Dialogue with Yoichi Ochiai: Defining "Skills to Generate and Implement Ideas" for Using AI

AI is a hot topic. While it seems that only overseas AI developments get the spotlight, Fujitsu Limited, a major Japanese IT company, has created the Zinrai AI after more than 30 years of Research and Development(R&D). In this interview, Yoichi Ochiai, one of the top young Japanese computer researchers, and Seishi Okamoto, Deputy Head of Artificial Intelligence Laboratory at Fujitsu Laboratories, Ltd. discuss AI's present and future.

Introducing AI Must Be a "Careful, Thorough Process"

Ochiai: I came here to talk about advanced AI technology, but I see cards with hand drawn illustrations in front of me. They do not appear to be related to today's topic. What are they?

Okamoto: These cards illustrate work improvements and solutions that can be achieved by AI and other technology. These are just a few examples; we have approximately 700 cards in total.

I can spend hours looking at these solutions for a wide variety of industries and jobs. The technologies depicted in these cards are primarily AI and IoT, and the number of AI-related cards has been increasing.

I think people have begun to develop an image of what AI can do, but only a few can imagine specifically how they can use AI in their businesses. My hunch is that most companies are "interested in AI but have a hard time figuring out how to use it."

Therefore, we help clients organize their issues. We assist in identifying and visualizing issues and then select and suggest suitable solutions to such issues. Such a "careful, thorough process" is important in the B2B technology business, especially with new technology.

To help clients think clearly and organize the information in their heads, we not only ask them about their needs as part of our daily marketing activities but also hold regular workshops to cover topics in AI and IoT.

Regardless of title, people from different age groups, divisions, and jobs sit at the same table and discuss problems to find solutions while we help them expand their ideas by introducing the possibilities that technology offers.

At each workshop, we are stuck close together for about six hours (laughs). Within this time, we tell participants to share their honest opinions and discuss matters in depth. Participants usually evaluate these workshops quite highly. Every workshop uses these cards; some of the cards have been created from opinions offered during the workshops.

No matter how excellent a technology is, it has no meaning unless it reaches clients and they use it. Our activities may seem low-key, but we believe that one of our responsibilities is to showcase technology in an easy-to-understand way.

Ochiai: Your ideas are really like those of a system integrator. I think these cards will be very helpful when people consider using new technologies. Oh look, this was my idea, too--"convenience store eyeglasses" to help people choose onigiri (rice balls) at convenience stores (laughs).

Okamoto: This idea came from a workshop with convenience store owners. They wanted to address the issue of visually impaired customers who had to pick onigiri randomly because they could not see the label.

Ochiai: That issue was on my radar, too. In my case, targets also include people with hearing impairments and those in wheelchairs. Among other research, I am promoting activities primarily in CREST, one of the Strategic Basic Research Programs.

My lab has some overseas students. For them, too, the Japanese text on onigiri packages is a mystery (laughs). That led us to talk about eyeglasses which enable users to learn what is inside. These kinds of ideas never develop unless you are actually there and experience the problem yourself. They are really interesting because they already have a lot of detail.

Two Steps Ahead Leads to Failure; Imagine a Step Ahead to Succeed

Okamoto: Let me ask a question. We usually produce solutions like these cards by applying our technology, which we continue to improve daily in the lab, and clients' opinions. To generate new value, what way of thinking do you consciously practice?

Ochiai: There are many, but I'd say "see two steps ahead but do not act on it; do what is one step ahead."

Let's say that at a mine you want to make all the trucks driverless. That's an idea two steps ahead. Realistically, it is impossible to turn multiple trucks into driverless vehicles. Upon reaching this point, people stop thinking and give up on making their ideas into reality.

You have to see a step ahead. In this example, a driver can operate the first truck, and you can program the vehicles that follow this truck to be driverless. This reduces the hurdle and you might be able to make an attempt.

When we encounter new technology, we often assume it can do anything and imagine a dream tale. However, the important point is to limit the dream and imagine realistic steps forward.

In reality, it is difficult for companies to put this idea into practice. They cannot know what technology can do unless they ask experts. That is why society needs a company like Fujitsu.

I think that Fujitsu's business platform will be quite powerful in the future.

The package of human-AI integration can cover many more business fields than cloud services alone.

The Age of AI Puts the Spotlight on Hardware Again

Ochiai: Shifting to a more specific topic, many experts in AI research, myself included, are very frustrated with the poor computing resources for using AI--namely, hardware.

In my lab, when a team of students and I execute deep learning, we use a server that costs about 6 million yen to produce results suitable for a top conference. However, it takes around two months before we see research results... The more we want to do with AI, the greater the data volume, which means very slow processing. It almost feels like the computing environments from a previous generation.

No matter how good AI-enabled unique ideas are generated, this inhibits their deployment.

This problem may have arisen for us because our research is special and uses a lot of computing power. Still, I am concerned that when companies begin to think that they want to use AI more strongly, they may face problems because the hardware is too expensive or their budget is insufficient to realize practical use.

I fear that poor hardware resources may hold AI back from becoming part of business and daily life.

Fujitsu remains committed to supercomputers and other hardware even as other IT companies are withdrawing from the hardware business one after another.

Okamoto: That is because we think that AI's evolution needs both software and hardware.

In the example you just gave of processing speed, deep learning has become much faster thanks to graphics processing units (GPUs). Learning, however, often takes a very long time. In addition to processing speed, power consumption is also a serious problem. I do not believe that current GPU architectures are capable of both high performance and saving power at the same time.

We began to address this issue some time ago and are now developing a chip for high-performance, power-saving deep learning. Commercialization of the Deep Learning Unit (DLU), a processor designed exclusively for deep learning, is scheduled for next fiscal year. This should help eliminate the processing power problem that you are experiencing.

Fujitsu has made a great scientific achievement by developing the K computer with RIKEN (the K computer ranked first in the Graph500, an international supercomputer performance ranking, five times in a row). The DLU was developed with techniques acquired by building the K computer. I believe that skills to accumulate and effectively use such hardware techniques enable us to positively influence the AI business.

Ochiai: Though it depends on how much AI-related technology must be incorporated into solutions, IT suppliers that do not build hardware will have a hard time. Since the start of the 2000's, the main IT business battlefield has shifted from hardware to software. Cloud services are now booming, but there will be business opportunities for hardware again in the future.

Okamoto: I am very pleased to hear that. Zinrai was the result of our 30 years of AI research. We want to make the most of having our own hardware in addition to creating a full lineup of attractive APIs and providing our own environment for AI development and operation.

What Apple, Google, and Amazon Cannot Do

Ochiai: Fujitsu is unique because it is a system integrator that has its own hardware.

Effective use of new technology requires ideas about how to use it and integration, or the ability to customize and introduce the technology. The ability to maintain technology is also important, although this is a low-key activity. AI requires an individual to integrate it.

Going back to the example of convenience store eyeglasses, many companies could develop such a product as a wearable device. In order for the device to sense a user's entry into the store and guide the user to the onigiri shelf, however, more is required than a device with certain technical capabilities; the abilities to develop a system to cover the entire shopping process and support device operation are also necessary.

Companies may produce new ideas as technology spreads wider and the amount of learning materials increases. However, user companies lack the ability to combine hardware and software capabilities. They need partners.

In this sense, companies like Fujitsu that integrate systems and create hardware, software, and AI have an advantage. Another advantage of Fujitsu is that, through its long years of experience, the company has acquired many client contact points and become highly knowledgeable about clients' issues.

In the future, you will greatly benefit from your familiarity with a wide variety of clients (worksites) and businesses as well as from having an extensive business portfolio because the field of AI requires specialized knowledge and skills. Apple, Google, and Amazon will not want to do what you have been doing.

Offering products to clients that perfectly match proposals is very important and can also cover many business fields. It takes many man-hours, but your platform greatly contributes to reducing the number of such man-hours.

There is still one area where I would like Fujitsu to make more effort--to fully communicate the company's strengths and roles in society to its clients. In other words, I refer to practical branding, including directing the B2B brand image to C and improving brand recognition in society.

Japanese companies are not so good at this. They should more strongly appeal their brands because they are indispensable in delivering new technology to end users.

Okamoto: We are working hard in that area, too!

You just mentioned that we are indispensable in spreading AI, which is a role we take great pride in.

We place importance on our "social implementation capability," meaning our ability to solve clients' workplace issues and implement technology instead of just individual component technologies.

Besides component technologies for machine learning and deep learning, Fujitsu also has hardware and software to make the most of these technologies. In addition, our strengths also include the ability to draw out clients' needs and our system engineers' ability to integrate systems to meet such needs.

Fujitsu's Zinrai AI consists of two parts: processing capability and a technical framework. Its processing capability achieves "perception and recognition," "knowledge formation," and "judgment and support," while the technical framework makes use of learning and the latest technology to elevate what has been processed. We believe the most important part is "actuation" (action) to deliver generated value to people, companies, and society. Zinrai is a "Human Centric AI" that is always centered on people and worksites.

In daily operations, our system engineers and sales reps create solutions through in-depth discussions with clients whom we have built long-lasting relationships of trust with. New companies offering only technology cannot imitate us and our solutions no matter how hard they try.

We believe we can deliver AI solutions that solve clients' and society's issues because we are also a system integrator that has provided services and systems for all types of industries.

AI is something we develop with our clients. Introducing it does not mean our service has ended. From this perspective, Fujitsu's comprehensive capabilities effectively contribute to clients' success.

(Produced by NewsPicks Brand Design Team / Interview by Tsuyoshi Kimura. Written by Toadyish Sugiyama. Photos by Hirokazu Hasegawa.)

[After the interview]

Even after the post-interview photo-shoot session, Mr. Ochiai lingered in front of the inspiration cards. He began touching various cards and said, "I would make this into a business. Why hasn't anyone done that?" He seemed to have a great time generating one idea after another.

He then asked, "Who makes these cards?" We answered, "Fujitsu Design." He was impressed.
When we asked, "Would you be interested in supervising the addition of new cards?" he responded immediately that "Many could be created in no time if my 50 students create 20 ideas each and I check them." Perhaps Mr. Ochiai will take charge of adding new cards in the near future.

Mr. Ochiai then closely examined each Client Worksheet posted on the wall describing the outcome of an actual AI workshop and remarked: "Company owners who understand the concept of mid-term and long-term visions must participate in these workshops."

He concluded by saying that this was a good learning experience.
Seeing him scan and continuously take in all the latest information regardless of the situation was amazing. Those who are on the cutting edge are truly different from the rest of us!
(FUJITSU JOURNAL Editorial Desk)

Yoichi Ochiai
Advisor to the President of the University of Tsukuba, Assistant Professor of the Faculty of SLIS, Head of the Digital Nature Group, and media artist
Yoichi Ochiai was born in 1987. He completed his studies at the Graduate School of Interdisciplinary Information Studies at the University of Tokyo faster than others and received his PhD (in Interdisciplinary Information Studies). He specializes in CG,HCI, VR, visual-audio-tactile information display methods, digital fabrication, automatic driving, and human body control. In 2015, he joined the University of Tsukuba's School of Library Information and Media Studies as an Assistant Professor and became head of the Digital Nature Laboratory. In 2017, he became Advisor to the President of the University of Tsukuba, Visiting Professor of Osaka University of Art, and Visiting Professor of Digital Hollywood University. He is also CEO of Pixie Dust Technologies Inc. He has been recognized a number of times both domestically and internationally, including the World Technology Award 2015 from WTN (USA) in 2015, the Prix Ars Electronica from Ars Electronica in 2016, and the STARTS Prize from EU. His major publications include "The Century of Enchantment" and "Survival Strategy for the Super AI Era."

Seishi Okamoto
Deputy Head of Artificial Intelligence Laboratory, Fujitsu Laboratories, Ltd.
Seishi Okamoto joined Fujitsu Laboratories in 1991. Since then he has been engaging in research and development of artificial intelligence, information search, and knowledge discovery. For three years from FY2012, he developed new business through big data analysis and trained data scientists at Fujitsu Limited. Since FY2015, he has been the director of the artificial intelligence project at Fujitsu Laboratories. He also worked as a graduate school Visiting Professor at the University of Tokyo (5 years from FY2008), Visiting Associate Professor at the Japan Advanced Institute of Science and Technology (6 years from FY2002), and Visiting Professor at the Japan Advanced Institute of Science and Technology (3 years from FY2008). He led a Grants-in-Aid for Scientific Research (A) project (3 years from FY2007). He has a PhD (Science).