Episode 4: What Made You Think to Use AI to Make Games More Fun? (Deep Dive with BNS Creators!)
Creators in BANDAI NAMCO Studios (BNS) work passionately to create games that everyone can enjoy. But what exactly is the source of their passion?
This series of articles will talk about how the creators at BNS apply things and experiences they like to game development.
Episode four of this series features a talk with Yohei Hase, a technical director in the AI Tech Unit’s Game AI Section, which is part of the Tech Studio’s Network System Department.
He develops AI technology for video games, and we hear he learned AI through self-study after joining the company. We got the chance to ask him about his thoughts on improving the games he works on and the kind of things he values in his work.
Yohei Hase, Technical Director
Affiliation: BNS Technology Studio, Network System Department, AI Tech Unit, Game AI Section
――― Could you tell us about your department and what you do for work?
I’m a Technical Director in the AI Tech Unit’s Game AI Section which is part of the Technology Studio. For work I research and develop AI used in games, and recently started implementing a system to find bugs by automating the player character’s movement with AI.
Recently we’ve been working on applying the system to BLUE PROTOCOL, and are in the midst of a cycle of implementation and revision.
I was a part of BLUE PROTOCOL until last year. I worked as the leader of the team in charge of implementing various forms of AI, including enemy characters and NPCs in the cities. In 2021 I founded a new department, the Game AI Section, and have worked there since.
――― You presented at CEDEC several times in the past. Speaking of which, you received the Award for Excellence in CEDEC AWARDS 2021’s Engineering Category—congratulations!
Thank you. The award was given for “Technological Foundations and Practical Applications of Decision-Making Technology for AI Characters”, which is the AI technology that automates NPC movements that I talked about earlier.
Recipient: Technological Foundations and Practical Applications of Decision-Making Technology for AI Characters
Comment： Despite a scarcity of concrete information on character AI planning technology in Japan and abroad, this presentation explained the system’s technological foundations and provided multiple examples of practical applications in games. It was also recognized for showing practical use cases and the low barrier to implementation in a game team.
CEDEC AWARDS 2021 Engineering Category: Award of Excellence
―――Until last year you oversaw technology to automate NPC movement, and you currently work on tech to automate tests. What sort of differences are there between the two?
There is a lot of common ground. However, the goal when working with in-game enemy characters or NPCs is to make the game more fun, so I try to implement the AI in a way that makes battles feel more responsive, or characters more lifelike. On the other hand, the goal of automating player movements for testing is to find bugs. In the case of the former, you add elements to the map that make the game more entertaining, but the information you add can turn into a breeding ground for bugs in the latter. We try to use AI as much as possible to address these bugs. They have different goals, which is why we use autonomous AI, which is similar to what’s used in robot vacuums. Robot vacuums clean your room from wall to wall without receiving any information about it. AI are similar, they automatically create their own maps to identify what type of dungeon it’s in or terrain features, then play through the dungeon. We make this kind of AI using the same technology as is used in robot vacuums.
―――That makes sense. Even if both types of AI control movement, the way they do so will differ depending on their goal. Normally when you think of debugging, you would imagine that a specialist would find bugs by playing the game. What are the benefits of automating this process?
One part of testing is repeating the same type of procedures. One example would be confirming whether changes to a program had an impact on other parts of the game. The largest benefit of using AI is automating simple procedures that don’t need to be done by humans to reduce the cost of labor.
―――So you need to separate tasks into ones that need to be done by people, and ones that can be done by AI. How would you describe tasks that well-suited for AI? I imagine that it would be proficient at checking whether things were running as intended, following a set of specifications.
Yes, I think that tasks that are well-suited for AI involve repetition, like clearing the same dungeon over and over again. We rely on AI heavily to find bugs, like the game crashing mid-dungeon. It’s boring and stressful for people to clear the same dungeon repeatedly. So something we often do is set the AI to automatically play through a dungeon, have them run it repeatedly all night, and it’ll have finished testing the following morning.
―――How often do you use this type of AI for work?
Currently (at the time of this interview) the AI is still mid-development, so we don’t often run many instances of it in parallel. We keep an eye on its performance and continually improve it bit by bit. However, we are considering using AI on a larger scale once they reach a certain level of completion.
The amount of content put into an individual game is on the rise, and so are the number of things that need testing. Automating parts of this work lets us focus manpower where it’s needed, which makes development easier overall.
―――You joined the company directly after graduation, but did you focus on AI when you were a student?
I did not study AI when I was a student. I studied graphics and AI as part of game development, but I didn’t study AI in depth at that point.
The first team I was assigned to in the company was Ace Combat where I worked on implementing the enemy’s AI. This was the first time I learned about AI in depth, and it was a very good experience for me. It was from then that I started studying AI on my own.
―――What do you find interesting about AI? Are there thing that make it fun to work with?
If you move a character from point A to point B without tinkering with it, they’ll bump into structures as they move, but implementing AI gives them the capability to move while avoiding structures. Also, I enjoy building up characters’ intelligence. The more you work on them, the more you feel them mature. This type of work is called character AI in the games industry. Since then, I’ve mostly worked on implementing AI that autonomously controls character behavior.
One major trend now is deep learning, and an example of where it proves effective is image processing, and that technology is available in commercial applications. On the other hand, game and character AI is highly compatible with robotics since they require characters to move autonomously. So I often research the technology used in robotics.
―――How specifically did you study AI?
Game AI is one of the domains of AI in which information in Japanese is most scarce. There is more information now than when I started studying ten years ago, and back then I tried my best to understand the English sources. I learned through information I found in documents online and in industry conferences, but it wasn’t exactly easy.
―――Do you have any personal rules that you stand by when working?
Automated QA in BLUE PROTOCOL
Let’s see. I try not to make compromises when I work. Development comes with time constraints, but I try not to settle with my current results, and to always look for even small things to improve in the time I’m given. And when I come up with things to improve, I try to see them to the end; I think nothing beats proactivity.
―――Do you have any recent examples of times you were proactive?
At one point, the staff in charge of scenarios and planning for each project at the company looked at examples by other studios and one of the things they started talking about was whether we should make a standardized scenario tool.
Everyone including those involved with planning agreed that we would benefit from having this kind of tool, but no one took action to create it. I realized that AI would be useful in making the tool, took the initiative to work on it, and am currently in the process of turning it into a project. I’m proud that I took a proactive stance then.
―――I think you often share information with the wider game industry, but what is your motivation to do so?
I personally often watch CEDEC presentations, so there’s a part of me that feels like I should give back by sharing my own knowledge, because I could say I’m where I am now by learning from external sources. Outputting my knowledge to people outside of the company is also a way to grow my network and gives me chances to receive new information, so I try to actively share what I can.
―――What do you want to try in the future?
Until now I worked on game projects, where I focused on improving the game with which I was involved. From this year I moved to the Technology Studio instead of to a specific project, so from now on I want to focus on using AI from a wider perspective to make all of the titles we create more fun.
―――To wrap up, could you tell us one of your guiding principles?
That would be a quote from Mitsuo Aida: “Learn all your life, stay young all your life.”
The first time I saw this quote, I felt it fit the way I think and act to a tee. I like that it embodies a desire to enjoy staying active and the humility to continue to study.
I’m currently (at the time of this interview) thirty-five years old, and as I get older and enter my fifties or sixties, I imagine there will be times when I will want to draw the line and say what I’ve done is enough. But I don’t want to give into that temptation, and hope to continue to study and try new things no matter how old I am.