different directions of user UGC and off-site secondary creation.

Game AI 1.0 can try to solve the problem of distribution volume, while 2.0 means that major manufacturers hope to use their technical advantages to further intervene in the “quality” of content, thereby widening the competition gap.

At this WAIC sub-forum, Tencent Tianmei J3 shared some content of cooperation with the Tencent AI Lab team, such as designing a set of general integrated multi-modal environmental perception solutions, so that AI can efficiently model 3D terrain and make decisions, from The simplest 1V1 mode starts to research and train models, and gradually develops to a more complex 5V5 mode.

This is actually directly linked to the training of the academic nature of game AI mentioned above.

The 3D environment in FPS is closer to the performance of human beings in the real environment, and has the value of providing cross-field research. The results of training, in addition to academic value, can be more directly transformed into commercial value in the game.

Another intuitive application is the AI action generation technology applied by Tianmei in “Reverse War Mobile Games” based on the autoregressive neural network algorithm model (ARNN model). Simply put, the NPC in the game will react according to the player’s real-time behavior , these actions can be independently generated and adjusted according to the principles of kinematics, thereby improving the immersion and playability of the game.

This technology is also in cooperation with Tencent Robotics X Laboratory. It was originally used to train AI to let the robot generate the action performance of simulating real creatures. It also accelerates the progress of AI training while applying games.

In the mobile game “Naishuihan” recently launched by NetEase, the application of AI includes not only painting and face pinching systems, but also 400 AI NPCs and over 40 AI lyrics check-in points. Hu Zhipeng, senior vice president of NetEase Group, was previously interviewed by reporters Shi said that all the interaction process and results between players and smart NPCs in the mobile game “Naishuihan” will be deeply integrated and bound with the game content, which will affect the overall follow-up direction of the game.

In short, big manufacturers have trained more AI models and obtained research results, which can be directly converted into the application of more game content and create more direct value. This also means the impact of game AI on games All-round deepening from quantity to quality.

After clarifying the direction, the impact of technology on the game industry is still accelerating.