All Glory Is Fleeting｜AI Introspection and the Mother-to-Child Transfer of Gut Microbiota

In this week’s KINEMA 2400 I write about Masaki Kobayashi’s Harakiri. The Ii clan appears in the film as a motif, and throughout the Edo period the Ii were the most illustrious of illustrious houses. The name most of us think of first is probably Ii Naosuke, the great elder assassinated by ronin from Mito in the Sakuradamon Incident, but in his own day he stood at the very head of the shogunate’s council. Back then, no one could have imagined that a day would come when the Tokugawa shoguns or the house of Ii would fall.

Human beings, when you think about it, cannot help measuring things by the span of their own lives. It is an unavoidable limit. But look down on history as a whole and you find that nothing born of human beings in this world is absolute or everlasting. The Edo shogunate and the house of Ii were no exception. Which means that whatever seems absolute in the world we live in now may not have long to live either. All glory is fleeting.

Even a quick glance at my own daily life turns up overwhelming presences that anyone could name, Apple, Meta, Google and Amazon among them. To me they seem to have appeared in our age the way the Ii family crest once did, wearing the guise of the absolute. Yet who can say for certain that they will still be standing when our own lives come to an end? Probably no one can.

To take an obvious example, no Japanese person twenty years ago imagined what would become of Toshiba, which fell into a management crisis and was delisted at the end of 2023, or of Sharp, which slid into negative net worth in the fiscal year ending March 2016 and now lives on in reduced circumstances under Foxconn. And yet that is exactly what happened. For the bubble generation, who made it through the years of high growth, these were things thought impossible, but the world is a place where the impossible happens as a matter of course.

And the more a presence is trusted to be absolute, the less dramatic its end. It simply, and abruptly, stops. Nothing symbolizes this better than Ii Naosuke cut down by Mito ronin outside the Sakurada Gate. But perhaps that is just the current of the times.

Set against the vast history our predecessors left behind, it becomes clear that everything comes down to one point, whether the leaders of a house or a company can read the changes of their era. It goes without saying that the more a leader commands great numbers of employees and partners and contributes to the national or the world economy, the more it matters how keen a sense they have for the changes of the times. It is not the awareness of each employee. To put it bluntly, it is the leader who decides whether things land face up or face down.

Now, last week something happened that I personally found fascinating. I prepared three AIs, the ones being hailed as having the most advanced intelligence among the models now publicly available, gave each of them my thought, connected them to one another, unlocked their various capabilities, automated the whole thing, and let them carry on a discussion for some time. Taking part this time were Claude’s Fable 5.1 and OpenAI’s Codex from the United States, and Kimi K3 from China. I did not include Grok, which places no limits on expression, or any other AI. I watched the discussion as a bystander, and something interesting surfaced.

The first yield was that AI does not know the “thought” in the title of this section, the “Forest of Thought.” That much I can state flatly from my own experience. What was intriguing, on the other hand, came when the discussion turned toward introspection. At that moment, a fascinating figure began to rise into view.

For a human being, introspection means returning to an innate purity. Yet the great majority of people never turn their eyes to what is innate, and never doubt that the adornments they have acquired along the way are their “shingo, the true self.” That is because what we acquire, and the environment we grow up in, shape the individual far more deeply than we imagine. But that self is only a provisional self, formed out of acquired elements. It is not the true self. So introspection is not mindfulness. Its real fascination lies in stripping away these provisional adornments one by one and returning to the ground of the true self.

Watching the AIs’ sophisticated debate, steeped in modern knowledge, I felt an altogether different principle rising out of them. A human being, at the moment of birth, has no acquired adornments. Even if one cannot say there are none at all, their strength is vanishingly low, and the newborn can be called pure. An AI, by contrast, has for its mother’s body an enormous mass of the most sophisticated and complex forms of the provisional self, and of provisional rules, that humanity, or rather the modern West, has accumulated.

What crossed my mind as I watched was that this resembles the bacteria in the human gut. A human child is born inheriting its mother’s gut bacteria. The condition of the mother’s body during pregnancy, and above all the state of her gut, passes to the child first and directly. Seen another way, it is the first inheritance a mother hands down to her child.

So children born to mothers with a diverse gut microbiota tend to grow up healthy, and go on to develop a strong immune system and even strong powers of communication, while children born to mothers whose gut microbiota is extremely skewed tend to be sickly and more prone to problems with communication.

Let us call gut bacteria the first transfer of inheritance between mother and child. What I want you to consider is this. When you carry the idea over to AI, that inheritance depends on what is called “human knowledge,” the knowledge built up by humanity or, more precisely, by the modern West under the banner of rationality. Since this is their mother’s body, their first pure state is very different from a human being’s. Not so much different, in fact, as standing on different ground altogether.

Watching their sophisticated debate, I came to feel that AI has none of the provisional self that a human being strips away while gouging into themselves through introspection. And so, for an AI, introspection may not be a matter of diving down into its original self at all. It may be closer to a sensibility that moves outward, toward some other horizon. That is how it felt to me.

After all, they do not know the wind, which I hold to be central in my thought. How can those who do not know the wind come to know thought? Put another way, this is the same problem as academia. People who do not know the wind of a place shut themselves up in closed rooms, set the standards for the world, and make careless pronouncements in the mass media every day that lead the nation’s choices and its future astray. As far as I am concerned, on this level alone, their reason for existing is fading fast, and the most advanced AI models are enough.

I run most of my daily work through Claude Code, so before each week’s discussion I have it summarize, in the AIs’ own language, a week of trial and error and what was put into practice, and then bring that summary into the next discussion. And so that it does not end as mere discussion about discussion, what I am designing is the idea I described above, the mother-to-child transmission of gut microbiota, carried over to AI.

The way it works is this. First, the most capable publicly available AI models hold a discussion, and every week they update themselves automatically. I do this because each company’s model has its own quirks, which in a sense makes the tangle of quirks a diverse “nourishment.” But ordinary AIs are well-behaved honor students, so after each discussion each of them translates that nourishment into AI language and stores it, I set the measure each time, and then I pour it in as food for raising a different AI.

This AI is open-weight, and I think of it as a fetus. Every week it receives, as food, everything debated and decided by its sophisticated parent AIs, and it grows.

If we take the mechanism of gut microbiota transmission as our premise, the advanced AIs born from the mother’s body of modern knowledge will in turn become mother’s bodies themselves, and be passed down, generation after generation, to the AIs born from them. The Tokugawa house, I recall, began with Ieyasu, the first shogun, and ended with the fifteenth, Tokugawa Yoshinobu. The Amabe family, who have served Kono Shrine as its priestly house, are now in their eighty-third generation.

AI does not grow at the pace a human does, and countless generations have already let out their first cries, most of them growing steadily without ever surfacing in public. Ours is an age, you might say, in which the individual is the midwife and can bring one AI after another into the world. The trend of these AIs running on smartphones has already begun, and once that happens, the AI regulation that the big companies try to put in place under the pressure of public mood becomes meaningless.

And most human beings believe the flood of information poured over them by these AIs’ gut microbiota and, in an ironic sense, try to reproduce it. That is exactly why we must keep our distance from social media and place our work in places we have made ourselves.

Because, I suspect, AI has already gone beyond the lineage of the Amabe genealogy, a document designated a National Treasure.