攻殻機動隊 M.M.A. - Messed Mesh Ambitions_

In his 1990 work “On the ‘End,'” literary critic Kojin Karatani writes about the parallels between Japan’s Meiji and Showa periods. For example, both experienced political uprisings, with Saigo Takamori’s Satsuma Rebellion in Meiji 10 (1877) and the February 26th Incident led by the Young Officers of the Imperial Japanese Army in Showa 11 (1936). Each era also saw the establishment of a constitutional system around their twenty-first or twenty-second year. And the Showa period lasted 64 years, roughly the same as the Meiji and Taisho periods put together, seeming to suggest a 60-year historical cycle.

When I was working in government, I believed grasping the trend of history was crucial to policy. In the early 2000s, when I was involved with initiatives to dispose of nonperforming loans (NPL), Karatani’s argument made a strong impression on me. Is the current economic slump so very different from that of the post war period some 60 years before? I believed that if we could overcome the adversity, a time of greater prosperity would be sure to follow. Not only that, but I noticed that a great number of people in Japanese business seemed to be just putting things off and making excuses. It was as though they had become stuck in the past, fettered by their experiences with bad debt. I began to see NPL disposal not as a simple economic problem, but as something that its settlement could liberate people, and I appealed to politicians about the importance of putting and end to the problem at the time with this reasoning.

Everyone knows what happened next. NPL disposal concluded by the middle of the 2000s. But Japanese society still seemed far from its old dynamic self. The 60-year cycle model of history was no longer applicable. From here on, I felt that global economic factors would have a far stronger influence on the country than ever before.

Around 2012, I was searching for signs that things might be changing. That’s when I met AI researcher Yutaka Matsuo. As I listened to Dr. Matsuo tell me about his expertise in deep learning technology, I could see that this was no ordinary technological breakthrough. Postmodernism—a popular school of thought through much of my career in government during the Heisei period—held that there were no longer any decisive changes. Yet here was something I intuitively felt would bring about definitive transformation. I wanted to fully explore the implications of this new technology. Before long, the new field of AI was a hot topic in the media, and Dr. Matsuo became a man of the day.

Before I continue, I would to like to give a brief account of AIs history of development. By 2010, AI research had already enjoyed two boom periods, followed by “winters” in which interest declined. In its first generation, AI was able to make inferences based on human-provided logic. Then second-generation AI was developed to emphasize the intake of expert knowledge. But both generations ultimately achieved little. The AI was only able to solve problems in extremely specific conditions. Despite the staggering cost of learning, it could only respond to things it had been taught. We had hit a wall.

In contrast, the third generation of AI we see today is being developed with a totally different approach. This generation is designed as a neural network that mimics the brain and given a layered structure—the “deep” in “deep learning”. And the AI is tasked to learn on its own using a vast amount of data.

This extremely unique approach to machine learning is almost the opposite of the previous ones where humans would feed rules and logic directly to the AI. For example, behind the rapid development of natural language processing are newly devised learning mechanisms known as “attention” and “transformers.” The way I explain the essence of these mechanisms is that they make AI to understand the “feeling” of the words. It’s different from giving the AI rules like grammar or syntax, which are understandable to humans. For instance, let’s think about the “feeling” of the word “Philosophy”. It might feel at home to be close with the words such as “Reasoning” or “Descartes,” but it wouldn’t be so comfortable with the word “Pasta.” Rather than use simple word order, the AI calculates the semantic distance between words. By digesting huge volumes of texts, the AI then learns the distances between the words, expressing them in the form of vector. The outcome resembles something like a multidimensional semantic space comprised of words.

These methods have led third-generation AI to remarkable success. Many people perhaps already know about generative AI tools like GPT-4, which can generate an answer for a specific question while understanding the context of the question, even if that specific question has not been directly learned from data beforehand. AI is now able to discover patterns in complex information and appropriately employ them in other contexts.

It’s not just the new learning mechanisms and commercial successes that have led me to see third-generation AI as potentially bringing about decisive change. I’ve examined a wide range of theories, and in doing so have come to find that this mechanism of self-learning is widely shared beyond the field of AI. And it’s this commonality that has led me to believe our world is about to change dramatically.

One such revelation came to me when I learned about Markov Blankets, which are a type of Bayesian Network originally developed by AI researcher Judea Pearl. Bayesian Networks describe probabilistic relationships between events, and are thus one of the crucial models supporting machine learning and deep learning by AI. When applied to words, think of it as a method for dividing and grouping words based on their distance to one another. What most attracted me to this concept was neuroscientist Karl J. Friston, who applied the model to biology more broadly.

Friston used Markov Blankets to describe the concept of “local actions” that occur between organisms and their environment. Using humans as an example, local actions refer to how we acutely react to the things that happen in our close surroundings, like in the home or workplace, but don’t pay much attention to what’s going on far out in the universe. Local actions are broken into two categories: prediction or inference about the surrounding environment, and actions performed on the surrounding environment.

So where does the name “blanket” come from? While it sounds like something having to do with sleep, “blanket” is actually used to indicate the barrier between the “outside”—the external environment that one is trying to predict—and the “inside”—the individual doing the predicting. The formation of cell membranes, which are the foundation of organisms, can also be explained this way. In attempting to make a prediction, the individual sees itself as separate from the environment, and a barrier between the two forms. Ultimately, the concept of the Markov Blanket suggests a world in which “knowing” and “being” are essentially the same phenomenon.

In addition, the fact that cells predict their external environment means that they recognize this environment not as a simple aggregation of separate sensory data, but as a pattern. If they weren’t able to sense a pattern, they wouldn’t be able to learn new things, since the environment would be sensed as a perpetually unknown state. Friston believes that organisms form “generative models” to predict their environment, and are constantly revising their models in response to data in order to minimize error. So, it’s easy to see the similarities between the basic mechanisms of organisms and those of third-generation AI.

And that’s not all. Friston’s research also focused on the process in which higher-tier systems are created from lower-tier systems, like how cells create organs, and organs create humans. He believes that these processes are structured like Russian Matryoshka nesting dolls—Markov Blankets stacked on Markov Blankets—that grow in their comprehensiveness. As multiple systems intertwine on a given level, a higher-tier system is formed to help recognize and adapt to the external environment. This phenomenon is known as “emergence.” The emergence mechanism increases the world’s complexity while simultaneously making it possible to respond to complex systems.

And so, the principals illustrated by Markov Blankets can explain the intelligence of humans, AI, and all other life forms, and I’ve come to believe that they are fully present in third-generation AI.

I’m convinced that third-generation AI, rooted in deep learning technology, is radically transforming the way we see the world. I was so inspired by this when I was working in government that I wrote a book with the economist Keiichiro Kobayashi and Yukata Matsuo called “Sotaikasuru Chisei” (Relativizing Intelligence), published in 2020. In it, I presented a philosophy based on Markov Blankets that I call “Strong Isomorphism,” exploring how AI is changing our perception of the world.

So, just what is isomorphism? As I mentioned when explaining Markov Blankets earlier, the basic mechanisms of life and AI are isomorphic. But it’s not just that. There are also questions of human intelligence, such as how humans are able to use words to recognize the world, and where those one-of-a-kind abilities came from. I should mention new research currently underway, which is using AI learning mechanisms to examine how humans acquire intelligence. Spearheading this is Alison Gopnik at UC Berkeley, who has been able to show that infants create a frame of patterns and abstracted perceptions of their environment before acquiring language. The same can be said for third-generation AI: it learns how to recognize patterns using locally observed sensory data before acquiring language or logic. In other words, human intelligence, organisms, and AI are all isomorphic.

Next, let’s look at the structure, which resembles a set of Matryoshka nesting dolls. Friston says that a highly complex system is formed as cells form into organs and organs form into a human body. Higher and higher layers emerge as the same mechanisms of the Markov Blanket repeat. The result is a hierarchically tiered structure. This hierarchically emerging form can be also be characterized as isomorphic.

And finally, “knowing” and “being” are also isomorphic. This point is relevant to the philosophical debate between nominalism and realism, and whether an object exists because humans perceive it, or whether the object exists on its own. The Markov Blanket starts with the boundary between inside and external. If something “exists,” then the organism should be distinguishable from its environment. Conversely, if it were perfectly synchronized with its external environment, it would be impossible to distinguish and would appear non-existent. In the case of cells, the cell membrane forms a barrier between its inside and outside. The cell membrane is able to sense and interpret data from the external environment and respond accordingly. As such, the cell’s interior is not exposed to “surprise shocks” from the exterior environment. Order is maintained. Or rather, entropy is kept at bay. In other words, the act of an organism “knowing” about its external environment and “being”—existing and being distinguishable from its environment—are essentially the same phenomenon. Therefore, “being” and “knowing” for an organism are isomorphic—they are the same.

What kind of world view does Strong Isomorphism offer us? First, it’s not dualism between mind and matter. Knowing and being are part of the same natural process. Second, it’s not elemental reductionism. In the world we live in, various patterns emerge, repeating hierarchically to create complex pattern and order. Order does not spontaneously emerge through the random motion of elements, like with the law of large numbers. Rather, with Strong Isomorphism, the idea is that order emerges from order. The third tenet of Strong Isomorphism is the computational view. Organisms predict changes in their environment to calculate and generate models, thus maintaining order. It is the same behavior that occurs when generative AI and humans engage in conversation. The fourth tenet is the relational view. Patterns are generated only when one thing interacts with another. Cells interact with their environment. I interact with you. And through this we recognize patterns amongst one another. And the fifth tenet is the realist view. Once pattern recognition is established, the question “does this really exist?” is relatively meaningless. The straightforward answer is: yes.

These kinds of ideas existed even before the advent of Markov Blankets. For example, we can look at commonalities between Eastern and Western thought as summarized in Toshihiko Izutsu’s work “Ishiki to Honshitsu” (Consciousness and Reality). Izutsu examines pre-modern Islamic philosophy, Confucianism, and neo-Confucianism, analyzing each while taking the position that reality can be tangibly verified—that cat’s are real, for example. In his analysis, he comes to an understanding of consciousness as having a multi-layered structure, not being merely contained within a single superficial layer. His idea suggests a hierarchal structure to reality. Likewise, while the Zen Buddhist concept of “karma” is often associated with the concept of “nothingness,” it should in fact be understood as a dynamic process composed of constituent parts of the self and the empirical world. Put simply, rather than “nothingness,” we should think about a dynamic and fluid “existence.” My philosophy of Strong Isomorphism shares much in common with a strain of Eastern philosophy that analyzes hierarchies and relationships to reach a theory of realism.

Strong Isomorphism is radically different from most modern thought, which assumes that only humans are capable of perceiving things in their totality. While it is true that humans are able to perceive even the deep edges of the universe, in essence, individual humans and mankind are simply adapting to their local environment and performing a set of computations. But even humans are unable to compute for all conditions. In this sense, we are essentially the same as other organisms, all of whom fundamentally occupy the same realm of existence.

The concept of isomorphism is one of relational connections. This view also happens to be reflected in concepts found in quantum theory. For example, in Carlo Rovelli’s “Helgoland: Making Sense of the Quantum Revolution,” rather than reduce reality to the individual, the theoretical physicist explains objects, humans, and all other life as existing through their relationships and appearance to one another. This idea also connects to the quantum theory’s famous double-slit experiment. Essentially, the approach is to explain the characteristics of an object through interactions, including human observation. This concept is highly compatible with Markov Blankets and the idea that organisms predict and recognize patterns about their environment. Strong Isomorphism is an attempt to interpret and connect these concepts.

Strong Isomorphism focuses not on the absence of things in the world, rather, it attempts to answer questions about why there is so much diversity and plurality, and why—despite the laws of entropy—complexity emerges and evolves across life, society, and information.

Earlier, I stated that understanding the flow of history is crucial when it comes to creating policy in government. It was during my time working in government that I encountered third-generation AI, which led to developing my philosophy of Strong Isomorphism. So, how did this development affect my work in government?

When I started writing “Sotaikasuru Chisei” (Relativizing Intelligence), I oversaw digital data policy at the Ministry of Economy, Trade and Industry. While the things I was thinking about when writing the book influenced policy considerations, and vice versa, perhaps the policy most closely associated with my Strong Isomorphism philosophy was the idea that governance will radically change. At the time, this was coined “Governance Innovation.”

Simply put, Governance Innovation stressed the need to change to a style of governance that finds patterns and evolves in order to adapt to the environment rather than applying fixed rules, similar to how third-generation AI functions in comparison to previous generations. Earlier, I also mentioned that Strong Isomorphism is not reductionist, but takes a relational view. This leads us to a view of governance in which the boundaries of enterprise and organization are not rigidly fixed, but are flexibly configured across relationships with other entities. Another way of putting it is that rather than trying to optimize planning through PDCA (Plan-Do-Check Act), governance shifts to an environmental observation model of OODA (Observe, Orient, Decide, Act). Similar to Friston’s principles for living organisms, the idea is to be aware of the local environment, then adapt and evolve. Thinking about things this way allows us to see governance as existing together within the same framework as humans and AI.

During this time, Japan hosted the G20 summit, and I oversaw the field of digital technologies. Interest in digital technologies among G20 nations is extremely high, but their interests can also conflict with one another. We needed to set an actionable agenda that all 20 countries—among them the United States, various European nations, China, India, and Russia—could agree upon. At the time, then prime minister Shinzo Abe had promoted the policy of DFFT (Data Free Flow with Trust). The configuration of DFFT left room for compromises between countries, and if it could be explained in a way that brought in the philosophy of Strong Isomorphism, we could draw a closer connection to the T in DFFT, Trust.

To ensure a stable and secure system for global data distribution, rules need to be created amongst international parties. Commitments are made, applied accordingly, and updated as necessary. This is all fine and well, of course. But, I think the concept of “trust” has a slightly different connotation. The word for trust in Japanese is “shinrai.” When we say a person is trustworthy in Japanese, we mean not simply that they keep their promises. Rather, for us, the essence of trust is grasping the general intent of the promise so that one is able to deal with a situation when it goes awry and live up to the other party’s expectations. The concept of trust thus goes beyond a rule-based framework.

The “Trust” in DFFT became intertwined with my Governance Innovation initiative. Rather than having a set of uniform regulations dictated by overarching general rules, the idea is for governance that incorporates emergence, which arises as a result of the totality of efforts taking place across individual parts.

The concept of Governance Innovation was well-received abroad, and in January of 2020, shortly before the start of the COVID-19 pandemic, I attended a conference in Paris held by the Organization for Economic Co-operation and Development (OECD). I have fond memories of the trip. However, I found that the term “Governance Innovation” was more ambiguous to those in the OECD than I had anticipated. To better convey its meaning, I borrowed from Lincoln’s Gettysburg Address to break down the term into three aspects: governance of innovation, by innovation, and for innovation. Respectively, this means “government oversight of AI innovations,” “using technologies like AI in government, and “government support for AI innovations.”

Another core concept we incorporated into our policies at that time was that of “architecture.” Specifically, we established an organization at the Information-Technology Promotion Agency (IPA) called the Digital Architecture Design Center, which designs systems for data sharing among enterprises and architecture for regulatory compliance.

The experience further impacted my understanding of governance. To be brief, I came to understand that in the digital and AI world, simply creating and amending rules and laws will not solve issues. For example, the internet is considered a decentralized system. But this isn’t because there was some rule that dictated that the internet be decentralized, and even if there were such a rule, it would not have achieved much. The internet is decentralized only because of efforts to design a decentralized architecture.

The privacy-conscious EU created the GDPR (General Data Protection Regulation) regulations to restrict the distribution of data across borders. Personally, I deeply respected the human-centric values of the policy. But when I explained Japan’s DFFT program to EU staff, I stressed the importance of designing suitable architecture. In other words, the appropriate architecture had to be provided in order to achieve the GDPR’s policy’s requirements and live up to its European values. My idea was that it would be better to have initiatives carried out multilaterally among nations, and in the public and private sector. Of course, my ideas weren’t particularly unique and I’m sure many Europeans shared the same sentiment, but the idea sparked interest and led to a major conference in the OECD.

This discussion about architecture also ties back in to the world view developed in my book “Soutaikasuru Chisei” (Relativizing Intelligence). When we envision the future that AI will bring about, one typically imagines a Big Brother-like surveillance state. The authorities build a giant computing machine to control people. Then they lose control and the machine goes on a rampage. This imagining of a powerful, centralized authority seems to have deep roots. Or perhaps some people recall the mainframe computers that appeared in classic science fiction stories. Our imagining of AI in actual society is deeply intertwined with how its depicted in fictional stories, and there’s no assurance that things won’t be dystopian.

At the same time, I am a pluralist, and I believe that AI is fundamentally pluralistic, so it can be used in many different ways. This conclusion is not a result of living in a democratic country. As I talked about earlier, third-generation AI differs from the previous generations, which attempted to come up with optimal solutions based on set logic, inferences, and knowledge taught by humans. Third-generation AI’s architecture is rooted in decentralized information processing modeled after the human brain, giving it the ability to correct and learn from mistakes. So, at its core, AI must be designed to be decentralized and multidimensional, just like humans do not individually direct their cells when moving their entire body. I believe that modern architecture is able to achieve this sense of commonality, openness, diversity, and evolution.

In my book, I argued that the mechanisms of third-generation AI currently under development are able to repeatedly perform local actions, recognize patterns, and then go through a process of emergence. This is an inherently pluralistic process. It also posits that one cannot calculate the optimal response from a position outside of their environment.

In his famous 1945 paper, “The Use of Knowledge in Society,” Austrian economist and philosopher Friedrich Hayek pointed out that there are some problems that cannot be solved, even when all knowledge is concentrated in a single location. He stressed the superiority of market mechanisms, emphasizing the importance of knowledge being decentralized and coordinated appropriately. It seems that we are now entering an era where—beyond this “market”—the potential of decentralized, pluralistic systems and governance is being questioned.

So what can we imagine for the future? When it comes to Governance Innovation, we have a real-life case we can examine: the India Stack, a digital platform being developed in India. This type of platform is also known by the more general term DPI, or Digital Public Infrastructure.

India Stack has a four-layer structure. The first is the foundational identity verification layer called “Aadhaar,” which utilizes biometric data such as one’s face, iris, and fingerprints to authenticate identity. While adoption of the system took nearly 10 years, its realization has made it possible to open bank accounts for a dramatically lower cost, thus increasing bank account adoption. The program has also helped with social inequality by making it possible to receive subsidies and other benefits at a rapid speed. For example, within two weeks of the program being approved, benefits had been provided to 500 million people. Aadhaar is also used to oversee work attendance of civil servants, eliminating the issue of truant government employees.

The second layer is payment systems. The wide adoption of bank accounts and the liberalization of access to the financial system has made it easy for new players to offer payment systems, and the country has seen rapid expansion and innovation in the field of financial services. The third layer is data management, which allows the individual user to manage data such as official certificates and other documents, eliminating the cost and time it takes to submit various applications. The system gives individuals control over their data while also offering convenience. The fourth layer provides software tools that integrate with various services. For example, in the field of education, modules for learning, teaching, and supervision are provided, and can be customized by students, teachers, and supervisors. Students can check the status of courses taken and graduation requirements. Educators can create curriculum using freely available information. The status confirmation tool was even used to oversee vaccinations for COVID-19, leading to a faster roll out.

When India Stack first launched, I didn’t know all these details. But during my time in government, I became deeply interested in the platform, so I went to India to talk to various ministries and agencies about it. In fact, I wasn’t given a very clear explanation when I first went in 2018, but six months later, when I visited again, an India Stack booth had been created in the ministry building and was being enthusiastically promoted. I found it kind of funny, but it actually makes sense. You see, the development of India Stack wasn’t led by a government bureaucrat. It was started by a non-profit think tank called iSPIRT, who I was also able to meet with.

When I spoke with the iSPIRT engineers, I was impressed by the strong sense of drive they had about their ecosystem. No matter how skilled Indian engineers are, without a capable platform of its own, the country will be overrun by the big platforms from overseas. And in fact, iSPIRT’s team includes world-class engineers who’ve worked for GAFA (Google, Apple, Facebook, Amazon), so I was more than convinced.

The India Stack approach is completely different from the way things are done in Japanese government and business. It’s these bold, fresh approaches that decides the success or failure of the digital transformation (DX). One such approach is structuring the government in layers, similar to the structure of Amazon’s AWS cloud service. Rather than have the government stacked in a vertical, pyramid structure with separate departments, ministries, national and local governments, it can be horizontally structured with a system or platform at the base, which is equipped with tools for various fields, supports cooperative thinking, and is setup to allow for interoperability. Having such a system requires government functions to be modular and horizontally oriented, allowing them to be freely combined with one another. The India Stack program is configured in just this way.

When configured in the conventional vertical way, education is segmented into elementary, secondary, and higher levels, along with divisions for subjects, textbooks, supplementary materials, and tests. In India, they are working on a radically different modular structure. It’s based on using the same modules for basic schooling, higher education, vocational school, and retraining programs. It ignores existing organizational models. In Japan, we don’t have many success stories for such cross-disciplinary platforms, even in the private sector. When companies talk about their plans to use AI, they often talk about trying out proof-of-concept models organized in a vertically segmented department structure. But it doesn’t matter how innovative the technology. If it’s bound by existing organizational structures, it will never live up to its potential.

Let’s now look at some successful DX examples not restricted by vertical organization. I’m involved with IGPI Group and Michinori Holdings, who are working on a reform initiatives for regional bus companies. Many regional bus companies are suffering financially due to population decline. They are also dealing with a major shortage of drivers and staff. But if bus routes are reduced, the local residents, including the elderly, will not be able to go shopping, go to the hospital, or live their lives.

The first step in the restructuring focused on operations in Yokosuka. Michinori Holdings operates several bus companies concentrated in the Tohoko region. Each company is located in regions with different population densities, snowfall, and other conditions. Conventionally, one might think that given these differences it would be easiest for the companies to be operated separately. But Michinori is taking a different approach. Using the same logic as the India Stack, Michinori broke down bus company operations to eight factors, including vehicle procurement, safety measures, and new campaigns to increase use. Best practices are adopted horizontally for each factor. As a result, operations have improved.

This foundational work was followed by the horizontal introduction of a digital technology called “dynamic routing,” an AI system that allows the bus to stop anywhere. In short, the routes are not fixed, but change according to the needs of the passengers, with routes shifting depending on peak times for school, medical appointments, and shopping.

The new technology transformed operations. Dynamic routing eliminates the need for separate bus routes between public buses, school buses, and buses provided for hospital patients. And if they happen to enter a residential neighborhood, they can even be used to collect and deliver packages, integrating passengers and cargo. In this way, AI is able to transform bus services into another service entirely. What comes next is a model known as “Mobility as a Service,” or MAAS.

For another example, we can look to initiatives at HITO Hospital in Ehime Prefecture. In 2018, the hospital adopted iPhones for communications amongst doctors and nurses. Before that they had used PHS telephones, PCs, and daily reports to communicate with one another and access electronic medical records, just like other hospitals. But the old way of doing things was slow, and it was hard for nurses to reach busy doctors as they attended to patients.

The switch to iPhones led to a dramatic transformation in communications. For example, the chat function provides a sense of security, since messages are available to read at any time. This has resulted in more proactive communication. Likewise, chats can be shared simultaneously with doctors and nurses on separate teams, horizontally integrating communications and making it easier to flexibly change teams. Moreover, since communications are saved automatically, there is no need to work overtime writing daily reports about everything. And employees now walk significantly less, since they no longer spend time trying to track down an available computer. Together, all of these factors have dramatically improved the work environment. Nurse turnover has fallen to zero. Most importantly, the iPhone altered communications between nurses and doctors in a way that has ultimately transformed the culture of the organization so it’s no longer top-down.

However, I have to add that true hospital innovation cannot be easily achieved by management and staff alone. Attempts to further expand this model would surely run into institutional barriers. Doctors and nurses have a strictly defined scope of practice, which limits the flexibility of their work. As with other fields, I believe that if we are to truly take advantage of digital technology, the required qualifications and scope of practice applied to doctors, nurses, and pharmacists will have to be made more flexible, and regulations will need to be reformed.

For the last 30 years, I’ve gained many insights through the books of sociologist Masayuki Osawa. One of Osawa’s most well-known concepts is that of “third party position,” which refers to a transcendent outside party who develops a criteria for values and standards for right and wrong. In his writings, Osawa uses this concept to explain the mechanisms of modernity and capitalism, and traces the histories of Christianity and monarchies, attempting to explain the process of their creation.

Simplifying the concept of “third party position,” I would explain it as the movement toward a hypothetical point in the future where all has been fulfilled—arriving at an ideal, universal point. In actual reality, we might set different intermediary goals like “becoming a developed country,” “catching up to the west,” or “becoming a unicorn company.” Pursuing these goals, and changing them accordingly, sets and stabilizes the direction of society and individuals. In short, it can be viewed as a pyramid-shaped societal model with an abstracted apex placed somewhere in the future.

Since the Meiji era, Japan has developed with this societal model. Particularly impressive results include the period depicted by novelist Ryotaro Shiba in his book “Clouds Above the Hill,” which is set in the Meiji era. We can also look at the period of rapid economic growth that occurred during the Showa period as another example. Japan’s model was one of catching up and overtaking the western world. To survive, it was necessary to advance through a top-down structure, viewing each challenge simply and concretely. Kojin Karatani draws parallels between the Meiji and Showa period, suggesting a model where the top-down structure is loosened once a goal is reached, then tightened for the process to start again.

Of course, the model explained in Osawa’s concept of “third party position” is not unique to Japan. It explains the nature of modernity and capitalism in general. So what happens when this model is taken too far? If we become overly attached to a future goal or ideal, we end up with a totalitarian, non-pluralistic society, since any movement deviating from the goal will not be tolerated. In addition, those successful in moving towards the goal will also accumulate a disproportionate share of the wealth. No doubt, this is the exact problem facing capitalism right now.

I believe third-generation AI will bring about an understanding of the world that could be a catalyst for change. My 2021 book “DX no Shikoho” (How to Think About DX) advocates changing the pyramid-shaped societal model, and also addresses the potential of third-generation AI. That’s the final topic I’d like to talk about.

Incidentally, Hegel is often brought up as the philosopher who came up with a reasoning for why the “third party position” mechanism can push societies toward totalitarianism. Reading works such as “Lectures on the Philosophy of History,” we can see that the world seems singularly drawn to an ideal of western modernity. However, in his book “Philosophy from World History: Contemporary Edition 1,” Osawa (who I talked about earlier) presents a different reading of Hegel’s ideas.

Osawa’s interpretation is tied to Hegel’s idea of “concrete universals.” Osawa uses concrete universals as a way of solving Russells paradox, a concept from mathematics that looks at contradictions in set theory. Essentially, Russell’s paradox looks at sets that contain themselves as elements within the set. So if we accept that set A contains a, b, and c, what happens if we try to place a, b, and A in a set? I’d suggest referencing the original text for details, but accepting this, we find ourselves unable to determine the elements in the set. According to Osawa, we can use concrete universals to understand a process in which set A transforms into a separate set B. This transformation occurs autonomously from the contents of set A. As such, “A” is the concrete universal, as it is both a set and a concrete element.

Now, what does Osawa’s theory tell us about shifting from a pyramid-shaped society. One hint can be found in a topic I covered earlier: the transformation of bus companies such as Northern Iwate Transportation and Fukushima Transportation under Michinori Holdings. Through its extensive efforts, the company has successfully dealt with the issue of population decline, ensuring continued bus operations. Ultimately, what resulted was a business that includes cargo transport and other functions that go beyond normal bus operations. Essentially, they are now more than just bus companies. This is just like the transformation from set A (a bus company) to set B (a bus company that is more than a bus company). Likewise, it is occurring autonomously, as the individual bus company attempts to protect its business. This is made possible through technology and organization: AI-based dynamic routing and an organizational structure that is layered rather than pyramid-shaped.

This discussion surrounding concrete universals also relates back to the Markov Blanket model and my philosophy of Strong Isomorphism. We can see how the idea that stable patterns develop in the world through the combination of local actions performed by individuals in their environment, and the idea that, in adapting to their environment, new patterns and hierarchies emerge, is ultimately the same as concrete universals. However, with Strong Isomorphism, there is no need to come up with a final destination as there is with the “third party position.” And it is not compatible with a cyclical world view where the same things repeat themselves. It’s possible to choose the action to adapt to the environment. And more and more complex, evolved patterns emerge as the layers increase. Isomorphism does not need a “third party position.” But it also differs from a cyclical world view.

In fact, in his book Social Systems, sociologist Niklas Luhmann also offers up a similar Hegelian interpretation. Elaborating on his theory, Luhmann points out that it differs from a world view in which there is a single center that maintains control and sense. He also suggests that it’s these differences that give meaning to the world, which is similar to the idea of individual systems adapting to their surrounding environment. Luhmann goes on to write that the world can be understood by including a third set of unrealized possibilities, possibilities that have not yet been selected, but could be in the future. He states that this is what Hegel was referring to with concepts such as “dialectic” and “sublation.”

Following Osawa and Luhmann’s interpretation of Hegel and the dialectic, I believe that it may be possible to develop Strong Isomorphism into a type of social philosophy. Humans, other lifeforms, and AI are connected through the repetition of local computations, and their systems evolve on their own by accumulating layers. While there is no static, optimal point we can calculate to evaluate whether we are making objective progress through periodic choices and changes, the system is not anarchic. Stable patterns and order emerges.

Of course, Strong Isomorphism is not fully developed. But if there is to be some form of “new capitalism,” as it is often called, there are insights to be found by diving into AI and digital technologies and looking back at the ideas of those who came before us. That’s how I see it, at least.