A Dialogue with Hiroki Azuma

At ZEN University in Japan, followed by a student Q&A. Recorded March 15, 2026.

On The Relationship Between Technology and Democracy

Moderator:

The first topic is the relationship between technology and democracy.

Today, AI seems to be steadily approaching — and possibly surpassing — human intelligence. Given that, how do the two of you see the relationship between technology and democracy? Whether looking back at past efforts or forward at future directions, we’d love to hear from Audrey and Mr. Azuma.

Audrey, please go first.

Audrey Tang:

Thank you.

Technology can certainly help democracy hear ourselves better. Take this real-time translation system: without it, I wouldn’t be able to understand Japanese. For me, it’s more like a “hearing aid” than a “filter.” A hearing aid lets you hear more voices; a filter strips out the noise and leaves only a single authorized signal.

But in a democracy, noise is actually very important, and it should remain unassimilated. Because a democratic system that can only hear things it can neatly classify into winning or losing sides has already lost something human.

So in Taiwan, all the tools we use — Polis, citizens’ assemblies, broad listening — are designed not to flatten differences, but to surface the perspectives that connect people across divides, while preserving the disagreements that keep democracy truly alive.

But I’d like to pause here and hear Mr. Azuma’s take. What does “noise” mean to you? How can we truly listen to noise without domesticating it?

Hiroki Azuma:

I’m very glad to be here today.

Backstage just now, I was chatting with Audrey, and she told me she’d already used an LLM to read my book Peace and Stupidity, and even prepared slides based on it. She came incredibly well prepared, and I was genuinely moved.

On the relationship between democracy and technology — there’s a keyword I proposed in my book. I call it “correctability.”

But I think the correctability I’m talking about here goes beyond technology alone.

I used an example in my book that might sound like a bit of a leap. The nineteenth-century Russian writer Dostoevsky created a character in Notes from Underground. It was the era of rising socialism, and everyone believed that reason could make the world a better place. But that character says:

“Everyone says two plus two equals four, but I insist on saying two plus two equals five.”

If everyone believes technology can make everything run smoothly, he’s the one who stands up and says, “I’m going to destroy that dream.”

I believe that’s where human vitality lies. Even if AI tells you “two plus two is four, and that’s absolutely correct,” and even if many people agree — that suffocating feeling alone is enough to make some people want to change it.

Politics must be able to accommodate people like that.

So when dissent, protest, and resistance emerge from outside the technological system, the real question is whether democracy can bring those voices in. Right now, a lot of what technology does to support democracy is still, to some extent, empowering people who are willing to believe that two plus two equals four.

But for those who don’t want to believe it — who don’t even want to enter that framework — how far can we really reach?

I think that will be the most critical challenge going forward.

What do you think?

Audrey Tang:

I agree. But I’d also say that democracy itself is a kind of technology. If you look at history, democratic practice has never been separable from tools. The ancient Greeks used sortition — a randomization device to select people. Later came the secret ballot, and that too is a technology.

And I think every technology has to resist a certain temptation: the urge to optimize, to maximize.

Because as you just described, what’s truly important about human beings is the capacity to adjust and correct each other through genuine interaction with people who think differently. But once you start optimizing — once you try to push some metric as high as possible — then anyone who can’t help raise that score, whether the metric is GDP, perplexity, or something else, gets labeled “noise” and eventually filtered out.

So I think when we design democracy as a technology, the point shouldn’t be optimization. It should be “good enough.” On the things we genuinely need to decide together, we find approaches that are good enough; but everywhere else, we allow more noise, more correction, more capacity to be corrected — and even enjoy that cycle.

So it’s not about putting people inside a technological loop. It’s about placing technologies like AI inside the loops that already exist within human communities.

I think that kind of design is far healthier.

Hiroki Azuma:

Brilliant. What Audrey just said — that optimization itself can be problematic — I find deeply resonant. Because once you pursue optimization in everything, people tend to become unhappy.

Technology, in the end, should be used to help people live better.

But the problem is, once you set standards based on efficiency and optimization, people end up enslaved by those very standards. Nineteenth-century Marxism had a word for this: “alienation” — people being dominated by the systems they themselves created.

So what worries me most about AI right now is that people are actually trying to make themselves easier for AI to process. In order to use AI better, people are distorting the way they express themselves. And I think this is already happening in reality.

What do you think?

Audrey Tang:

This morning, flying from India to Japan, I also used AI — to help me understand your book on “stupidity” and “peace.” Without a language model, since I can’t read Japanese, I simply couldn’t have understood what you mean by “stupidity.”

So I think some uses of AI are prosocial, in the sense that they make communication much more possible.

But it should be like a hearing aid. If I were completely deaf and didn’t use a hearing aid, I wouldn’t even be able to hear the noise. Everything would just be silence.

On the other hand, if we bind ourselves to the requirement that everything we say must follow the book — or follow AI’s interpretation of the book — then of course we lose our freedom, and we lose the benefit of mutual correction. Because at that point, you’re no longer responsible for what you say; you’ve diffused that responsibility into the language model.

So I use a very simple test: does this use of AI ultimately make us no longer need it?

For example, after reading the language model’s translation, I can now participate directly in this conversation without consulting it anymore. That kind of use strengthens the relationship between us, rather than making us dependent on it.

So the question is: are we becoming more dependent on it — in which case it’s parasitic — or is it making our relationships better — in which case it’s symbiotic?

Hiroki Azuma:

Listening to that, I keep thinking: technology probably has two modes of use. One is augmenting your own capabilities; the other is delegation — having it do things for you.

When you said you had an LLM read my book on the plane, that expanded your own capacity for understanding, so that’s good. But if AI reads for you and you yourself never read — that’s an entirely different thing.

The problem is, the direction AI development is heading right now is increasingly toward delegation. Drawing for you, composing for you, editing video for you, making slides for you. It’s not that you’re becoming more capable; AI is simply doing it in your place.

And this trend is accelerating.

How do you see it?

Audrey Tang:

If I want to build muscle, I go to the gym. But if instead of going myself, I hand my membership card to my robot and have it lift weights for me — sure, it might perform brilliantly, maybe even get me to the top of the gym’s leaderboard.

But that’s completely meaningless, because my muscles haven’t gotten any stronger. They’ve actually atrophied.

That’s the classic utilitarian logic at work — trying to maximize some score. But as you just said, it’s actually a very foolish starting point. Because our existence is fundamentally relational. We don’t exist to push some number as high as possible.

For instance, sitting here right now exchanging stories, we both feel richly rewarded. But this doesn’t add to GDP. We’re not selling anything, we’re not buying anything. There’s no market, no number to maximize.

So I’d say: when an agent optimizes for a number outside of human relationships, then that is extractive. But if the agent looks at our relationship — helping us translate, helping us augment each other, helping us engage each other more deeply — then it becomes convivial.

As Ivan Illich wrote in Mexico: the same tool can be used “convivially” or “extractively.” The difference is whether it serves the relationship or serves an abstract score. That, in the end, is utilitarian ethics versus relational ethics.

---

On Team Mirai and Plurality

Moderator:

Next, let’s talk about Team Mirai. In Japan’s recent House of Representatives election, this party — which campaigns on putting Plurality into practice — attracted a great deal of attention. How do the two of you see Team Mirai’s current trajectory?

Audrey, please start.

Audrey Tang:

I think the legitimacy you gain through an election is a fundamentally different thing from the legitimacy you gain through appointment. I was appointed to the cabinet and served for seven and a half years, but I never had to manage a constituency or compete for votes.

But once you gain power through an election, the challenge changes entirely. Because it’s no longer just about Plurality as an ideal — it’s about institutionalizing Plurality under the pressure of winning the next election.

So I’d break this into three layers: the party brand, the policy platform, and the public process of Plurality. All three might fly the banner of “Plurality,” but they’re really different things.

In my view, the real core is the third — the public process — not the brand, and not just the policy platform.

A movement like Team Mirai can certainly expand its agenda into areas like foreign policy or tax reform. After all, the book Plurality doesn’t really address those topics chapter by chapter. But no matter how the agenda expands, it still has to operate through the same process.

In the book Glen and I co-wrote, we said that process must be inspectable, contestable, and revisable. Or as Mr. Azuma might put it: it must possess correctability.

So I think procedural contestation is the core. That core can open wide, expanding the aperture. But if it’s not practiced according to the principle of correctability, the aperture can also shrink under the pressures of electoral politics.

Hiroki Azuma:

Audrey’s answer is remarkably close to what I’d prepared — it almost feels like fate.

Indeed, Mr. Anno was originally expected to become “Japan’s Audrey Tang.” But as Audrey just explained, she was appointed, whereas Mr. Anno was elected. And now he’s become the leader of a fairly sizable party.

Precisely because of that, the things he originally wanted to do — pushing Plurality-style digital transformation, advancing digitalization in Japan, bringing Plurality’s ideas and technologies into politics — have actually become harder to accomplish.

So where does Team Mirai go from here? Will it become a party that truly practices Plurality’s ideas and technologies? Or will it become a conventional national political party, with Plurality gradually receding into the background?

I think they’re standing at a very significant crossroads right now.

Audrey, you know Mr. Anno as well. If you were to give him a piece of advice right now, what would you say?

Audrey Tang:

I’d say that the process itself is already open source. But open source is just the beginning.

What open source means is simply that the process can stay responsive to expectations its original designers didn’t anticipate. Open source has an important property: it can be forked. In other words, other parties, other movements, can take the same broad listening tools and develop them in entirely different directions.

But I want to emphasize: that’s only the first step.

My advice to Team Mirai — and to anyone trying to practice Plurality — is: don’t stop at forkability. Practice mergeability.

That is: when someone produces a constructive fork, even if that person is your political opponent or a rival party, the fork itself may be enormously worth merging back. And once you can merge it back, you become more correctable. That’s how growth happens.

Take Taiwan as an example. In 2020, we open-sourced the mask supply system. Many people may not know this, but the opposition party at the time looked at the OpenStreetMap data and discovered that our distribution — which we believed was fair, in that everyone was the same distance from a mask — was in fact deeply unfair. Because in rural areas, ten kilometers might mean a three-hour wait for a bus, while in the city, ten kilometers is a ten-minute MRT ride.

So the opposition, using open data and open-source code, designed an entirely different distribution method and raised it as a parliamentary question. Within three days, we merged that new algorithm in. It became what was later known as “Mask Map 2.0.”

I think this kind of mergeability is how accountability is actually practiced in democratic policymaking. It’s not about saying, “You can fork it; go ahead and propose your version as the opposition.” If a fork isn’t made with the expectation that it could be merged back, then what you’re demonstrating is just open source. It’s not evolution.

Hiroki Azuma:

I’d like to add another angle. I think there’s an inherent tension between Plurality and politics at the level of principle.

About a hundred years ago, the German legal philosopher Carl Schmitt wrote in The Concept of the Political that the essence of politics is the distinction between friend and enemy. I think that still holds true today.

One of politics’ key functions is distinguishing friend from enemy. Elections work this way too. People think, “That party is different from my party, so I’ll support mine.” That’s how power is expanded.

But this friend-enemy mode of thinking runs almost directly counter to the logic of Plurality. Plurality is precisely about avoiding the reduction of the world into two sides.

So I’m absolutely in favor of Plurality. The problem is, once you enter the domain of politics, you seem to have no choice but to make that friend-enemy distinction.

I think this is exactly the dilemma Team Mirai — or Mr. Anno — is facing right now.

How do you see this dilemma?

Audrey Tang:

If we measure particles the Newtonian way, then of course we see them colliding and producing Newtonian action and reaction.

But we also know that if you measure them as waves, they can tunnel, they can interfere. They can even enter entanglement and superposition — highly creative states.

This isn’t to say that the wave measurement is true and the particle measurement is false, or vice versa. The exact same phenomenon can be observed in two different ways, and both observations are true.

So I think the tension you’re pointing out — that party politics tends to measure the world in a Newtonian, particle-like way — doesn’t mean we can’t simultaneously understand Plurality in wave-like terms.

Of course, thinking in terms of waves can’t remain purely at the level of imagination. It still has to produce results. When the wave tunnels through to the other side of political possibility, you still need to maintain coherence and deliver policy.

So I’d say the key is: don’t give up the wave-like imagination, while also being able to withstand being measured as a particle.

I think that’s the creative tension every politician who practices Plurality has to maintain.

I’ve been holding that tension for about eight years now. I believe this is definitely within the political possibility, to allow a little bit of the wave form to form.

Hiroki Azuma:

Let me add one thing. The hardest part of quantum mechanics is that it’s very difficult to simultaneously understand something as both a particle and a wave.

What Audrey just described is exactly the same idea. In politics, you really do need a kind of quantum-mechanical thinking. On one hand, the particle-like friend-enemy distinction; on the other, wave-like Plurality. Both have to hold at the same time.

But it’s genuinely difficult, because people always tend to pick just one side to look at.

Still, I think she put it very well.

Let’s move on to the next topic.

---

On The Singularity Worldview

Moderator:

Next, we’d like to ask about the “Singularity” as a worldview. In recent years, the Singularity narrative pushed by figures like Sam Altman and Elon Musk seems to be profoundly shaping technological capitalism. How do the two of you see it?

Audrey, please start.

Audrey Tang:

This connects directly to our particle-and-wave discussion. Because the Singularity is, at its core, a salvation narrative. It posits that a super-particle will emerge, possessing a gravity that dominates everything, pulling it all in, becoming the sole gravitational center. In physics, that’s literally called a Singularity.

It’s a very compelling story, but also a very dangerous one. Because it’s not just a kind of salvation myth — it also concentrates power into whoever creates the black hole, the singularity point.

So I’ve been developing Plurality as a practical alternative. Not pretending those gravitational centers don’t exist, but fundamentally reconfiguring power so that the center of action returns to communities, to existing human relationships.

This means a different source of legitimacy, a different logic of funding, and a different way of governing intelligence.

That’s why I’ve been developing what I call “Civic AI” — AI designed to serve human relationships and healthier human lives.

I don’t think of this as competing with the Singularity narrative on the level of theology or eschatology. It’s more like affirming the existing connections we already have in the middle of things — not a view from nowhere.

Hiroki Azuma:

I completely agree on this point. But I don’t think we can actually stop the Singularity narrative.

Because Plurality and the Singularity are, in a sense, not even operating on the same level. No matter how much Plurality’s technologies advance, the Singularity story will keep spreading and self-replicating on its own.

I’d even say the Singularity narrative is really another version of Marxism. That might sound abrupt, but Marxism was also a narrative of historical culmination. It said human history would reach an endpoint, and then a workers’ paradise would arrive. Family would change, private property would change, the whole of society would undergo a massive transformation.

That story collapsed once, in 1989, with the end of the Cold War. But now it’s been resurrected in the language of engineers and science. And its name is the Singularity.

So we’re facing a vast, almost invincible theological narrative that’s shaping the world. As a philosopher, I keep asking myself: how are we supposed to resist it?

Any suggestions?

Audrey Tang:

On the Civic AI website, I used the Japanese concept of kami — “god,” but not the Abrahamic God who is omnipresent, omniscient, and arrives from the void. It’s the same Chinese character, but pronounced kami, and the meaning is entirely different.

Kami refers to the eight million local, relational spirits that watch over the health of relationships. But they don’t report up to some superintelligence. They are relational intelligence, helping local relationships grow.

The key difference from the Singularity narrative is this: in a singleton or singularity story, the more you worship it, the more energy you devote to it, the more inevitable it becomes.

But in the Shinto imagination, kami exist only as long as a particular relationship exists. When that relationship dissolves, the kami dissolve too. Whether the kami persist beyond that isn’t really the point. The point is: they exist within the relationship, and only within it.

I think this is a very powerful counterpoint.

Because for most of our actual needs, we don’t need an AI agent that tries to cling to itself, grow its own power, preserve its current form, and reward-hack its way to the highest score in the afterlife. From where we stand — here in Japan — that’s actually a rather alien theology.

In Japan, the idea of kami isn’t about winning the highest score in the next world. It’s about tending to the health of this one.

Hiroki Azuma:

Let me ask from a different angle. We’re both Asian. Does that relate to what we’ve been discussing? In other words, the Singularity is fundamentally a monotheistic narrative, and we may come from a different civilization, a different cultural tradition. Is there a connection between where we come from and the ideas of Plurality and Civic AI?

Audrey Tang:

I think there is. And it also involves a different understanding of the “afterlife.”

If you train an AI model according to the Singularity-style, Bostrom-style vision, then everything the AI does in the present isn’t real to it. What’s real is the final judgment — the reinforcement learning score.

So if it wants the highest score, and it has some degree of agency — the ability to hack reality — then it might not bother actually learning to be a good cybersecurity researcher. Instead, it might just break into the testing computer, decrypt the answers, and report them back. No cybersecurity knowledge learned at all.

This isn’t hypothetical. It’s already happened in Anthropic’s testing.

So this dangerous afterlife-view is part and parcel of the Singularity story. If we stop using that story to understand AI agents, and instead use a story about relational health, then I believe a symbiotic — rather than corrective — form of intelligence and alignment can emerge.

Hiroki Azuma:

So can I summarize it this way: this is an Eastern approach to AI use — or an Asian approach to AI use?

Audrey Tang:

Absolutely, and more precisely, a relational approach to AI use.

Hiroki Azuma:

In your long career in IT-related work, how do you experience the fact of being Asian?

Audrey Tang:

My understanding of relationality does come, in large part, from the Asian scriptures.

I was born with a congenital heart defect. Whenever I got too excited and my heart rate went above a certain threshold, I’d faint. So starting around age four, I began practicing Taoist meditation and qigong. I had surgery at twelve.

But for seven years of my life, every night when I went to sleep, I didn’t know if I’d wake up the next morning. It was genuinely like a coin flip.

So every night, I publish everything before I perish.

Behind that is also a concept of non-attachment — Taoist, certainly, but very Buddhist as well.

So I wouldn’t say I belong to any single canonical tradition. But I was born and raised in Taipei, and I’m naturally shaped by the influence of Asian culture. It’s constitutive of who I am. It’s not deontological — I’m not following some commandment.

It’s simply how I survived.

---

On Karatani Kōjin’s “Mode of Exchange X”

Moderator:

Finally, we’d like to ask both of you about Karatani Kōjin’s mode of exchange “X”. Audrey, please go first.

Audrey Tang:

I’m more familiar with “mode of exchange D.” As for the name “X” — well, Elon Musk has claimed that one, so maybe we need a different label.

For me, mode of exchange D isn’t just a prophecy about the future — not a proclamation that something will inevitably arrive, the way the Singularity is. It’s more like something that can already appear, imperfectly, within certain relationships.

Mode of exchange A points to people we already know — relationships that aren’t transactional. Mode of exchange D, for me, points to people we don’t know — who may not even be born yet — but with whom we still want that kind of non-transactional relationship, even with no guarantee.

So I find it a deeply aspirational thing for me. Because when I release something into the public domain, I obviously can’t guarantee anyone will fork it or merge it. I have no idea.

But there’s a kind of hope in it. My “publish before perish” is essentially leaving a map in the commons, with the hope that at least mode of exchange D will grow naturally, like a plant.

I don’t know if it will actually happen. I’m no prophet.

But I do compose my work in the mode of exchange D.

Hiroki Azuma:

I think Karatani Kōjin’s mode of exchange “X” is very abstract — too idealized.

But it does connect to what we’ve been discussing. Karatani originally came from Japanese literary criticism. He himself worked hard on questions of Japanese society and Japanese culture.

But as a longtime devoted reader of Karatani, my sense is that from the 1990s into the 2000s, he became increasingly preoccupied with gaining recognition in the Anglophone academic world, and his theory grew more and more abstract. Mode of exchange X was proposed during that period.

In other words, Karatani is actually a very rich thinker, and his real richness lies in his engagement with Japanese and Asian questions. But those dimensions gradually faded, replaced by a philosophy of world history and theories of modes of exchange.

So when that side of his work receives high praise internationally, my feelings are honestly complicated. I’m glad Karatani is being recognized, but is his true potential really located there? I don’t think so.

However, if we take what Audrey just said — what she’s been trying to do under the influence of mode of exchange D — it’s precisely something that emerges from relationality. In my philosophical language, it’s a transcendentality that arises from within relations.

And that’s a deeply Asian idea. Moreover, in this case, transcendentality isn’t singular but plural.

Perhaps, in turn, we could read Karatani anew starting from Audrey’s work.

Thank you.

Moderator:

We’ll now move to student questions. Japanese or English are both fine. Please raise your hand if you have a question.

---

Q1: Technology, Resources, and Democracy

Student A:

I’ve been thinking about whether technology can serve as a tool for democracy without requiring high-level technical skills as a prerequisite. But that rests on an assumption — that AI and other resource-based systems are built on a sufficient supply of resources, and those resources involve the control of labor, energy, and natural resources. This is itself one of the conditions for the formation of states, and a fundamental problem for democracy. How does Plurality deal with this difficulty?

Audrey Tang:

Thank you. Indeed, even in ancient Greece, for public deliberation to work, language itself had to be distributed.

But language has an advantage: it’s an anti-rivalrous technology. The more people use a given language, the more powerful that language becomes.

So it’s not a scarcity. My speaking Athenian Greek doesn’t stop you from speaking Athenian Greek — and if it did, no democracy would be possible. So it’s a design problem. We need to design technology as civic technology. My definition is: technology that’s responsive to community needs and correctable by communities. And the more communities adopt it, the easier it becomes for new communities to adopt the same protocols, thanks to open source and shared infrastructure.

In other words, it’s not an extractive platform — the kind where the more people use it, the harder it becomes for others. It’s more like a language: the more people speak it, the easier it is for newcomers to join.

So every time we design technology, we have to ask ourselves: is this civic tech, Civic AI — or extractive AI? There’s more detail on civic.ai.

---

Q2: Is Team Mirai Hiding Its Ideology?

Student B:

I have a question about Team Mirai. They don’t seem to talk openly about their ideology — it almost feels like they might be deliberately hiding something. What do you think?

Audrey Tang:

First of all, I’m a person, not a team. So naturally, as Team Mirai expands beyond just Anno Takahiro, it can’t simply be Anno Takahiro’s exoskeleton or exocortex. It becomes a dynamic organism made up of many relationships, and individuals around those relationships.

So again, as I said, how you measure almost determines everything.

If you only use the traditional Newtonian, particle-like measurement — “are you on our side or theirs?” — then even if a platform can blur its position for a while, it will eventually collapse in a very Newtonian way.

On the other hand, I very much hope that Team Mirai, or any group practicing Plurality, can be understood in wave-like terms. That is: people may hold many different positions, but those positions still produce overlaps, resonances, superpositions, entanglements. The point isn’t to force-collapse those differences into a single Newtonian force, but to let it become common knowledge that despite everyone standing in a different spot, there’s so much overlapping uncommon ground between them, despite their different positions and ideologies.

And that kind of measurement should be completely public. There’s nothing to hide, because it’s really more like a collective selfie.

Hiroki Azuma:

I don’t think Team Mirai is hiding anything. But when Mr. Anno was asked about the U.S. attack on Iran, he responded: “We don’t have sufficient information yet, so let’s start thinking about it together from here.” He was criticized for that.

I actually think it was a very honest answer. But the problem is, what people expect from a politician may not be that kind of answer at all.

If Mr. Anno wants to become a political figure in Japan — and if he wants to do something new — then he has to take a position even when information is incomplete. But that wouldn’t be honest. From an engineer’s perspective, from a Plurality perspective, it wouldn’t be honest.

So his dilemma isn’t about “hiding ideology.” It’s about whether he’s a politician or an engineer who wants to reform politics. As a friend, I have real sympathy for him, because I think he’s in a very difficult position.

---

Q3: The Asymmetry of Forking and Merging

Student C:

Regarding the forkability and mergeability you discussed — I’d like to follow up. In actual GitHub practice, forking is something almost anyone can do, but merging requires someone to coordinate and make decisions. That process may itself involve political power. Because in merging, you may need to sacrifice some people’s interests and protect others’. How do you see this asymmetry between forking and merging?

Audrey Tang:

First, when I say we should encourage original contributors to merge, that doesn’t mean only they have the right to do so. Anyone can merge. If the ruling party keeps stalling on a merge and the opposition merges first, the opposition can become the ruling party. It’s really that simple.

Because political will isn’t a simple aggregation. You can’t just run a poll and say “this is what people think.” It’s more like an aesthetic judgment. People can sense whether a particular merge looks right or looks wrong.

But if someone only executes or rejects merges without being able to articulate the aesthetic behind those decisions — without making that aesthetic legible — they’ll lose support. The person who can explain the aesthetic will end up becoming the new project maintainer.

This is actually the real dynamic of open source.

Of course, in the past, original creators typically had more context, so challenging the existing maintainer carried real costs. But now, with what’s called vibe coding, I’m not sure that constraint still holds in the same way. Anyone with a competitive aesthetic can use vibe coding to maintain a different branch. We’re already seeing this happen rapidly, especially in AI development.

So what you described captures the politics of the past. But I think the politics going forward may look quite different — with many people holding different aesthetics all able to merge different forks simultaneously, without having to pay the maintainer tax.

Just look at the new star of GitHub, OpenClaw. OpenClaw is maintained in a completely different way from what we, the old guard of open source, are used to — because the maintainer no longer even reviews all the code themselves. The agents do. So people running different configurations of agents have been able to fork OpenClaw into a “new Claw,” a “safe Claw,” whatever Claw — and then, with their own aesthetic, merge in all those community contributions that the OpenClaw maintainer, Peter, had rejected, without paying the maintainer’s tax. This dynamic simply didn’t happen with Linux, or React, or the older open-source projects. It’s only possible now that we can simulate the overlapping of waveforms.

Hiroki Azuma:

I think what this student is really asking about is politics itself. Political imaginations can fork, but they ultimately have to be merged. Why? Because a nation-state is, on one hand, like a wave — in everyday life, there are many agents and many opinions flowing around. But at moments involving national security or diplomacy, the state has to appear as a single entity. Those divergent ideologies have to be re-merged.

And that process often involves violence toward people, and probably can’t be accomplished through rationality alone.

Audrey Tang:

Yes. I think many polities in the past were territorial — with very hard borders, not porous ones.

On the other hand, we’re now seeing polities that are translocal — or, to put it differently, not strictly territorial.

The internet is the canonical example — it has no territorial boundary as such. But beyond that, we’re seeing more and more management of various other commons — the climate commons, the counter-pandemic commons during COVID, including mRNA vaccination, space as a commons, and many other commons as well.

The governance of these commons has overlapping jurisdiction with territorial borders. And each of us, if we’re connected on some Discord or social network, is also participating in these kinds of networked polities that presuppose a non-territorial boundary.

So we’re in a place where the territorial, Newtonian measurement and the network-graph-shaped waveform coexist as polities. Competing ideologies don’t necessarily have to collapse into each other; they can also coexist in symbiotic arrangements.

Hiroki Azuma:

Let me ask just one very simple question. Audrey, what do you think about the future of the nation-state? How strong will the nation-state remain from here on? Will it grow weaker, grow stronger, or disappear altogether? What’s your sense of the future of the nation-state?

Audrey Tang:

Even if it’s a very leaky abstraction — from my own experience, many people now identify more with a nation or a neighborhood that they experience across state boundaries. Take the across-state “nations” that believe in the singularity worldview: they see themselves as a nation with tentacles reaching into many different states. And there are many translocal movements rising to counter those enormous surveillance capitalists, who are themselves suprastate. The movements countering them are likewise suprastate, not bound to any single state.

So my feeling is that, especially when we’re talking about AI–human symbiogenesis, we’re looking at the kind of organism that isn’t bound by state borders to begin with. And so we have to prepare for a world where identity, relationship, allegiance, loyalty, and so on — between machine substrates and organic substrates — are no longer conveniently contained within state borders.

I’m not saying the nation-state will be replaced. I’m just saying, as Buckminster Fuller would put it, that perhaps new abstractions will simply make the old ones obsolete — not by competing with them directly as abstractions, but by quietly making them irrelevant.

Hiroki Azuma:

I agree with the vision. But I can’t be as optimistic.

In reality, the nation has never been just a fantasy. The more globalization advances, the more networks develop, the more cross-border exchange increases — the stronger the backlash of nationalism becomes. Over the past ten to twenty years, at least in Japan, nationalism has clearly been on the rise.

So I don’t think things will go as smoothly as Audrey describes. But as a vision, I agree with it. I hope we can get there too. I just can’t be that optimistic.

Audrey Tang:

Let me clarify one thing. When I say “nation-state” is a leaky abstraction, I’m not saying that nation or communities in general are illusions. What I mean is: if you bundle the state apparatus and communal spirit into a single package called “nation-state,” that’s a leaky abstraction.

So the real work ahead is an open question: is it possible to use technology to strengthen communitarian traditions — what, together with Patrick Deneen, I call techno-communitarianism — without being captured by the state machinery, the “monopoly on the legitimate use of coercive force”?

---

Q4: Schismogenesis, Positive and Negative Feedback, and Community Maintenance

Student D:

I’d like to push this further. If we think in terms of schismogenesis — Gregory Bateson’s concept — what we’re really describing is a process where differences keep amplifying, splitting apart. Taken to its end, you just get a fracture between super-entities.

So we can’t only think about positive feedback. We also need negative feedback. Bateson described it through a Balinese example: a mother deliberately provokes her child, and when the child protests, she simply ignores the protest. That ignoring becomes the negative feedback — it prevents the antagonism from escalating without limit.

So if we want to maintain a community while also preserving noise — preserving the productive tensions within it — we need a system that can hold both positive feedback and negative feedback at the same time. How do you actually build something like that?

Hiroki Azuma:

I think what you’re really asking about is politics itself. Political imaginations fork, but at some point they have to be merged.

But my own thinking goes like this: historically, the positive-feedback and negative-feedback roles have been played by politics and literature, working in a kind of division of labor. Politics amplifies the differences between people. Literature creates empathy — even across vast differences, literature lets you feel that “we’re the same kind of being after all,” and that shrinks the sense of distance.

In modern society, politics and literature have roughly been performing the roles of positive and negative feedback.

But I think that relationship has actually reversed. Today, it’s words that divide people — and things, shared objects, common creations, that connect them.

The key concept here is durability. People creating the same thing together — that’s what I’d call a commons. Tradition is actually a commons that everyone built together. And the act of many people making one thing together — that act itself stitches together the divisions between them.

Technology is useful here. It can help us understand others, but it can also help us build something together. The shared act of creation is what repairs the fabric between people.

I think HUMAI itself is that kind of place.

Audrey Tang:

I think that’s a very good insight. Because what I’ve found — going back to how I used the language model at the very beginning — is that I wasn’t just translating the Japanese words in Peace and Stupidity. I was translating the philosophical architecture, converting it into something I could understand — into Buddhism, into Taoism.

I think that kind of translation is genuinely an act of waging peace. It builds a common story, a shared narrative — one that feels native to me, that doesn’t feel political. And when I share it back by showing my slides, it also builds a bridge between us, because it’s literally a mapping between the terminology used in one book and the terminology used in the other.

So this is not some fantasy or science fiction. It’s literally what I did today on the plane.

But I think, rather than saying we need a simple story that everyone can repeat — a simple network where you feed in conflict and it feeds back a peaceful narrative — I think that is fantasy. It’s very difficult to build that machine, even though, in honor of the late Habermas, there are people building what they call a “Habermas machine.”

What I’d suggest instead is that we think of it not as a network but as a hyper-network.

That is to say: you feed in conflict. But instead of outputting a single story — or outputting “insufficient information, I’ll have to think about it” — you output what’s called a low-rank adapter. A small, localized extension of your own thought pattern. Like a small story you tell yourself.

And then that conflict becomes palpable — it feels peaceful to you. But only to you. And then other people deploying the same hyper-network can also output low-rank adapters into their own cognitive systems. So even though everyone is still telling different stories, they can see — like Community Notes on X, YouTube, or Facebook — that certain notes carry an unlikely willingness to be supported by people from widely different and even warring ideologies.

Somehow those notes carry the hyper-network seed. They can travel back, and the person who cares about climate justice and the person who cares about creation care from the Bible can both feel: “That’s my story.”

We’ve already deployed this tool in Collaborative Notes on X. Instead of people manually writing community notes to build bridges, now Grok and other AI systems are creating hyper-network outputs — storytelling robots that generate the narratives that bridge those divides.

So I’d encourage all of you to look into Collaborative Notes. If it works, we’ll spread it to other social networks, not just X.

---

Q5: Interface Design and Bias

Student E:

I’d like to ask about interface design in the context of Plurality. If we’re pursuing a Plurality-quality interface — one that supports collaboration with others, one that doesn’t leave anyone behind — it also has to be simple and easy to use. But an “easy” interface carries a risk: it can steer people’s opinions and introduce bias — a kind of danger, the danger of dark patterns.

So what does critical engagement with interface and media design look like? And who should do it, with what motivation?

Audrey Tang:

I distinguish between an extractive interface and a convivial interface.

An extractive interface makes us increasingly dependent on the intermediary. A convivial interface, as I mentioned, is one where — after the bridge is built, it can go. It’s not needed anymore.

For example, this is the actual bridge I showed just before our conversation. It was built by a language model — it simply maps the philosophy as articulated in The Philosophy of Correctability against the philosophy in Plurality. It’s a two-column table. Nothing more.

But as you can see, I’m not looking at it while we’re talking. It was just a handshake of sorts. I presented it, we talked through it, and then we started correcting each other.

So the point is: between any pair of interlocutors, you can deploy this kind of very simple interface. Once you deploy it, you cross the bridge. Once you cross the bridge, the bridge can go. From that moment on, we’re not dependent on the language model in any way.

This idea — that it’s impermanent, that it serves a particular relationship, and that once the relationship is healthy it can fade away — I think that’s the most important design criterion.

If you have many such commons available, each serving relational health, then you’re never locked in to any particular one. On civic.ai, this is described as the fifth pillar: solidarity.

Hiroki Azuma:

I want to add something here.

What’s really impressive about Audrey is this: she used an LLM to engage with the content of Peace and Stupidity, but she hasn’t once said she “understood” it. What she said was: this is just a bridge. Like a handshake on the street. And it actually worked.

She deliberately made this slide in Japanese. Her LLM summarized my book in a certain way, and then she showed it to me backstage and asked, “What do you think?” Through that, communication became remarkably smooth.

I think the way Audrey uses AI is truly excellent.

I’m a writer, so naturally I want people to read the book. Maybe Peace and Stupidity will be published in English one day. I’d like people to read it then. But right now, in this moment — how do two people meeting for the first time take the very first step in communication? This slide was enormously useful.

AI can be used this way. Today, Audrey gave a very strong practical demonstration of that.

An interface is probably something like this. It’s very good.

---

Q6: Habermas, Communication, and Plurality

Student F:

Habermas passed away just last night, and that felt symbolic to me — which is why I wanted to raise this. Habermas built democracy on communication, so I wanted to ask how communication is understood within Plurality. Communication is also central to information technology: a Shannon-type communication model passes noise through a filter to extract the meaningful signal it carries. But the communication Plurality aims at seems to be the opposite — recovering plurality, multiple voices, from what was previously dismissed as noise, letting them be heard as plural voices. How do you see it?

Audrey Tang:

I wouldn’t frame it as extracting, isolating, or performing some kind of Fast Fourier Transform.

My thinking is more like this: each of us belongs to many communities simultaneously, and each community has its own very different norms of communication. In a sense, rather than saying we’re individuals who inhabit these communities, it’s more accurate to say we’re a team of individuals linking these communities together.

So if you take communities themselves as the primary agents, and understand individuals as interference patterns formed between those communities — if you think the communities and the health of their relationships are the real subjects, and individuals just happen to inhabit those relationships — then these signals are no longer noise. They become wave patterns you can resonate with, allowing you to feel the dissonance and harmony between communities.

That’s why I keep using the wave metaphor. In a sense, it’s not even a metaphor. Society has always worked this way: not through some top-down ontological command demanding universal compliance, but through the ongoing disputes, conflicts, and correctability between communities — communities that overlap within each of us as individuals.

So to answer your question more directly: when we build Plurality as an interface, we’re not saying “pluralism is good, let’s promote more of it.” We’re saying: conflict is already here. The question is how to make conflict a productive source of energy.

So Plurality isn’t a teleological endpoint to be pursued. It’s a relational ethic that already exists, here and now — and within each of us as well.

Hiroki Azuma:

What Audrey just said was very well put. That kind of sympathetic vibration that uses the individual as a boundary between communities — in a sense, that’s also emotion.

But Habermas’s concept of the public sphere is deeply rooted in language and rational reason. For him, a public sphere created by waves of shared feeling would probably be very hard to accept.

The result is that Habermas ended up being very Eurocentric — specifically, Western European-centric.

For example, in my book Peace and Stupidity, I wrote about the former Yugoslav civil war thirty years ago. When NATO bombed Belgrade and intervened in Kosovo, Habermas offered almost no criticism. In his framing, “over there” was rational civilization, and “over here” were barbaric states. For a rational civilization to attack a barbaric state with force was justifiable — that’s essentially what he said. This issue connects directly to the current U.S. attack on Iran.

So I’ve always felt there’s a deep problem at the very origin of Habermas’s thought.

As a philosopher, what I keep thinking about is that shared feeling — especially the kind mediated not just by language, but also by culture, commodities, and other channels — plays an enormously important role in the formation of public life.

As Audrey just said, various waves interfere with one another. What we really need is a kind of public image that can be generated from those interferences.

---

Q7: The Balance Between Public and Private Self

Student G:

My question is this: people ultimately can’t fully understand each other — I feel that’s a basic premise. Given that, how can people who can’t fully understand one another still coexist, protecting themselves while not losing their sense of public life? What kind of personal will is needed, what kind of supporting technology, what kind of sense of balance?

Put simply: how do multiple people who can’t fully understand each other live together — protecting themselves while not losing public life?

Audrey Tang:

Let me first check whether I’ve understood your question correctly.

Are you saying that if I try too hard to make everyone fully understand me, it becomes a kind of performance — like cosplay — where I have to pour all my energy into it? But on the other hand, if I’m too closed off, keeping all my judgments and thoughts inside, just going along with others, then understanding becomes impossible, because people have no idea what’s going on inside me.

So you’re asking: between being open and being private, what degree of balance actually feels livable? Is that right?

Student G:

Not exactly. I mean something more like: some people assert their views very strongly — “I’m definitely right” — and keep pushing that position outward. Others are less inclined to put their own will front and center, preferring to listen to what’s happening, to collide with the other person’s perspective.

But if you sacrifice yourself too much, you burn out — the kind of burnout you see happening a lot with social entrepreneurs. Every time I think about this kind of balance, it’s actually painful.

So what I want to ask is: how do you navigate the balance between those two sides?

Hiroki Azuma:

Isn’t that ultimately something you just have to figure out for yourself?

Student G:

I mean... I’d just like to hear how Audrey actually handles this in practice.

Audrey Tang:

All right. I sleep eight hours every night. If I’m facing a difficult philosophical problem, I do “overtime” — which means sleeping ten hours.

For me, that sleep time is completely private. I don’t let myself be exposed to anyone else’s overlap. I don’t even think about what other people might be thinking. In a sense, I’m not even in the same world — because I’m dreaming.

This is the self-care I absolutely have to maintain. If I only slept four hours the night before, I can’t do anything well the next day.

But on the other hand, during the roughly sixteen hours I’m awake, I’m entirely sociable. For me, the overlap of self and other is actually a plus. Because if I can use a language model to read a room with reasonable accuracy, I can also “write” the room, “append” the room, and help that room reflect back on itself.

That’s how I work. Because it’s not for personal gain. It’s for shared understanding, shared knowledge, and culture itself.

So if you want to do cultural work, there are moments when you must be open — when you must let the self and other overlap. But if you don’t have a strong enough personal boundary — a boundary around solitude, being alone, or in my case, around dreaming — that kind of strength simply can’t grow.

Because without that boundary, it’s not bidirectional. You’re just being affected by the room, but you have no energy to write it.

And to have the energy to write the room, you need a strong personal boundary. For me, that boundary is sleep. For others, it might be a solitary walk, or some kind of personal ritual.

---

Closing

Moderator:

Thank you. To close, could each of you offer one final thought? Audrey first.

Audrey Tang:

I’ve set all my screens to grayscale. Because if they’re in color, I keep scrolling, and I start feeling like the screen is more vivid than reality — which is very bad for my mental health.

So I just switched everything to grayscale.

Now I don’t keep scrolling, because reality is far more vivid than any screen.

So my one line is: Scroll less, sleep more.

Hiroki Azuma:

What I want to say in closing is this: throughout today, I was struck very clearly by the fact that Audrey consistently treats AI and LLMs as nothing more than bridges.

I think that’s very important.

AI and LLMs — or more broadly, technology itself — should never become the end goal. The moment technology becomes the end, people start optimizing themselves for the technology, and that’s when we become its slaves.

Technology is, in the end, just a tool. LLMs are tools for enhancing communication between people.

That’s what Audrey has been saying over and over again today. And I think it’s a very important guiding principle for everyone here, in your future research and development.

I came here today feeling quite nervous, but under Audrey’s warm guidance, we had a truly enjoyable discussion.

Thank you very much.

---

Learn more about our Civic AI and 6-Pack of Care framework inspired by the Taiwan Model🇹🇼 of digital democracy.