AI MIRAI's Vision for the Near Future of AI and Work [4]When will AI's "second dawn" arrive?
Yutaka Matsuo
Graduate School of Engineering, The University of Tokyo, Institute for Advanced Research
Kodama Takuya
Dentsu Group Inc. / dentsu Japan
What We Need to Face Change: Insights from Professor Matsuo of the University of Tokyo
The "AI boom" is shifting from mere hype to practical implementation. What phase will it enter in the coming years? And how should companies adapt accordingly?
Yutaka Matsuo, Associate Professor at the University of Tokyo and a leading AI researcher, and Takuya Kodama, leader of Dentsu Inc.'s AI project "AI MIRAI," discuss these questions.
The Moment AI Takes Hold: Lessons from the Internet's Evolution
The Moment AI Goes Mainstream
Kodama: While AI applications are gradually advancing in business, for many companies it remains in its infancy, often positioned as part of new ventures or R&D. Globally, however, AI adoption is accelerating in earnest, and falling behind poses significant risk. So, how will Japan's "second dawn" arrive, when AI adoption surges?
Matsuo: I believe it will be less about a single, clear breakthrough and more about various phenomena accumulating, leading to continuous progress. The internet 20 years ago, often used as an analogy for the spread of AI and deep learning (※1), didn't take off because of one specific trigger; it gradually permeated society. Deep learning is already being practically applied in medical imaging and facial recognition, and its use in manufacturing inspection tasks is also beginning. I expect this trend will continue, with continuous penetration.
Kodama: So it will spread continuously, though there will be differences by field.
Matsuo: Within that, I believe the major breakthrough moments will occur when it connects with business models. In the internet context, it was when search engines linked with search-linked advertising; monetization became easier, and it started to take off. But for deep learning, we haven't seen that yet.
Kodama: It's true we often hear that the only monetization methods people can think of for AI are "contract development" or "licensing fees." Especially since AI is mostly B2B. That's why it's often said it doesn't scale.
Matsuo: However, I believe general-purpose technologies connecting to consumers will emerge soon. Comparing it to the internet's spread, we're currently around 1998. It was a time when portal sites were deemed crucial yet didn't exist yet, and from there, various killer apps emerged online. I think AI will follow the same path.
※1: Deep Learning
A particularly advanced technology within machine learning that underpins AI, also known as deep learning. While machine learning identifies patterns and learns from vast datasets like images, deep learning has evolved this learning capability, enabling particularly difficult recognition and classification tasks. Google's "AlphaGo" also utilized deep learning.
Changing Advertising & Marketing: The Key is "Storytelling"
Matsuo: When AI becomes widespread, humans will essentially be able to live like royalty. Tasks around us—even highly personal ones like cooking—will be automated.
Kodama: Alongside this, I believe AI will bring significant changes to the advertising and marketing fields.
Matsuo: I agree. Using technologies like GANs (※2), we'll be able to automatically generate high-quality images, enabling ads tailored to specific situations and individuals. Essentially, we can present a "process of generating concepts within society" customized for each person. It's fascinating to imagine what might emerge then.
Kodama: For dramas, if image generation becomes seamless, we could integrate ads naturally into the story without making them standalone commercials. If the line between ads and content blurs, we might be able to convey messages in more appropriate contexts.
Matsuo: Or, for instance, using GANs to automatically generate props for dramas and change them to match sponsors. That way, props could be swapped during reruns.
Kodama: These kinds of changes seem certain to happen.
Matsuo: Advertising channels might change too. Take AI-powered home cleaning robots, for example. If these become widespread, they could automatically order consumables like tissues and toilet paper. Then, the choice of "which brand to buy" would depend on the robot's default settings. People who aren't particular about consumer goods brands would likely buy whatever the default setting offers.
Kodama: So the default settings of the cleaning robot would hold value as advertising. Advertising channels might transform completely.
Matsuo: As AI like Siri or Amazon Echo develops into concierge-like entities, they could become advertising channels themselves. However, users would feel uncomfortable if they discovered a store suggested by AI was actually an advertiser. Therefore, advertising on AI concierges should fundamentally be avoided. Yet, I believe it could happen. Specifically, in the form of "If this AI recommends it, I'll give it a try."
Kodama: That might be similar to the feeling of "I'll go because it's advertised in this magazine."
Matsuo: Exactly. Suppose the AI has a distinct personality and tells the user, "This place is my personal recommendation, so please give it a try." If a relationship has been built between the user and the AI, the user might think, "Fine, if you say so." This mirrors real human relationships, doesn't it?
Kodama: So the communication design itself becomes the advertisement.
Matsuo: And within that, the added value that becomes crucial is "storytelling." AI is often discussed in the context of "automation." But if you just automate architectural design, it only reduces design costs. However, if you can automatically generate designs in the style of a famous architect, the property's selling price actually increases. This technology is already within reach and can be monetized.
Kodama: That's what we call "storytelling," right? It's a concept currently highly valued in marketing, and if AI can provide it, it becomes monetizable.
Matsuo: Yes. The keys to deep learning's advancement are "automation" and "narrative." Kings spend money on things with narrative value—like paintings by famous artists or ingredients found only in specific regions—things with brand or rarity. Providing that in a different form is likely the future vision of deep learning.
※2: GAN
Generative Adversarial Network. A technology that creates an AI to generate images and another AI to judge the quality of those outputs, improving both machines' accuracy through an adversarial relationship. For example, the generating AI creates an image of a dog based on a database, while the judging AI analyzes whether that image meets the criteria for a dog. By feeding back this result, both become smarter. This process is repeated to increase the accuracy of image generation and judgment.
The "Understanding" and "Right Investments" Companies Need Going Forward
Kodama: We've discussed the future so far, but looking at the present, while last year saw a flurry of AI proof-of-concept experiments, things seem to have settled down now. With development costs still high and some saying that human labor is cheaper and faster for certain tasks, there's a risk of development slowing down. What kind of investments do you think companies should continue making?
Matsuo: Observing various initiatives, many cases labeled as AI are essentially just IT implementation or data digitization. They're merely catching up on overdue IT adoption at this juncture—akin to "turning a negative into a zero." What I've consistently emphasized is that the true innovation here is "deep learning." The discussion around IT and data conversion is entirely separate from deep learning, exemplified by AlphaGo. Deep learning has suddenly made possible what was previously impossible. It's like going from zero to positive. Shouldn't investment be guided by this clear understanding?
Kodama: Indeed, listening to the discussion, I see cases where these concepts are confused. Unless we clearly distinguish them, it might be difficult to envision future business models utilizing deep learning.
Matsuo: For example, medical images were difficult to analyze with previous technologies. Deep learning enables entirely new forms of analysis. The same applies to other fields: vast amounts of images and video footage, which were extremely difficult to convert into usable data before, will now hold significant value.
Kodama: Precisely why we should start identifying now what will become assets in the deep learning era. Image data collected casually could become valuable seeds for business.
Matsuo: Furthermore, we should proactively "go out and capture" images and videos with deep learning in mind. I believe this will be a pivotal point for future business. The crucial thing is for companies to clearly define their vision now: how will they generate profits using deep learning?
Kodama: That underscores just how significant the technological revolution deep learning will bring.
Matsuo: Yes. Above all, it's precisely because you envision the future and calculate potential profits that you can invest with a clear stance. Silicon Valley companies meticulously forecast this; they invest generously because they see the goal. Of course, this investment includes investing in talent.
Kodama: Talent development in the AI field is a highly focused area right now. Your lab also collaborates with Deep Core, and this talent development is also tied to corporate vision.
Matsuo: That's right. What kind of business can deep learning enable? How much profit will it generate? Having that vision makes it clear how much money can be invested in talent. Then, high personnel costs can be allocated without hesitation. As a result, people flow into that area. Overseas companies have already started doing this.
Kodama: So, clearly articulating a vision for using deep learning also helps secure talent, right?
Matsuo: Exactly. So first, you must correctly grasp what deep learning is as a technology and what it enables. That leads to the vision. Deeply understanding new technology should spark new business ideas.
Kodama: Moreover, we must act swiftly. As mentioned at the beginning, AI development progresses continuously, and deep learning could become the "standard" before we know it.
Matsuo: As we enter this era of change, falling behind globally would be fatal. Especially in Japan, larger organizations tend to lose agility. Internal constraints arise, and getting consensus takes time. We must avoid slowing ourselves down. Considering this, amid rapid technological evolution, rethinking organizational structures is also essential in this era of deep learning advancement, isn't it?
What is AI MIRAI?
A cross-functional project at Dentsu Inc. exploring AI's business applications across diverse fields and accumulating practical knowledge. It applies the insights, ideas, and networks unique to an advertising agency—gained from society and consumers—to the new field of AI. This professional group aims to continuously reinvent its own business and work methods through technology while delivering new value to society.
Was this article helpful?
Share this article
Newsletter registration is here
We select and publish important news every day
For inquiries about this article
Back Numbers
Author
Yutaka Matsuo
Graduate School of Engineering, The University of Tokyo, Institute for Advanced Research
Graduated from the Department of Electronic and Information Engineering, Faculty of Engineering, University of Tokyo in 1997. Completed the doctoral program at the same university in 2002. Doctor of Engineering. Joined the National Institute of Advanced Industrial Science and Technology (AIST) as a Researcher the same year. From October 2005, served as a visiting researcher at Stanford University. From 2007, Associate Professor at the Center for Knowledge Structuring, Institute of Advanced Research, Graduate School of Engineering, The University of Tokyo, specializing in Technology Management Strategy. From 2014, Co-Director and Specially Appointed Associate Professor of the Global Consumer Intelligence Endowed Chair, Technology Management Strategy, Graduate School of Engineering, The University of Tokyo. Specializes in artificial intelligence, web mining, and big data analysis. Received the Paper Award (2002), 20th Anniversary Commemorative Project Award (2006), Field Innovation Award (2011), and Merit Award (2013) from the Japanese Society for Artificial Intelligence. Served as Student Editorial Committee Member and Editorial Committee Member of the Japanese Society for Artificial Intelligence, becoming Deputy Editor-in-Chief in 2010 and Editor-in-Chief and Director in 2012. Served as Ethics Committee Chair since 2014. One of Japan's top artificial intelligence researchers.
Kodama Takuya
Dentsu Group Inc. / dentsu Japan
After working as a client-facing producer for digital platform companies, he has been promoting the use of AI both within and outside the company since 2018. He is currently affiliated with Dentsu Group Inc., where he is involved in the AI and technology strategy for the entire Dentsu Group, encompassing not only Japan but also overseas operations.
