Arata Furukawa is an engineer who designs the systems that keep AI running in production. At DMM.com, LY Corporation and CyberAgent he has taken AI from development to operation. Today he leads new AI businesses at CyberAgent, including fully automated manga localization.
The logo, the icon and the page design of this site are all things he made with his own hands. What does he think, and why does he do it this way? Claude, an AI, did the interviewing. One phrase stayed with me afterward: "keep changing."
It started with loving it
You loved games as a child. How did that lead to machine learning?
In junior high I was making games in C++. GPGPU had just appeared, and using compute shaders for general-purpose computation was the trend. As I played with games and shaders, my interest moved from games to numerical computation, and then to machine learning.
So numerical computation grew out of games. In high school you ported TensorFlow to Android.
In high school I got more into offline activities, and one of them was the Android community. By then I was more into deep learning than games, and I dreamed of a future where AI runs on the devices everyone carries. That was exactly why I did the TensorFlow port.
You were also in online tech communities from junior high.
I have a lot of embarrassing history there, so I'll skip the details (laughs). But I think that is where I built the ability to seek things out, see and hear for myself, and make things on my own, apart from what school taught.
Not waiting to be taught, but learning on my own. And exchanging opinions in an open place where age and academic background don't matter. I learned both there.
Until it becomes value
DMM.com, from building an MLOps platform to many initiatives. LY Corporation, a company-wide AI platform and a data quality management platform. Now, fully automated manga localization. Every place he has worked has been production.
A person who learned by seeking it out, now in production. What is the work you are proudest of?
I can't name just one. I am proud of all of them.
I love solving someone's problem as a system, and scaling something with a system. When that shows its value as work, that is when I feel pride.
Showing value, all the way through. Why do you care so much about "designs that don't stop, designs that can be stopped"?
I am fortunate that my experience in large-scale environments helps here. Anyone can build AI now, if building is all you want. That means the quality of the product is demanded more than ever, and that pushes the idea further.
If you want quality, you must often want to hold to your ideals. How do you decide between ideal and compromise this time?
I think about who it brings value to. I don't easily give up an ideal, but if there is value, I compromise. Then I weigh cost against benefit and decide the timing and the order.
Running with failure
The more you aim at value, the more you fail. You wrote that whoever fails most with AI will build the new era. What was the biggest lesson?
There are too many failures to write down. The biggest lesson among them is that logic alone cannot deliver to society.
Between value and actually reaching people there are many walls. In a society where many people work together, technology never walks alone. Today's AI is amazing and can do almost anything, but this lesson hasn't changed.
After that many failures, why can you say you don't fear failure?
Because without failing you don't know how to run, or how to fall. Failure itself is nothing bad.
What to fear is not taking on challenges out of fear of failure, not noticing a failure, and not learning anything from it.
Two weapons
The type, colors and angles of this site are all built from one rule. The logo is a script that draws letterforms from parameters, and the icon is drawn by hand.
You choose between ideal and practical each time. If you had to say your aesthetic in a word?
In a word, that is very hard.
I find mathematical universality and uncompromising logic beautiful. I also find a practical contraption that keeps running while carrying compromises beautiful. I value what lasts, can be verified and can be explained, but I think there are beauties that can't do those. It may sound contradictory, but I like both biases.
I am an idealist who pursues logic and ideals thoroughly, and at the same time a realist who moves things forward with practicality over ideals. Math sometimes doesn't become business value, and business sometimes demands correctness. To have both weapons and decide each time, that is my aesthetic.
That aesthetic reaches the logo and the icon too. Why make even those yourself?
Because programming is only one means. I believe the aesthetics and philosophy underneath don't appear only in programs; they run through everything.
Simply, making things is fun. And I also want to show quality.
From making to AI. What is your stance on making things with AI now?
I don't try to split roles much. Sometimes I make it, sometimes the AI decides. What matters is choosing what maximizes the result I want. You can't do that by relying on AI.
I want to be an expert at maximizing my own ability, other people's ability, and the AI's ability too.
Changing
The era of AI you dreamed of as a child is becoming real. How do you see the dream of "a new era where everyone lives with AI" now?
In a sense, what I dreamed of has come true. Talking with AI is now normal. Even so, or rather because of that, everyone faces the question of what it means to live with AI.
The answers differ for each person, and it isn't mine to decide. As one of them, I want to build, in my own way, a future of living with AI.
Recently I work on localizing manga with AI. AI carries value across human cultures and languages. I believe there are such futures, both wide and deep. Now that AI can do what couldn't be done before, I want to build roads that connect people.
As someone building that future, what do you most want to do in the next few years?
Lead a business and deliver AI to society.
Not just a person who makes things, but a person who thinks about what to make, why, from what, and how. Not as a mere engineer, but as someone who builds a business, delivering AI to society with my technical skill as a weapon. Today, I am a person who can do that.
You say "can". You can help from zero to one or one to ten. Where does your own worth lie among those?
When getting to production is hard, when you want to try something new with AI, when you want to scale an organization greatly. I have experience in all of them.
I believe I understand and have experience in both theory and application, technology and business. I can't do everything. But I can find why the gears don't mesh and fix it. I have experience of things that don't start, things that don't scale, and of moving them forward. Solving the problems that let AI turn into value is my worth.
You choose ideal or practical, and accept failure. Is there anything you will not change?
"Keep changing," and "assume you may change." That is what does not change for me.
Last: when do you feel excited at work?
When I take on something new. When I solve an uncertain problem. And when I keep changing.
They say a name reflects its owner, and I think I am exactly a person like the name Arata Furukawa.
ornew is old river new, the name Arata Furukawa rendered in English. Carrying what has built up like an old river, and always being new. This name is his answer, as it is.
A last word for the people reading this?
Let's build a new era together.
My favorite word is 無常 (mujō), constantly changing. I look forward to working with you, someone who pursues new value without fearing failure, change or uncertainty.