In April, the University of Tsukuba entered a partnership with the University of Washington (Seattle, WA), NVIDIA, and Amazon in the field of artificial intelligence (AI). This initiative will receive support totaling US$ 50 million (approx. 7.5 billion yen) over the next 10 years. A signing ceremony was held at the Department of Commerce in Washington, D.C.
AI is now a word used by the media every day. SAKURAI Tetsuya, Director of C-AIR and a professor at the University of Tsukuba's Institute of Systems and Information Engineering played a key role in this partnership.
As a young boy, he was fascinated by machines and loved to take things apart. SAKURAI then discovered computers in college, which led him to where he is today. In this interview, we asked him about the future of AI.

(Interviewer: TAKAI Takaaki, Deputy Director, Bureau of Public Relations)

SAKURAI: In Seattle, I visited the UW and the Amazon Headquarters to talk to some people there. We're now at a stage to have more concrete discussions on the basic agreement that had been laid out. After Japan's Obon holiday in August, people from the UW, NVIDIA, and Amazon are visiting us at the University of Tsukuba. We've agreed to have our final discussions then.

SAKURAI: The University of Washington is located in Seattle so it has strong ties with Microsoft. Seattle is also home to Allen Institute for AI, founded by Microsoft co-founder Paul Allen. I believe these circumstances contribute to the university's robust AI research. The UW is also collaborating with Boeing, another company based in Seattle. This means, they're able to apply AI in various fields such as manufacturing.

SAKURAI: This agreement will give our university momentum to make great progress in AI research, and I believe we'll be moving toward the next step.

SAKURAI: American investment in AI research, including private sectors, is extraordinary, making the country highly advanced in this field. Meanwhile, it is essential for Japan to catch up in the field of AI. I believe it would be more advantageous to work closely with the U.S. in order to make further advancements, rather than trying to do everything on its own.

SAKURAI: In the past, my research focused on finding this function using nonlinear approximation methods. Deep learning, a method in AI, is essentially a form of linear approximation. You have input, deep learning processing, and output. What happens inside must be decided. So my research was very closely related to what is now known as deep learning.

SAKURAI: Of course, the concept of AI had existed for a long time with various methods being explored.
The term "AI" was first proposed in the 1950s. At that time, people thought that AI could be developed rather quickly, but it turned out to be much harder than expected. Back then, a method was proposed for determining if AI had been successfully created. If you couldn't tell whether you were talking to a human or an AI on the other side of the wall, then the AI was considered successful. But in practice, it became clear that this was not so simple.
The concept of deep learning had also been around for quite some time. There have been various attempts over the years, but didn't work as well as hoped. In the past, there were several waves of deep learning boom up until today.

SAKURAI: I think the first wave occurred when the concept first emerged and people were saying, "Wouldn't it be amazing if this could be done?" But when it was actually attempted, it turned out to be harder than expected, so the excitement faded after about 5 or 10 years.
After that, there were several occasions where people had high hopes for a breakthrough, but again, they didn't quite go as planned.

SAKURAI: Yes, the idea of AI acting like humans or having conversations, as in science fiction, did exist.

SAKURAI: Yes. The availability of large amounts of data on the internet and the significantly faster computational power have played a role. Also, the performance of deep learning boosted around 2015, drawing a lot of attention. That was around 10 years ago, and it was thought that this boom would end quickly. However, that wasn't the case. New developments kept emerging and this surge has continued ever since.

SAKURAI: I can't say if the current approach can achieve everything, but with various processes using AI, we'll be able to do things we never imagined before. Instead of just being a trend, I see AI become more like a fundamental part of society's infrastructure.

SAKURAI: For example, hospital data includes patient privacy, so it cannot be shared outside the hospital. However, data from a single hospital is often not enough and AI accuracy would improve, if it could learn data from as many hospitals as possible. In such cases, we cannot simply gather the original data. Instead, it would be useful to have a technology that transforms the data so that it can be used for AI learning, but cannot be understood by humans or be reverted back to the original data, thus protecting privacy.
The kind of technology which lets us use data while keeping AI safe and reliable, would be helpful in a wide range of fields, including healthcare, finance, manufacturing, agriculture, education, and drug discovery.

SAKURAI: While it's not the same approach as ours, there are methods such as encrypting data, or sharing only the machine learning model while leaving the data as is, which Google has proposed. Another method is adding noise to the data to hide private information while allowing for some level of machine learning. Various researchers are working on such methods.

SAKURAI: Well, I'm really into mechanics. I love things like cars and motorcycles, anything mechanical. I enjoy tinkering with amateur radio, and synthesizers with lots of knobs and buttons are also fun for me.

SAKURAI: Just round that time, my parents bought me a synthesizer. I was still a young boy so I tinkered with it without really knowing what I was doing. I didn't understand how synthesizers produced sounds, but when I got to college and learned about the Fourier transform, I realized that that was the principle behind it.
At my university, I was in the engineering department majoring in applied physics. Information technology was offered only in graduate school at the time, not at the undergrad level.
When I was in junior high school, the school only had one computer that looked like a giant calculator, and I occupied it all by myself to create programs.
So computers were already my hobby. During college, I bought my own computer, taught myself through books, and programmed it on my own.

SAKURAI: Yes, I chose a lab that had plenty of access to mainframes and large computers. It was a math lab that focused on developing algorithms for computers, so I was doing mathematics for computing.

SAKURAI: Yes, looking back, I've always loved mechanics and machines. For example, motorcycles... they have an engine with a complex structure. I would ride on top of the engine and see how fast I could make a curve. It was pretty fast!

SAKURAI: When I was a boy, I tried to make an electromagnet. I thought that wrapping more wire would make it stronger, so I made a gigantic electromagnet. I knew that using one or two batteries wouldn't be enough, so I connected it to a 100-volt power line and it exploded! The room got scorched in several places. It was a pretty scary experience.
That reminds me of other incidences. Vacuum tube radios have a very high voltage. As a kid, I didn't really understand that and I used to get excited about the sparks. And of course, I took apart all the clocks in the house.

SAKURAI: Exactly. The key is delving into the principles and understand what's happening inside that space.

SAKURAI: I guess you could say that, now that I look back.

SAKURAI: In the world of AI, there have been many instances where things that were said to be impossible to achieve in 10 years, became a reality just 1 year later. For example, beating top players in the game of Go, or the recent advancements in large-scale language models. It's really unpredictable when something will become possible. So I believe there will be a number of cases, where things we now think are impossible, will suddenly become possible. In that light, I hope that in 10 years, the University of Tsukuba establishes itself as a hub for AI, producing research results that can make a global impact.
First, we hope to establish a base for exchanges, where researchers and students from the University of Tsukuba can visit places like the University of Washington, NVIDIA, and Amazon to work on joint research there. Likewise, we want to welcome people from those institutions to our university and collaborate on common themes and tackle challenges together. Essentially, we want to create an environment where there is ongoing collaboration in each location.