Astronoka – 1998 Developer Interviews

Yukihito Morikawa - Director
Yousuke Saitou - Producer

―Morikawa-san, do you see Astronoka as an "AI game" in the same vein as Ganbare Morikawa-kun 2?

Morikawa: Yeah. That game uses a neural network which we modeled after the human brain, but this time we're using a genetic algorithm modeled after evolution. The AI in Astronoka is roughly divided into 3 types, and one of them is the "GA", or genetic algorithm, which uses DNA to encode and transmit small bits of information and in doing so "solve problems" in the environment. I thought it would be really fun to have a system like this in a video game. I had this idea for Ganbare, but there was no real place to use it there so I shelved it, but now we're giving it a go in Astronoka.

―And what are your thoughts on it so far?

Morikawa: Well, it was supposed to be an easier way to make a game. (laughs) What I mean is that, in RPGs, for example, fine-tuning all the monster stats is very tedious. You basically have to match the player's equipment and such to the monster's strength, maintaining a certain standard of difficulty all the way through. With the genetic algorithm I figured all that would be handled by the AI and it would be easier for me--but midway through I realized the error of that thinking. There's no easy way out. (laughs)

―What have been some of those challenges, specifically?

Morikawa: The fine-tuning of the traps, I'd say. There were bugs up to the end, and the evolutions I had expected to occur didn't happen. But I was unable to determine if this was an actual bug, or an issue with the underlying speed of the genetic algorithm. We had an incredible number of people working on bug reports, similar to what you'd have in an RPG. Eventually I got a hint of what the problem might be, and in turn I went and made all the traps stronger--and it was right then when we found the actual bug. I remember thinking, geez, remind me what was the point of using this AI again…? (laughs) You could say it was a lot more difficult than it needed to be.

Another challenge was the early groupings of Baboo. The very first twenty Baboo all start out the same, but fine-tuning the initial strength of that group proved to be extremely difficult. The very first stage of their evolution is the only one we can precisely verify. Once the trap battle starts, there's no way to figure out exactly what's going on under the hood. That's because the Baboo start evolving dynamically depending on the way you play and the placement of the traps.

―Do you have a favorite vegetable?

Morikawa: The shimaimo (sweet potato), because almost all of the woes of this development began with it. (laughs) I have a deep affection for them now. I continue to grow them even in the latter half of the game. It's meaningless then, but still… I'm too attached to get rid of them. That's why for me, this game begins and ends with shimaimo. (laughs)

―And your favorite Baboo?

Morikawa: It's the orthodox choice but I like Bub Baboo. I drew a lot of different rough drafts of both Bub Baboo and Shimaimo together. (laughs) I like their white coloring, I guess, despite being a little unassuming and plain. Also, I really like Bubtron.

―How has Astronoka been received at Enix so far?

Saitou: I think they were able to achieve most of what they set out in the original planning documents. One example and this is a very basic thing (and also something Morikawa had wanted to do for a long time) but instead of a "package tour" kind of experience where the developer holds your hand and guides you from one event to the next, Morikawa wanted to make more of a "free tour" experience that would reflect the subjective individuality of each player. I am also a fan of much older games, so I like that freedom. (laughs)

In that sense, one thing we focused a lot on was the on-boarding experience for players. For a lot of people the word "simulation game" is an immediate turn-off (laughs), so we wanted the game to be friendly. If you're just staring at stats and reading dry text descriptions then I think people will right away think this game is too hard. So we focused a lot on that presentation, and making sure that the writing was also interesting and fun. We asked the programmers to make the visuals and movements as fun as possible too. The different Baboo species all have subtly different reactions, for example. Honestly, we're going to be polishing and re-working Astronoka up to the very end… I'm already steeling myself for the crunch-time tears. (laughs) Our goal is to make Astronoka a game that's just as fun to watch as it is to play.

―What other challenges have you faced in the development so far?

Saitou: The most difficult thing would have to be the verification work. For this game, we're not merely working with simple numbers in a table; to adjust the balancing we have to actually change more complicated equations. So the verification work involves looking at Morikawa's initial specifications and planning docs, and confirming whether those things are actually being reflected and implemented in the equations. But the hitch is that the evolution speed is extremely fast, making it hard to track.

Also, the balance of the game shifts easily depending on what the individual player does. The vegetables and Baboo are handled quite differently, for example. The vegetables are, to a certain extent, easy to treat in a traditional gameplay sense, but adjusting the Baboos is like trying to grasp at clouds… at the same point in the development, we'll hear someone say "the Baboos aren't evolving" and then another person will say "the Baboos are too strong". (laughs) It's just impossible! But what's been incredible to me is the work of the Enix trap battle team during the debugging. They've managed to pull off all these different traps, and realize a system where you truly can evolve the Baboos how you intend.

―Do you have a favorite vegetable?

Saitou: It's a safe choice but I like the cosmic carrot. In the first year, in the cosmo carrot competition in Tengaron Village, when all the carrots are still the same shape and size, Yamada-san puts out a differently shaped cosmic carrot. If players feel the surprise and excitement from that moment I'll be very happy.

―Do you have a favorite Baboo?

Saitou: Babooks. I also like the standard Baboo with the elongated mouth. The white colored one. Their movements are cute to me somehow, the way their ears move and such. They don't move in tandem, right? The way just one ear will quickly go "bloop!" is real cute. (laughs)

---

Astronook - 2019 Developer Interview

excerpted from the full interview at denfaminicogamer.jp

Yukihito Morikawa - Director
Rie Narusawa - Producer (Enix)
Yoichiro Miyake - AI Developer (Enix)

―One year after Ganbare Morikawa-kun No. 2, you published Astronoka in 1998. This time, a "genetic algorithm" is used, but where did you get the idea for that?

Morikawa: It had a very specific inspiration. There is a artificial island in Tokyo built from a waste landfill called Yumenoshima, and at the time it was having a problem with flies. Each time the flies appeared they would spray them with insecticide, but eventually they became resistant to it. So the following year they'd have to use a stronger insecticide, and the whole pattern would repeat and escalate…

When I heard about this story, I thought, "this is how monsters evolve!" And instead of insecticide, I thought, "wouldn't it be interesting if there was a game in which the monsters became more and more resistant to the traps set by humans?" Then I thought, why not apply this idea to qualities other than poison resistance--like what if the monsters were smart enough to try evading the traps by, for instance, taking a detour in one case, or hopping over it in another… and that was when I realized a genetic algorithm was the perfect fit for this idea.

―Miyake, as an AI researcher, what aspects of Astronoka do you find compelling?

Miyake: Well, I don't want to sound like I'm being too critical, but Ganbare Morikawa-kun was a game that mostly appealed to AI enthusiasts: "whoa, there's a neural network in this game!" But for those who didn't know that it used AI, I think it was a little difficult to understand what made it fun. Astronoka, however, is a very interesting game even without the player knowing anything about AI. I don't think there has been a game made with genetic algorithms that surpasses it.

―What did you find interesting about it?

Miyake: The axis of interest in Astronoka revolves around growing the vegetables. The players take great care in growing their vegetables, so it's super frustrating when the Baboo pests eat them. (laughs) This motivates the player to put more sophisticated traps up, but then the Baboo evolve and become smarter… and that gameplay loop is a very fun one.

The way AI principles are embedded into the game design is truly innovative. It's so natural that the player doesn't even realize there's an AI there, nor does it matter for the gameplay. All games should aspire to use AI so naturally, and Astronoka is the game that taught me that.

Morikawa: I also feel that it achieved the most natural match between the gameplay and the AI functionality. Nowadays there are a lot of social games with farming elements, but there was nothing like that back then. I admit, when the idea first came to me I thought, "Maybe I really AM a genius…!", but then the lukewarm public reception to Astronoka sobered me up. "Ok, maybe not." (laughs)

Miyake: No way, I'm telling you, it really was a work of genius! (laughs)

―Narusawa, you first met Morikawa when he was making Astronoka, correct?

Narusawa: Yes, that's right. At the time, my boss, Yosuke Saito, was the producer of Astronoka, and when I first joined the company I nervously accompanied him to a meeting where he showed me the characters specifications for Astronoka. I was really shocked.

―What kind of specifications were they...!?

Narusawa: It was just one Excel sheet. Normally in other games, if you have 100 different characters you will have specifications and details written out for every character.

Astronoka has over 1000 different variations of Baboo, with different colors, shapes, and ways of avoiding traps, so I was expecting to see planning documents that were hundreds of pages long. Instead, all he had was a single Excel sheet with the basic 20 types. "The rest of it will be created by the genetic algorithm," he said. That was my first encounter with Morikawa and game AI.

―Aren't there infinite combinations of vegetables and Baboos?

Narusawa: Yes, and that's why Astronoka is a game that never ends. There's a tremendous amount of re-playability, and you can go on and on and on… it's a game that makes you realize what "game addiction" means. You can keep making vegetables and fighting Baboos all the time, or you can get really into growing one specific kind of vegetable.

In those days most games had a very clearly delineated ending. While Astronoka also has an ending, I believe Morikawa created it with the notion that "a game using an AI like this doesn't need to end."

―Which is precisely the kind of thing AI allows for.

Narusawa: I also think it allows you to play at your own pace. When you've played a lot of games, you inevitably run across some where "the world the game designer dreamed up" doesn't match your sensibilities. But in Astronoka the very frame and breadth of the gameplay is something you adjust yourself. That breadth of freedom is one of the biggest appeals of AI, I think.

Miyake: I think one of the fascinating things about Astronoka's gameplay is that it essentially consists of just one action: setting traps. After the player lays down a trap he can walk away and let the game do its thing.

Morikawa: Actually, during the development we had a lot of requests for more varied, interactive gameplay. For example, people were requesting a punching machine trap where you could press A and knock out the Baboos. But I ended up cutting all those out.

Narusawa: The appeal is almost similar to gambling, I think. For people who don't know what going on with the AI, Astronoka is almost a "prayer simulator." You set up traps praying that they won't get eaten, and if they are, it's like doh! But there's a thrill in repeating this process again and again and seeing what happens each time.

―But the thing about gambling is that if the player starts to feel its unfair, that the deck is stacked against him, it ruins the whole thing. In that sense, I imagine developing a game with AI at its core must have been very challenging...?

Morikawa: Yeah, there were moments in the development that left a big impression on me. Sometimes the intelligence of the Baboos would exceed the strength of the traps the player had set. Then no matter what kind of traps you set you wouldn't be able to stop their invasion.

When the testplayers discovered this, I thought, "Damn, this is a real big problem." But one day, one of the debuggers solved the issue by removing all the traps. In nature, when an animal has a trait such as wings or a tail that does not get used, over time that trait degenerates and then eventually disappears. We had actually included a "degeneration" function in the algorithm to account for this. And if you removed all the traps, the Baboo's abilities would drop at once. The effect of this was to make it impossible to overcome traps that had already been overcome.

Morikawa: In magazine interviews at the time I boasted about how "I meant to do that!", but in reality it was the discovery of a single, excellent debugger. The debuggers made many other discoveries and revisions, so it's no exaggeration to say that Astronoka was a game developed in large part by the debugging team.

Narusawa: At the time, the debuggers at Enix were quite special. Basically, we were a company that didn't produce games in-house and only had producers, but we were a really strange company, and for some reason the only other on-site development staff we had were debuggers. They were also famous for being extremely talented, and instead of having the developers disclose all the specifications to them, they would use their brains to work out the really deep aspects of the game.

They would even suggest things like "you should do this to make it more interesting", so it was kind of like every debugger was also a planner. (laughs) So I believe Morikawa is correct, that Astronoka would not have reached the quality it did it not for Enix's efforts there.

―By the way, I believe a genetic algorithm like the one in Astronoka requires several trial passes before it "learns" what to do... does this mean that there is another algorithm or "brain" working behind-the-scenes in Astronoka?

Morikawa: That's right. On the surface a new Baboo arrives everyday, but it's not the next generation of the one from the day before--in reality there are at least 10 generations changing behind the scenes, and the best one of them appears. That is the time when manual adjustments are made, like "if it hasn't evolved much, we'll release a slightly more advanced version."

Miyake: If they hadn't done it that way, the player would have to wait through 10 generations themselves. Instead, a function handles that in the background, so that the player feels like the evolution is taking place--and that kind of "cheating" is exactly what makes game AI special. What is most important for the player is that each round he feels there is some evolution taking place with the enemies.

―That's the same approach used by Ganbare Morikawa-kun. When a learning algorithm is presented as entertainment, there's no need to stick to academically honest methods: it's more important to think of good "cheats", if you will. Morikawa, in the course of developing these games, what was the appeal that you saw in using AI in game development?

Morikawa: To put it simply, the fact that it was not deterministic. I think normal game development follows an "outline process". Using various flags and variables, the planners painstakingly craft the game in line with the planning documents. In contrast, making a game with AI is much more like a free-for-all, no-rules cage fight. It's fun because there's this thing you've made, but you don't know if it's going to act the way you intended it to!

―Things can happen that even you can't foresee, in other words.

Morikawa: Sometimes there would be moments when I would suddenly feel the autonomous intelligence or emotions of the algorithm, and I get creeped out. It's almost something like a life or a soul, if you want to put it that way. Of course, strictly speaking it's not intelligence or emotion or anything--but I can't help feeling that way. The moment I felt it for the first time, it was a like a shock ran through my whole body. And I remember thinking, "Ah, I want to deliver this feeling to the players."

It was particularly interesting when we were making Astronoka. The Baboos would often find novel ways of evading the traps, things a human would never think of doing. I would watch them almost with a feeling of respect for the ideas they came up with. (laugh) In this way, I wanted players to experience those same things I experienced when I was making the game. When players have felt that for themselves, I think, they will have truly encountered the AI for the first time.

If you've enjoyed reading this interview and would like to be able to vote each month on what I translate, please consider supporting me on Patreon! I can't do it without your help!