The AI department was established on January 1, 2022 at Square Enix Co., Ltd. (Square Enix), which provides many popular games such as “Final Fantasy” and “Dragon Quest”.
How will game AI evolve as digital industries such as the Metaverse and smart cities develop?
This time, we interviewed Mr. Yoichiro Miyake, who has been appointed as General Manager of Square Enix AI Department.
Table of Contents
- Into media beyond games | Background to the establishment of the AI Department
- An original game for each person | The future of games drawn by Square Enix
- From games to the real world | Possibilities of game AI
Into media beyond games | Background to the establishment of the AI Department
ーーPlease tell us about the background to the establishment of the AI department.
|Originally, there was a department called the AI Unit, but as the demand for AI increased within the company, the organization grew, and before I knew it, the members and areas of work had expanded beyond the scope of the unit. In recent years, the AI field has been expanding rapidly in areas such as machine learning, metaverse, and natural language dialogue in games. It was established to fulfill the role of delivering content.
ーーWhen it comes to the Metaverse, what do you pay attention to, Mr. Miyake?
|Today, people have high expectations for the game itself, and the digital game itself is changing. Until now, game AI has been associated with the behavior of enemies and allies, and the image of combat. However, from now on, metaverse-like community elements in games and various cultural elements outside of gameplay will become important.
For example, AI functions are required for loose interaction with users, such as conversations with store clerk NPCs (non-player characters) and holding in-game concerts.In particular, machine learning and deep learning are gradually entering digital games, and although they are not as fast as other industries, I believe they will definitely change digital games themselves. Until now, the parameters were created by designers and automatically adjusted by AI, but there were some details that we couldn’t keep up with. We are entering a phase of full-fledged efforts, such as adapting it through AI learning.
An original game for each person | The future of games drawn by Square Enix
ーーIn terms of machine learning in games, how are trends in game AI changing?
|Machine learning has traditionally been the technology of choice for game developers. For example, it has been introduced from development assistance, such as assisting artists and extracting candidates for regulation players in online games.However, recently there is a commentary culture, so just watching the video makes everyone feel like they have played. Therefore, by using AI, we believe that even if 80% of the game is the same, 20% can be changed to adaptive content customized for each person.
We believe that users have the conflicting desire to have the same experience as everyone else and to have a different experience. I don’t think it’s possible to satisfy both of them with unidirectional media such as movies, but I think interactive games have potential. Interactivity in a deep sense. It is a method of flexibly changing the content by estimating the personality, play style, and current psychology of each user.
Also, although it is still in the research stage, we are currently developing an AI character that draws pictures. The AI character has an emotion map, so you can see what state of mind he’s in. Emotions change depending on what the character sees, and when he draws a picture, a completely different picture is generated depending on his psychological state and the direction he is looking at. Until now, when generating a picture, it displays what the artist has prepared in advance, but this character uses deep learning to learn the touch of various painters. By doing so, we succeeded in generating a different picture each time, with the character really working deep learning in the game.
In this way, we are entering a stage where a painter-like character is created in the wide world of the game, and content is generated on the spot.
video playerーWhat you mean by “content generation” is not only graphics, but also movements, stories, and more importantly, dialogue, and everything else.
|I agree. However, although it has become possible with the power of AI recently, learning in the world means learning newspaper and media data for sentence generation, so it is necessary to customize it for games after learning. For that reason, it is necessary to perform *3 fine-tuning, and to make it learn to become familiar with the world view of a fantasy game.Deep learning technology enables us to do various things, but there are problems with machine learning, and there are things that we cannot do with data. Regarding this issue, reinforcement learning is more suitable for learning systems, and I believe that there are things that can be learned from various trials, so I am also proceeding with such research.
*3 Fine-tuning: Overall fine-tuning of some existing trained models and newly added models