Translated Interview with Naoki Yoshida on AI in Game Development

Couple weeks back, I came across an interview on エンジニア type of Yoshida having a discussion with an AI researcher from the University of Tokyo, Shota Imai, and sharing his (corporately limited) views on AI regarding its current usage, perception and ethics within Square Enix.

As a Final Fantasy XIV player, I respect Yoshida for his achievements in the game industry and his activeness in community interactions. I was interested on his thoughts on AI, while at the same time keeping abreast of the current uptake and usage of AI in the Japanese game development industry.

Shota Imai (left) and Naoki Yoshida (right)

Introduction

A quick intro to Naoki Yoshida. He’s the Producer and Director of the now renowned massively multiplayer online role playing game (MMORPG) Final Fantasy XIV at Square Enix. A little over a decade ago in 2011 or so, he was working on Dragon Quest projects, but had to take over the reins of a failing FFXIV that was causing Square Enix a financial disaster. His team somehow managed to continue provide bandaid patches to the original 1.0 to appease the surviving player base, while rebuilding a brand new version 2.0 in the background. The original 1.0 version of the game then ceased operation with an in-game calamity event that ties to the lore of the story. The servers were shutdown with this cutscene going up on Youtube right after. The rebooted 2.0 launched 10 months later in Aug 2013 and has been doing very well since, earning several awards, including the Game Awards Best Community award in 2021 and 2022.

The researcher doing the interview somehow happens to also be a decently experienced player of FFXIV, with 3000 hours over 3 and a half years, and has cleared several Ultimates (hardest content in the game when they were relevant).

The Interview

Let’s get right into it, what is extent of usage of AI at Square Enix (SE)?

Since the previous President, through academic-industry collaboration, SE has been moving forward to adopt using AI. Within the organisation, we have an ‘AI and Arts Alchemy’ team that specialises in the research and development of AI technology, so I’d like to think that we are a relatively early adopter of the technology. Within the FFXIV development team, adoption is similar.

I see! Somehow, I had the idea that FFXIV wasn’t really using AI much..

It’s understandable for it to seem that way. If you ask if FFXIV is being developed fully using AI, the answer is no. To be frank, we’re still far from having all team members, from engineers, producers, designers, to planners, to have a consolidated understanding of AI or whether adoption and usage of AI has permeated into their daily tasks.

Also there’s still an atmosphere of hostility towards AI in the entertainment industry.

Hostility towards AI?

Especially from last year, where there was a movement of Japanese voice actors and Hollywood actors coming together to campaign against having their voices being ingested by AI and then cloned. In the current situation, if anyone were to go around saying ‘We’re using AI’, they would quickly be mobbed on the internet. This is especially so because AI is such a vague term encompassing so many things and people still have different understandings about it. I think that this is something that can’t be helped at the moment.

I can imagine… The usage of AI in the entertainment industry is difficult to understand, and it tends to become controversial with concerns like “Won’t it take away jobs?”, or “Won’t it infringe copyright?”.

On one hand, if we talk about the definition of AI, programmers, and game developers who have been writing code for years would think, “Writing scripts, providing some input parameters and having it output something, isn’t that already a simple AI?”.

As a test, we’re also trying out having artists in our department, as part of their work, work on illustrations to be fed as learning data, to find out the quality of AI generated work possible.

Since these artists are paid by the company to do this work as part of their salary, the copyright of the data rightly belongs to the company. The model’s ability to produce works this way would then be copyrighted by the company. However, since it is create by the use of ‘Generative AI’, if word about this goes out, the risks would indeed be high.

‘Making decisions based on input parameters’ is the basic operation of a computer program, and in a broad sense, this also falls under AI. Right now, I think there’s a huge gap in understanding whether the ‘problem’ with AI lies with the input training data, how the model is used, or what and where the model is used for.

Because of the current situation where there is a large difference in the understanding of the technology and the ethical usage of it, the sentiment of ‘AI is evil!’ or ‘Ethical rules should be established!’ has been gathering attention and echoing throughout the population.

However, I believe that the research and verification on the usage of AI must be actively carried out, not only by the company, but also within the (development) team. There’s a global race to adopt AI, but how can we adopt and use it while maintaining safety is the question. Practical application of AI, particularly for creative work, is very troubling and the situation is currently in chaos.

Would it be possible to make use of AI in areas that would not be visible to players? Right now in FFXIV, there seems to be cases where players solve raid mechanics in ways that were not intended by the developers. Would AI be useful to find such unexpected bugs or behavior like that?

First of all, we think positively of the ways players solve raid mechanics that were not anticipated by the developers, as long as they are not due to bug abuse. It goes to show the amazing creativity of the players, and that’s what makes games fun. But, if we then talk about using AI to discover bugs, that’s where it gets tricky. (;´･ω･)

For example, if you use a skill that makes you invincible, but yet you still took damage. That could be the result of bug in the game, or it was intended design by the devs to not let you cheese it by having gaps in the invincibility.

In other words, even if there is a ‘bug’, or an event that that is unexpected based on the player’s past experience in the game, there are cases where it is deemed a bug and fixed, but also cases where it is deliberately kept as intended. This is because, from our creator’s point of view, we prioritise things that are ‘new’ or ‘not done before’ and that keeps things fresh.

Since there’s a possiblity that this scenario was intentionally created by the game designer, the final decision can only be made by the game designer. Of course, in this case, in the written specifications it does state that ‘This skill goes through invincibility.’ However, expecting AI to read through those specification documents and then making a call on whether it’s a bug in the implementation - that’s still not quite feasible right now.

That’s true, it seems difficult for AI to make judgements based on human common sense or to decide on the conventional way of clearing a game.

Regarding solving raid mechanics from before, we of course, create solution patterns in advance, and adjust accordingly to those patterns. However, I as mentioned earlier, the imaginative ability of players to trial and error different possible strategies are truly amazing.

The enjoyment of the game isn’t just all about playing the game with just that one ‘intended solution’ set by the developers, and that players are to execute exactly as follows. Instead, when players use their creativity to push the boundaries and overcome the challenge, they feel a sense of achievement and discovery, emotions and fulfilment.

to be continued another time