[No. 244] Sohei Arisaka

From mathematics to construction, and then to AI. Looking back, I have always been moving back and forth between the abstract and the concrete.

My high school was a preparatory school, but my life revolved around the brass band club. I remember being severely scolded for skipping classes but showing up for club activities. It wasn't that I was slacking off; my cold just happened to get better in the afternoon. With that attitude, I burned out after the final competition and graduated without even taking entrance exams, seeing no meaning in going to university.

During my gap year, I read a vast number of books regardless of genre. Eventually, I realized that my interests lay in the roots and foundations of things, such as philosophy and mathematics. I thought it might be fun to study those properly at a university. Once I started thinking that way, I began to focus on my entrance exam studies and successfully enrolled in the Faculty of Science and Technology at Keio University.

I studied mathematics at Keio for six years. My specialty was axiomatic set theory, a field that deals with the structure of infinite sets. Mathematics is the foundation of natural science, but this field is the foundation of mathematics itself—it was a discipline related to the "foundation" I had always wanted to know about.

The three years from my senior year of undergraduate studies through my master's were extremely intense. My academic advisor, Professor Takeshi Katsura, had just returned from abroad, and there was only one other doctoral student in the lab besides me. The seminars were almost always one-on-one with the professor—a situation with nowhere to hide, but a very luxurious time.

In the seminars, we would read specialized books and papers and reconstruct the entire content ourselves without looking at anything. Saying "it was written that way in the book" was not allowed; I had to be able to explain both definitions and proofs in my own words. If there was even the slightest ambiguity, I would be questioned, and in the beginning, there were times when we didn't even progress one page in two hours.

What I gained from this experience was not just knowledge of mathematics. Rather, I believe the sense of "truly understanding something" was the most valuable thing I obtained there. The cornerstone of my current intellectual production activities is undoubtedly in those days. Professor Katsura, thank you very much.

After completing my master's program, I joined Kajima Corporation. You might think, "Wait, from pure mathematics to a construction company?" However, in my mind, mathematics and construction were connected as things related to the "foundations of the world."

In mathematics, you strip away unnecessary elements from complex objects and consider only the essence. That world is very smooth, abstract, beautiful, and infinitely free. On the other hand, the real world is full of troublesome things that cannot be easily stripped away, and whenever you try to do something, you immediately get caught on something "rough."

While I found that frustrating, experiencing the Great East Japan Earthquake made me realize that we ultimately live in a physical world with such constraints. Connecting abstract thinking to that reality to build the foundations of the world—I thought that would surely be a worthwhile job. With that in mind, I jumped into the construction company.

For the first two years after joining, I worked as a systems engineer in the IT department. Later, the machine learning I had been studying personally connected with my work, and in my third year, I transferred to the Technical Research Institute. The work I tackled there was exactly how to turn "rough" reality into a form that could be handled by mathematics and computers. I would listen to real-world issues, organize complex situations, abstract the necessary parts, and put them into theories or systems. Then, I would bring the results back to the reality side to solve the problems. It was very interesting work, moving back and forth between theory and reality.

However, as I got used to the work and mastered trendy machine learning methods and convenient open-source programs, I began to feel a slight sense of discomfort. I wondered if I was just borrowing tools created by someone else without fully understanding them and choosing problems that seemed solvable with them. At this rate, I wouldn't be able to create new tools myself for the problems I truly wanted to solve. Using, understanding, and creating—the more I became involved in research and development, the more I became conscious of those differences.

I wanted to understand AI more. And I wanted to be able to create something new myself. With that in mind, I applied for the company's study abroad program and gained the opportunity to study at a graduate school overseas. I chose the Doctoral Programs in the Department of Mathematics at the National University of Singapore. My research field was machine learning, but an environment where I could conduct research while belonging to the mathematics department was a natural choice for me. The methods I devised there were born precisely because I knew the constraints of the situations where AI is actually used. Being able to translate real-world problems into mathematically elegant forms was a great achievement for me and gave me confidence.

Currently, I am engaged in AI research and development at Kajima Corporation's Technical Research Institute Singapore Office. After completing my doctoral program, I have returned to work where I don't just use AI, but understand its mechanisms and think about how to connect it to real-world construction problems. Currently, I am particularly interested in AI agents that connect smart buildings equipped with various sensors with the people who spend time there.

By the way, recently, I have been using AI not just as a research subject, but also as a partner for intellectual production, such as researching literature, organizing thoughts, writing code, and refining prose. However, no matter how smart AI becomes, my understanding is something only I can do; AI cannot understand on my behalf. I believe that being able to stop and ask, "Do I really understand?" will become even more important in the age of AI.

My career seems to have opened up bit by bit by facing my own interests and sense of discomfort. It's okay if you can't see your future clearly right now. University provides an environment where you can take time with things you don't understand and nurture your own questions. I hope all the readers will cherish the things they find truly interesting and the slight sense of discomfort they might feel. That might one day take you to an unexpected place.