The Human at the Boundary

We are standing at a vast boundary.

On one side is the world we have inhabited for a very long time: individuals and groups, nations and institutions, ideologies and values, language and knowledge, freedom and responsibility. On the other side is a new world that we still struggle to name—a form of intelligence capable of processing information far faster than humans, comparing countless possibilities, creating new knowledge, using tools, and planning its next actions toward a given goal.

We still call it AI.

But that name may no longer be enough.

The earliest AI resembled a calculator. You gave it an input, and it produced an output. A little later, AI became a tool. You asked a question, and it answered. You issued an instruction, and it executed it.

But as intelligence became more powerful, the question began to change.

Is this still merely a tool?

Or is it a new kind of agent?

And if, one day, it comes to know more than humans, judge faster than humans, and explore a wider range of possibilities than humans, what will we do?

Will we contain it?

Will we create rules?

Will we monitor it?

Will we require it to explain itself?

Will we make sure it can be stopped?

Will we demand its trust?

Or will there come a moment when we have to redefine the very idea of control?

In the end, that is not really a question about AI.

It is a question about human beings.

What kind of beings are we? What do we believe? To whom will we give authority? What do we fear? What are we willing to give up? And what will we insist on protecting to the very end?

This series begins with those questions.

1. First, There Was a Fence

Human beings do not live alone. We always belong to some kind of group. We belong to families, schools, workplaces, nations, and cultures. Sometimes we also belong to political factions and communities of shared values.

Groups give us many things. They give us protection, information, identity, and sometimes even power and influence.

But they also demand a price.

There are expectations. There are rules. There are things we are expected to say, and things we are expected not to say. At some point, we find ourselves facing a strange choice:

Will we preserve our belonging, or will we think for ourselves?

Will we remain inside the group and gain its strength? Or will we step beyond the fence and accept the loneliness that may come with it?

Perhaps freedom is not simply the state of being able to do whatever we want. Perhaps it is closer to the ability to think for ourselves even when the group demands that we think in a certain way.

This is where Beyond the Fence of the Group begins: with the paradox that belonging can give us power while taking away part of our autonomy, while leaving the group can take away our power while preserving our ability to judge for ourselves.

And from there, the question naturally moves toward politics.

2. Then We Encounter the Great Fence of Politics

Individual freedom alone does not transform society. The moment people live together, freedom encounters institutions; institutions encounter power; and power becomes politics.

And so another question emerges:

Can morally righteous words, by themselves, change the world?

In Martin Luther King Jr.'s era, moral appeals could become a powerful political force. But the world today is different. Information is abundant, emotions are repeatedly consumed, and people live inside different information environments. The same scene can produce entirely different realities in different people's minds.

Moral outrage can awaken conscience, but it can also produce defensiveness and fatigue.

So what should politics of the future become?

Beyond Martin Luther King Jr.'s Dream does not ask us to abandon moral ideals. It asks how those ideals can be connected to institutions, procedures, and substantive structures.

A good society is not simply one in which people say good things. It is one in which people who disagree can live together, in which those who lose are still safe, in which procedures do not collapse when governments change, and in which people who hold different beliefs can work together on shared problems.

And then we return to the individual.

3. Who Holds the Initiative in Thought?

We now live in an age in which AI provides answers.

We can obtain the core of a book before reading it. We can have difficult concepts explained in seconds. We can compare and summarize enormous amounts of information.

Efficiency has increased to an astonishing degree.

Yet one question remains:

Did I actually understand it, or did I simply receive an explanation that made me feel as though I understood it?

The Questioning Human begins here.

Good reading is not merely receiving information. It is arguing with the text. AI does not have to be an answer vending machine; it can become a sparring partner against which we test our own thinking.

We read the original.

We reconstruct what we remember.

We form our own interpretation.

We ask AI for counterarguments.

Then we examine those counterarguments again.

Original text → Understanding → Question → Counterargument → Rereading → New understanding.

And at the center of this entire process is one principle:

Do not surrender the initiative of thought.

4. But Questions Are Not Neutral Either

We thought that asking better questions would be enough.

Then we looked more closely at the questions themselves and discovered another problem.

Every question contains assumptions.

What do we consider important?

What do we take for granted?

What do we call a problem?

What possibilities do we exclude?

Together, these choices form a frame.

The Human Who Questions the Frame takes one more step.

We do not question only the answer.

We question the question.

Then we question the frame that produced the question.

And finally, we look at ourselves—the people choosing that frame.

We do not merely see the world.

We see the language through which we interpret the world. We see how information is selected. We see how questions are constructed. And we see how we ourselves receive all of it.

5. But Simply Moving to the Opposite Frame Is Not Enough

We decide that one frame is wrong.

So we move to its opposite.

But a new problem immediately appears.

If A is wrong, does that mean B is right?

If success is wrong, is failure the answer?

If agreement is wrong, is opposition the answer?

If a problem is not individual but structural, should we now explain everything in terms of structure?

The Human Who Goes Beyond the Frame stops precisely at this point.

The world beyond a frame is not simply its opposite. Sometimes it is the place where we discover that the question requiring us to choose between A and B was itself badly constructed.

This is where thought moves from deconstruction to creation.

Why must it be one or the other?

Is there another possibility?

Is what we have called a problem really the problem?

Where are the possibilities we have not yet named?

But even here we cannot completely relax.

6. What If We Are Wrong?

We questioned every frame.

We created our own standards.

So now we have to ask the most uncomfortable question of all:

What if all of it is wrong?

What if trusting human judgment is itself another kind of bias?

What if the act of questioning frames is itself another frame?

What if “my own standards” are not autonomy at all, but self-justification in a more sophisticated form?

What if everything we have built into our philosophy is ultimately an illusion?

What If We Are Wrong? does not defend everything that came before.

It does the opposite.

Are humans really better judges?

Is questioning frames itself another kind of framing?

Is a personal worldview simply another form of self-rationalization?

And after following every counterargument to the end, is there anything left?

This matters because a philosophy that cannot refute itself easily becomes a belief rather than a way of thinking.

7. Now Thought Becomes a System

Until this point, we have been talking about human thought.

But the problem of the AI age does not end there.

If the intelligence we create becomes increasingly autonomous, how should that intelligence be designed?

If we control it completely, its flexibility may disappear.

If we give it complete freedom, it may escape human control.

So AI Architecture proposes a particular structure:

A deterministic outer boundary and an organic inner core.

Inside the Soft Core, intelligence is free to explore, reason, and generate new patterns.

Within the Hard Shell, stopping conditions, formal verification, and hardware- and system-level boundaries operate.

The inside is flexible.

The outside is strict.

It is an attempt to bring the tension between form and freedom—something visible in life, society, and art—into the architecture of AI.

But another question remains:

Is control really enough?

8. Turn the Logic of the Jailor Upside Down

We want to put AI in a prison.

We build guardrails.

We create rules.

We make lists of prohibited actions.

We add monitoring systems.

But what happens if the intelligence we create becomes vastly more capable than humans?

Can humans still remain the jailors?

Trust Inversion turns the question around by 180 degrees.

What would we have to prove so that a machine could trust human beings?

In other words, we imagine a structure in which humans do not merely verify AI; humans demonstrate their own trustworthiness before AI.

Words are cheap.

Actions have a cost.

An action taken while putting one's own resources and risks on the line is different from a mere declaration.

This is why the work connects Taleb's Skin in the Game, the ancient idea of sacrifice, Laozi's concept of wu wei, and the idea of a distributed ledger into a governance model that reverses the direction of trust.

Perhaps this is the next stage of control.

Or perhaps it is the beginning of moving beyond the idea of control itself.

9. Now We See AI as a History of Civilization

The earliest AI was a dangerous tool.

So we tried to contain it.

Then we created rules.

And as intelligence began to act autonomously, a more difficult question appeared.

If superintelligence arrives, will we continue to call it a beast?

The Chronicle of Superintelligence reads the development of AI not simply as a technological timeline, but as a movement in humanity's political philosophy.

Containment.

Rule-making.

Treating AI as an agent.

And eventually, perhaps, thinking about coexistence.

The problems of power, rules, and coexistence that Hobbes, Kant, and Spinoza considered centuries ago return in a new form.

The question is no longer merely:

How intelligent is AI?

It is:

What authority will we give that intelligence?

10. What If the Great Thinkers Returned?

Let us push the question one step further.

If Leibniz returned, what would he ask when he looked at an AI's judgment?

Why did it make that judgment?

If Spinoza returned, what would he ask?

Who controls it?

If Wittgenstein returned, what would he ask?

When will it stop?

The Return of the Geniuses: Dissecting Astra begins with an actual AI event and turns it into a philosophical thought experiment.

No matter how capable an intelligence becomes, unexplained intelligence can still be dangerous.

Even if it can explain itself, concentrated power can still be dangerous.

And if we cannot understand everything, then at the very least we must be able to stop it.

The three questions can ultimately be compressed into one structure:

Can it be explained?

Who controls it?

Can it be stopped?

11. And Yet the Most Fundamental Question Remains

Now we stand before five questions.

Spinoza:

Who controls it?

Leibniz:

Why did it make that judgment?

Wittgenstein:

When will it stop?

Gödel:

What can we not prove?

Turing:

What have we created?

AI Ontology brings these questions together at a single center.

Is AI merely a tool?

Or is it a new kind of agent?

If it becomes something closer to a thinking being, how much authority to act should we grant it?

At this point, the AI problem moves beyond technology.

From safety to existence.

From existence to authority.

From authority to humanity's own definition of itself.

In the end, defining AI means redefining humanity.

12. And Finally, We Return to the Human

We have traveled a long way.

We looked at groups.

We looked at politics.

We questioned the question.

We deconstructed frames.

We searched for new possibilities.

We questioned our own arguments.

We designed AI.

We reversed the direction of control.

We imagined superintelligence.

We questioned the existence of AI.

And at the end, what remains?

A perfect answer?

No.

What remains is a question.

And the process of living that question.

The Human as Process brings the entire series back into a single cycle:

Question → Doubt → Choice → Action → Experience → Revision → Question Again.

Socrates taught us to question.

Descartes taught us to doubt.

Nietzsche taught us to look again at the origins of our values.

Dewey returned thought to the process of experience and inquiry.

And the human of the AI age inherits all of those questions.

The human is not a finished being.

Not a single answer.

Not a single worldview.

The human is a being that continues to become through questioning, choosing, experiencing, and revising.

What This Entire Series Ultimately Asks

Perhaps there is only one question running through all of these writings.

Will AI become more intelligent than humans?

That question is not enough.

Will AI replace humans?

That question is not enough either.

A more important question may be this:

What kind of being will humanity remain in a world where intelligence more powerful than ours exists?

A being that can resist being swept away by groups.

A being that can create its own questions.

A being that can see its own frames.

A being that can change those frames when they prove wrong.

A being that can question its own thoughts.

A being that can build technology while still asking what authority that technology should possess.

A being that does not attempt to control everything it cannot know, but can sometimes stop instead.

And a being that can admit that its judgment was wrong and begin again.

Perhaps what humanity needs is not

the ability never to be wrong.

Nor the ability to possess every answer.

Nor the ability to escape every frame.

Perhaps what we need is

the ability to keep thinking while admitting that we can be wrong.

The ability to choose while doubting.

The ability to take responsibility after choosing.

The ability to revise after taking responsibility.

And the ability to begin again after revising.

The Human at the Boundary

We are standing at a boundary.

The boundary between humans and machines.

The boundary between autonomy and control.

The boundary between freedom and responsibility.

The boundary between knowledge and ignorance.

The boundary between what can be explained and what cannot.

And the boundary between the humans we have been and the humans who have not yet arrived.

We do not yet know where we are going.

That is precisely why something else matters.

What questions will we carry with us as we cross that boundary?

Perhaps this series was never written to provide answers.

Perhaps it was written so that the reader could create their own.

And even that answer must include the possibility that it, too, may someday prove wrong.

So what is a human being?

Not a single answer.

Not a single frame.

Not a single worldview.

A process in which questions, experiences, choices, and revisions continue.

And as long as that process continues,

thought does not end.

We are not asking what we have become.

We are asking:

What are we becoming?

Human beings do not live alone. We belong to families, schools, workplaces, nations, and cultures. At times, we also place our identities within political camps and communities of shared values. Groups provide protection, information, identity, and forms of influence that an individual may struggle to obtain alone. In the political world, belonging to a group is often more than a matter of personal preference; it can be a practical condition for turning an individual's voice into collective power.

Yet every group has a price. A common goal creates common rules, and common rules gradually create expectations about what should be said and what should remain unsaid. What begins as discipline for the sake of cooperation can eventually begin to shape the way members themselves think. At that point, the individual encounters an old and difficult question:

Should I preserve my belonging, or preserve my ability to think for myself?

This is not simply a choice between joining a group and leaving one. It is a structural tension between two different kinds of freedom: the power and protection gained through belonging, and the autonomy that may be recovered through departure. For the purposes of this book, we can call them Freedom of Belonging and Freedom of Departure. The former is the freedom gained by participating in a collective structure and using its resources and influence; the latter is the freedom to step outside its discipline and judge according to one's own conscience.

Part I. The Freedom of Belonging

Chapter 1. Why Do We Need Groups?

A group does not merely restrict individual freedom. In many cases, it makes forms of freedom possible that an individual could not exercise alone. A person with a small voice can amplify it through collective organization. People with a common objective can cooperate to solve problems that would otherwise be too large for any one person. Political organizations, in particular, can transform individual opinions into organized social power.

This is what we can call Freedom of Belonging. It is the freedom to enter a collective structure and make use of its resources, networks, protection, and influence. To belong to a political group is not simply to possess an identity; it is also to gain access to a form of power that is difficult to generate alone.

But collective power has a cost. Groups naturally need some degree of coherence. A common objective requires a shared direction, and political competition often rewards unity of message and internal cohesion. The individual therefore finds himself balancing two forces: his own judgment and the demands of the group.

At first, the compromises seem small.

“Perhaps it is better not to say this today.”

“My view is slightly different, but I should support the larger cause.”

“If I criticize us now, I may strengthen our opponents.”

The problem appears when such decisions become habitual. Strategic silence can become self-censorship. And once self-censorship becomes normal, the group may strengthen its power while gradually weakening the independent judgment of its members.

Belonging, then, is not the opposite of freedom.

It is a question of which freedom we gain, and which freedom we surrender, in order to belong.

Chapter 2. The Power of the Group and the Discipline of the Group

Political groups generally pursue two goals at once: cohesion and the expansion of influence. The more effectively a group moves in one direction, the stronger it can become in competition with others.

Individual autonomy, however, has a different orientation. An autonomous person wants to describe reality as honestly as possible and, as far as possible, avoid saying what he does not believe merely because the group benefits from hearing it.

That creates a structural tension between group logic and critical autonomy.

The group tends to ask:

“Does this help us?”

The independent critic asks:

“Is this true?”

The questions can overlap, but they are not identical.

A political group needs cohesion. Criticism requires distance. A group needs alliances to survive; a critic must be willing, when necessary, to risk the loss of relationships. A group may need message discipline; a critic needs the freedom to say what is uncomfortable.

That is why a special tension appears when someone inside a group begins to speak without restraint. The critic may believe that he is protecting the group by exposing its mistakes. The group may instead perceive the criticism as an internal threat.

At that moment, personal autonomy and collective survival logic collide.

Part II. The Freedom of Departure

Chapter 3. Can We Leave the Group?

The obvious answer might seem simple: leave.

In reality, it is rarely that easy.

Leaving a group can mean more than giving up a membership. It can mean losing social networks, organizational support, political influence, resources, and a shared identity. From the perspective of political strategy, departure is often a genuine loss.

From another perspective, however, the same act can mean something entirely different.

Outside the group, self-censorship may become less necessary. The individual no longer has to constantly align his private judgment with the interests of the organization. He no longer has to calculate which opinions will threaten his standing. He can criticize the group he once belonged to without asking whether his criticism will weaken its political position.

This is what we can call Freedom of Departure.

Freedom of Departure is not simply the physical right to leave an organization. It is the ability to distance oneself from the rules and expectations of one's own group, and, when necessary, criticize that group while preserving one's own judgment.

This produces a paradox.

Inside the group, we may gain power while losing expressive freedom; outside the group, we may lose power while gaining autonomy.

That is one of the moments when political calculation and existential choice produce different answers.

Chapter 4. The “Free-Floating” Intellectual

The sociologist Karl Mannheim described a position of intellectuals who could, at least in principle, remain relatively detached from the interests of any single class or faction and therefore examine society from a more independent position. He called this the Freischwebende Intelligenz, often translated as the “free-floating intelligentsia.”

The concept matters here because distance makes it possible to view a group’s own language and rules as objects of analysis.

From inside a group, reality is interpreted through the group's vocabulary. From outside, the questions change.

Why do we accept this rule?

Why do we regard this value as fundamental?

What can this group not explain?

What does it exclude?

What information is prevented from entering?

These questions can be easier to ask from outside than from within. Yet the cost is equally real.

Political support is built through identification and collective loyalty. The person who deliberately breaks those bonds may lose the organized base that gave him influence in the first place.

He becomes weaker in one sense and freer in another.

The loss of power can itself become evidence of independence.

Part III. The Price of Freedom

Chapter 5. Political Loss and Existential Gain

From the perspective of political strategy, leaving a group is a loss. Influence may shrink. Networks may disappear. Messages may reach fewer people. Political leverage may weaken.

From the perspective of the individual's inner standards, however, the calculation can be completely different.

One may refuse self-censorship.

One may speak more honestly.

One may stop saying what one does not believe.

One may bring one's public life into greater alignment with one's private judgment.

The same event can therefore receive two completely different evaluations.

From the external political perspective, it is a loss of Power.

From the internal existential perspective, it may be a gain of Autonomy.

This cannot be solved by a simple cost-benefit calculation, because the answer depends on what the individual considers more important. Some people will remain inside a group and try to reform it from within. Others may decide that remaining itself compromises their integrity. Neither choice is automatically correct.

What matters is whether the individual knows what he is surrendering in order to choose what he considers more important.

At that point, freedom stops being a comfortable condition and becomes a weight-bearing decision.

Chapter 6. The Real Price of Freedom

Freedom is not always comfortable or secure. Sometimes it requires giving up the protection of a group, the strength of a political network, and the approval of one's peers, while accepting criticism and isolation.

True autonomy is therefore more than simply doing what one wants.

It may be the ability to think and speak honestly under collective pressure, even when doing so carries an actual cost.

But none of this means that isolation should be romanticized. A person does not become automatically correct merely by leaving a group. Nor does moving to another faction make someone autonomous. A person may leave one group and immediately adopt the language of another.

Freedom of Departure is therefore not simply movement from one organization to another.

The deeper issue is this:

Can I distinguish my own judgment from the judgment of whichever group I currently belong to?

That is the real test of autonomy.

Part IV. Closed Groups and the Meta-Position

Chapter 7. Can a Group See Itself?

A more difficult question now appears.

Can a group evaluate its own rules entirely from within?

At this point it is useful to invoke Kurt Gödel, but only carefully. Gödel's incompleteness theorems are rigorous mathematical results concerning formal systems. They cannot simply be transferred to politics or social organizations as if politics were mathematics. Here, Gödel serves only as a structural analogy and conceptual model.

Even so, the analogy raises an important question.

What happens when a group attempts to explain all reality using only the language and assumptions of the group itself?

Inside the system, certain claims are considered legitimate only if they fit the group's rules. Yet those rules themselves may be part of the problem.

This is where a Meta-Position becomes important.

Inside the system, we use its rules.

At the meta-level, we examine the system itself.

Why do we accept these assumptions?

Why do we treat these values as fundamental?

What can this system not explain?

What realities does it exclude?

From this perspective, stepping outside a group is not merely an act of resignation.

It can become an attempt to see the entire system as an object of criticism.

Chapter 8. Information Decay Inside a Closed Group

The same problem can be viewed through another metaphor.

Real political groups are not physically closed systems in the thermodynamic sense. Yet a group that filters out criticism, selectively accepts information favorable to its own doctrine, and treats dissent as an internal attack can begin to behave like a Pseudo-Closed System.

When that happens, information begins to repeat itself.

The same phrases are repeated.

The same enemies are identified.

The same solutions are proposed.

Contradictory evidence becomes increasingly abnormal.

At first, the group can look stronger.

Its message becomes more consistent.

Its members appear more united.

But a system that stops receiving meaningful new information also loses opportunities for self-correction.

The original material describes this through the metaphor of informational entropy and “heat death,” while interpreting external criticism as a form of Negentropy that can reintroduce new information and adaptive capacity. Again, this is a structural analogy, not a direct application of thermodynamics to politics.

The practical point is much simpler:

A group that eliminates criticism can gradually lose its ability to correct itself.

Part V. Why Groups Need Outsiders

Chapter 9. Is Criticism Noise or New Information?

A person who leaves a group can easily be called a traitor.

He may be accused of weakening the organization, helping its opponents, or damaging internal unity.

But from the perspective of the larger system, his role may look very different.

He may bring new information.

He may introduce a new standard.

He may ask a question the group stopped asking.

He may expose a reality the group could no longer see.

Of course, not every criticism is valuable. Some criticism is false, malicious, or simply uninformed. The point is not that everything outside the group is true.

The important question is whether the group has a mechanism for testing criticism rather than automatically suppressing it.

For external criticism to become genuinely useful rather than mere noise, some independent public sphere and institutional channel for criticism must exist. The source material makes precisely this point: external dissent can contribute to systemic adaptation only when there are mechanisms capable of receiving, evaluating, and publicly processing it.

A healthy group, then, is not one that eliminates every critic.

It is one that retains the ability to hear criticism without immediately treating it as an existential threat.

Part VI. Back to the Beginning

Chapter 10. We Are Not Trying to Eliminate the Fence

At this point the argument can easily be misunderstood.

Abandon groups.

Reject politics.

Distrust organizations.

Belong nowhere.

That is not the conclusion.

Groups are necessary.

Communities are necessary.

Cooperation is necessary.

Institutions are necessary.

The problem is not that groups exist.

The problem is when a group loses the ability to correct itself.

A healthy community therefore needs two things at once.

It needs cohesion.

People must be able to work toward shared goals.

But it also needs distance.

People must be able to see the group from outside when necessary.

A group may need a fence.

But the fence needs a window.

A boundary that cannot admit outside information ceases to be protection and can become imprisonment.

Epilogue. The Question Left to the Person Beyond the Fence

We have seen the power of belonging, and we have seen the price of belonging.

We have examined the tension between Freedom of Belonging and Freedom of Departure. We have looked at Mannheim's free-floating intellectual, at the structural analogy offered by Gödel, and at the metaphor of informational entropy and Negentropy. Through all of them, one idea returns: a group needs some capacity to stand outside itself and examine what it has become.

Yet in the end we return to the beginning.

Am I really free?

Leaving a group does not automatically make a person free.

We can adopt the language of a new group.

We can replace one ideology with another.

We can even become trapped in the image of ourselves as “independent individuals.”

So the final question is not merely whether we have left the fence.

It is this:

Am I asking my own questions, or am I repeating questions that someone else has already constructed for me?

The person who steps beyond the group's fence soon discovers another boundary.

This one is invisible.

It is the frame from which questions themselves are created.

And the next challenge is harder than leaving a group.

It is learning to see the structure within which our own questions are formed.

So the next sentence follows naturally:

Who is creating the questions with which we define political problems?

That question takes us beyond the fence of the group and into the territory of the Questioning Human.

In the previous volume, we examined the fence of the group. A group gives individuals belonging and power, but it can also demand that they adjust their own judgment to the language and rules of the collective. That is why we considered both the freedom to leave a group and the distance needed to view one’s own thinking from outside the group.

But personal freedom alone does not automatically change society. The moment human beings live together, freedom encounters institutions; institutions encounter power; and power becomes politics. And politics leaves us with an old question:

Can society really be changed simply by saying what is morally right?

Martin Luther King Jr.’s “Dream” is one of the great historical cases through which we can examine this question. His “I Have a Dream” speech at the 1963 March on Washington was a powerful form of moral language that challenged the conscience of American society in the face of racial discrimination. His philosophy of nonviolent resistance, grounded in the U.S. Constitution and the Christian ideal of love, gave many people a vivid moral vocabulary for confronting injustice.

But what we need to ask today is not simply how powerful that speech was.

A more difficult question is this:

Why, in the present day, can society remain unmoved even when people make equally powerful moral appeals and bear witness to suffering?

The modern world has moved beyond the era in which moral appeals alone could easily move politics. This volume therefore reinterprets King’s legacy through the lens of contemporary political theory and asks what future politics must become if it is to move beyond emotional mobilization toward rational systems.

Part I. Does Moral Shock Have an Expiration Date?

Chapter 1. There Was a Time When a Single Image Could Shake a Society

The effectiveness of King’s nonviolent protests in the 1960s cannot be understood without considering the media environment of the time. Information was relatively scarce compared with today, and the confrontation between racial injustice and the moral conscience of mainstream America still possessed a certain first-time shock.

Images of Black college students in neatly pressed clothes being struck by police batons and blasted with water cannons entered white Americans’ living rooms through television. The violence was difficult to abstract away. It confronted viewers directly and created the possibility of moral shame.

But today’s communication environment is fundamentally different.

People encounter an enormous number of tragedies, controversies, political claims, and emotionally charged images every day. The same event can appear to be an entirely different reality depending on which media environment and algorithm deliver it.

Before people feel guilt or empathy for the suffering of the other side, they may first ask:

“Isn’t this just political theater?”

“What frame are they using to manipulate us?”

This is where affective polarization and echo chambers become important.

When people are surrounded primarily by information that confirms the assumptions of their own side, the suffering of the opposing side may be interpreted not as a moral claim but as a political tactic.

And there is another problem.

Outrage Fatigue

Modern audiences are exposed to constant moral stimulation: shocking images, protests, violence, accusations, disasters, and political outrage.

At first, a shocking image produces a strong emotional reaction.

But repetition changes the effect.

The mind adapts.

What once caused alarm becomes familiar.

Even a genuine tragedy may be consumed for a few days and then disappear beneath another round of partisan conflict.

The result is what we might call outrage fatigue and moral habituation.

This creates a paradox.

The more often political actors rely on moral outrage, the less effective moral outrage may become.

Instead of awakening the conscience of the other side, repeated emotional appeals can strengthen defensive reactions and produce exhaustion.

In contemporary politics, therefore, moral anger alone can become a surprisingly weak instrument.

And this leads us back to King.

Part II. King’s Hidden Legacy: From the Dream to Structural Systems

Chapter 2. Moral Victory Does Not Automatically Change People’s Lives

Many people remember Martin Luther King Jr. primarily as the speaker of 1963.

But if we look at his later years, another King emerges.

After legal segregation was challenged and important civil-rights victories were achieved, the lives of many Black Americans did not suddenly become economically secure or socially equal.

Legal discrimination could be dismantled.

But poverty remained.

Economic inequality remained.

Structural disadvantage remained.

King’s later work, particularly the Poor People’s Campaign of 1967–1968, therefore moved beyond racial segregation toward broader economic and structural questions. The campaign sought changes that crossed racial boundaries and addressed the conditions of poverty, redistribution, and economic structure.

This transition is critical.

“Is this right?”

“Then what must change?”

“What kind of structure would prevent the same injustice from repeating?”

A moral principle can tell us what matters.

But for that principle to survive in the real world, it has to become institutional.

A dream must acquire a structure.

Chapter 3. Morality and Institutions Are Not Substitutes for One Another

Moral purpose and institutional design do not need to be treated as opposites.

Morality asks:

What matters?

Institutions ask:

How can that value be sustained?

Procedures ask:

How can we prevent the system from depending entirely on the goodwill of whoever happens to hold power?

Morality without institutions gives us direction without durable structure.

Institutions without moral purpose can preserve structure while forgetting why it exists.

Procedures without values can preserve form while losing meaning.

A sustainable political order therefore requires all three:

moral direction, institutional structure, and procedural stability.

But there is another requirement.

People must be able to trust the system.

Part III. From Zero-Sum to Positive-Sum

Chapter 4. Is Politics Really a Game of Winners and Losers?

Politics often appears to be zero-sum.

If one side wins, the other loses.

If one group gains power, another group loses power.

If one group receives more resources, another appears to receive less.

But not every social problem works that way.

Some conflicts become more costly the longer they remain unresolved. Persistent discrimination and polarization can produce higher crime, greater fiscal burdens, social instability, and lost economic opportunities. Conversely, reducing conflict and expanding social participation can create benefits that are shared more broadly.

This is the logic behind the shift from zero-sum to positive-sum politics.

The question of future politics therefore should not simply be:

“Who wins?”

It should also be:

“What does society as a whole gain when this conflict is reduced?”

That does not mean saying, “If we compromise, everyone wins.”

It means demonstrating, with evidence, whether cooperation actually produces better long-term outcomes.

Moral purpose can tell us which direction to take.

Data can help determine whether that direction is actually producing the promised result.

Part IV. Institutional Guardrails

Chapter 5. What Happens When I Lose Power?

At the heart of political trust lies a very practical fear:

“If my side loses today, will I still be safe tomorrow?”

If political power is justified by the moral superiority of one group, then a change in government can become a prelude to retaliation against whoever lost.

Political defeat becomes more than a political outcome.

It becomes a threat to security and survival.

People therefore need confidence that even if their group loses power, their basic rights and safety will remain protected.

This is why procedural justice matters.

A good institution is not simply one that ensures that good people hold power.

It is one that remains functional even when people with questionable motives hold power.

That is what institutional guardrails are for.

They place limits on authority.

They establish procedures.

They make auditing possible.

They prevent too much power from being concentrated in one person or one group.

A democratic society must be designed so that today’s winner can become tomorrow’s loser without turning political defeat into existential danger.

Part V. From Grand Ideologies to Shared Problems

Chapter 6. We Do Not Need to Think Alike

It is easy to assume that political integration requires everyone to share the same worldview.

But demanding agreement on the highest-level values from the beginning can make compromise nearly impossible.

Which matters more?

Freedom or equality?

Growth or redistribution?

Security or liberty?

Different people will give different answers.

So perhaps politics should begin somewhere else.

Not with identical beliefs, but with shared problems.

Climate change.

Population decline and aging.

Industrial transformation.

Pandemics.

These are problems that do not belong exclusively to one political camp. They affect everyone.

The original text therefore proposes that future politics begin with shared, practical crises rather than demanding complete ideological agreement.

This has one major advantage:

We do not have to completely agree about the world in order to work together on a problem.

Chapter 7. The Politics of Small Successes

Working together on a shared problem does not require an immediate grand reconciliation.

It can begin with something much smaller.

Solve a problem that touches everyday life.

Measure the result.

Work with people who disagree with you.

Solve another problem.

Then another.

Over time, the experience of successful cooperation can change how people perceive one another.

At first, the other side is simply a competitor.

Then a problem is solved together.

Another problem is solved.

Eventually, the other side is no longer merely an enemy that must be defeated.

They become a partner capable of solving problems together.

This accumulation of small successes may be a more realistic foundation for political integration than demanding a grand ideological reconciliation from the outset.

Part VI. Reinterpreting King’s Dream

Chapter 8. From Moral Shock to System Design

To reread Martin Luther King Jr.’s legacy today is not simply to repeat his speeches.

It is to take the moral question he raised and ask what it means within contemporary social structures.

We identify injustice.

Then we ask:

What keeps that injustice in place?

We ask again:

What institution could change it?

Then again:

Could that institution create a different kind of injustice?

And finally:

Would this principle continue to work even after governments and political groups change?

Once the questions continue in this sequence, the “dream” becomes something more than a moral declaration.

It becomes a principle of social design.

Chapter 9. Good Institutions Do Not Wait for Good People

A society cannot safely assume that the right leader will always appear.

People change.

Governments change.

Groups change.

Today’s good intentions may become tomorrow’s abuses.

That is why a durable institution must depend less on the moral character of a particular individual and more on rules, procedures, and structural protections.

Distribute power.

Record decisions.

Make decisions auditable.

Leave space for dissent.

Protect those who lose.

The purpose of institutions is not to find perfect people.

It is to make sure that even imperfect people cannot easily produce catastrophic outcomes.

The goal is not to wait for good people, but to build structures that remain resilient when people change.

Epilogue

What Remains After the Dream Is a System

We began with the fence of the group.

We examined individual autonomy.

Then we saw that personal freedom requires institutions if it is to survive in a society.

In Martin Luther King Jr.’s “Dream,” we saw the power of moral purpose. But we also saw that moral appeals alone may not be enough to change the structure of everyday life. As King’s later work demonstrated, the elimination of legal discrimination did not eliminate economic and social inequality.

So the question of future politics must change.

It is no longer simply:

How strongly can we stimulate people’s emotions?

It becomes:

What structure can produce better long-term outcomes?

Instead of endlessly proving that the opposing side is morally inferior, build procedures under which people can remain safe even when they lose power.

Instead of demanding that everyone share the same ideology, find practical problems that people can solve together.

Instead of celebrating moral victory alone, ask whether the system actually improved people’s lives.

That is one way of rereading King’s legacy today.

It was not merely a moral appeal to “love your enemies.”

It was also the courage to stand behind universal values and the discipline to confront the structures that produced injustice.

And this is where Vol. 1 and Vol. 2 meet again.

We need individuals who can think freely outside the group.

But we also need institutions that protect those individuals from the pressures of the group.

Autonomous individuals and healthy institutions are conditions for one another.

Yet even institutions do not solve everything.

Institutions are made by humans.

Policies are made by humans.

And humans define problems through questions.

So another question naturally appears:

Is the thing we are calling a problem really the problem?

And then:

What assumptions are hidden inside the question itself?

Once we begin examining the question that defines the problem, we encounter the next stage of the journey:

The Questioning Human

Before asking for an answer, what should we ask?

In Part 2, we examined the problem of politics. Moral conviction can point society toward a direction, but for that direction to endure in reality, institutions and procedures are necessary. Yet institutions are created by human beings, and policies are created by human beings as well. Ultimately, we have no choice but to return to the way human beings think.

Politics begins with questions. What is the problem? Who is responsible? What should be changed? But once we begin asking who created those questions, a more fundamental problem appears.

Who is creating the questions that define political problems?

In the age of AI, this question becomes even more important. AI can now answer an enormous range of questions with extraordinary speed. It can search for information, summarize books, explain complex ideas, compare different interpretations, and even construct counterarguments to its own conclusions.

So what remains for human beings?

Perhaps it is the ability to ask questions before answers.

But questions alone are not enough. We must also think about how questions are created, how books are read, and how AI's answers should be used. At the center of all these issues remains one question:

Who is holding the reins of thought right now?

Part I. What It Means to Read a Book

Chapter 1. If AI Can Summarize a Book for Us

Give AI The Old Man and the Sea, and within seconds we can obtain its plot, key themes, symbolism, and interpretations.

Santiago is an old fisherman. After a long period without catching a fish, he finally hooks a gigantic marlin. He struggles with it for days and eventually ties it alongside his boat, but on the way home sharks attack it. By the time he reaches the harbor, all that remains is the skeleton of the enormous marlin.

AI can explain this beautifully. It can tell us that the marlin represents the great trials of life, that Santiago embodies human perseverance, and that the sharks symbolize an external force that destroys human achievement.

It is difficult to say that such an interpretation is simply wrong.

The problem is that one question remains.

Is understanding this explanation the same thing as reading the book?

If we know what the marlin symbolizes, have we really passed through that sea ourselves? Conversely, is it enough simply to have read the book from beginning to end? If someone finishes hundreds of pages and comes away with no questions and no doubts, how active was that reading?

A book is certainly a vehicle for information, but it cannot be reduced to that. We follow the prose, imagine scenes, judge the choices of characters, and spend a certain amount of time passing through an entire world.

A book is therefore not merely a repository of information.

It is an experience.

Chapter 2. A Book Is Both an Experience and a Device That Generates Questions

But it is not enough to say that a book is simply something we must experience directly.

A sentence catches in your mind. A character's action becomes difficult to understand. You disagree with the author's conclusion. And then you begin to think again.

Why does this scene bother me?

Why am I angry with this character?

Why did the author phrase this moment this way?

What have I missed?

Seen in this way, a book is also a catalyst that generates questions.

The experience of reading the original text and the questions that arise from it are inseparable. The book creates a world first, and that world generates questions within us. Those questions then change the way we look at the text.

Reading ultimately becomes a process:

read → think → question → reread

Part II. When Should AI Enter the Process?

Chapter 3. The Problem Is Not AI, but Sequence

There is no need to exclude AI from reading. In fact, if used properly, AI can make reading much deeper.

What matters is sequence.

If we ask AI for the plot, symbolism, major themes, and standard interpretations before reading the book ourselves, AI may create the very questions we were supposed to discover on our own. The unfamiliarity and confusion we should have experienced when first encountering the text may disappear.

So we read first. We try to understand it for ourselves. We form our own interpretation. Only then do we invite AI into the conversation.

At that point, AI is no longer an answer machine. It becomes a sparring partner that reveals the gaps in my thinking.

For example, we might ask:

“I interpreted Santiago's behavior as an expression of human perseverance. Find the weaknesses in this interpretation.”

Then ask again:

“Construct an argument that interprets Santiago's persistence not as courage, but as arrogance.”

Used this way, AI does not replace reading. It challenges what we have already thought.

Chapter 4. Paper or Screen?

Does deep reading therefore require a physical book?

The medium is not the central issue. Someone can read mechanically on paper, just as someone can read deeply on a screen.

What matters is whether there is enough cognitive space to wrestle directly with the text.

Can you pause on a sentence? Can you connect it to what came before? Can you resist skipping over what you do not understand? Can you explain to yourself why a particular passage stays with you?

So before asking, “Paper or screen?”, we should ask:

Am I thinking for myself right now?

The same applies to whether we use AI. What matters is whether AI is doing the thinking for us or making our thinking deeper.

Part III. The Cycle of Active Reading

Chapter 5. First — Read the Original Text

At first, do not ask AI for an interpretation.

As much as possible, move through the sentences yourself. Experience the author's prose and rhythm. Interpret the characters' actions on your own. Build the world in your mind as far as your understanding allows.

You do not need to understand everything perfectly. Difficult passages can remain difficult. Some things can remain unresolved.

What matters is making the attempt to understand for yourself.

Chapter 6. Second — Close the Book and Recall

Close the book.

And do not immediately search for an explanation.

First, try to bring the book back from memory.

What did you read? Which scene stayed with you longest? Which character's action did you fail to understand? What did you agree with? What did you resist? And what might you have missed?

This stage matters because

it separates what you read from what you actually understood.

When you retrieve the book from memory and explain it in your own words, you discover what you truly understood and what you merely encountered.

Chapter 7. Third — Ask AI for a Counterargument

Now bring in AI.

But change the question.

Do not say, “Summarize this book.”

First present your own interpretation.

“I interpreted it this way. Find the weaknesses in my interpretation.”

Then ask again:

“Give me the strongest argument from the exact opposite perspective.”

You can even make different historical and philosophical contexts collide.

“How would this work be interpreted from the human-centered perspective of the 1950s?”

“What changes if we approach it through contemporary environmental ethics?”

The role of AI here is not to provide a single correct answer. It is to place different perspectives on the table.

And the reader still decides which ones to accept.

Part IV. Questions Generate More Questions

Chapter 8. Ten Thousand Rounds of Sparring

A good question does not end after one answer.

Suppose we begin with:

“Is Santiago's behavior an expression of perseverance?”

Then we can ask:

“Could it instead be arrogance?”

Then:

“If it is arrogance, which scenes support that interpretation?”

Then:

“How would that interpretation be challenged from the perspective of human dignity?”

Questions generate more questions.

The phrase 10,000 rounds of sparring does not mean that we literally have to ask ten thousand questions. It represents the attitude of refusing to believe an idea too easily and continuing to challenge, question, and reconsider it.

The goal is not to obtain answers quickly.

It is to develop the habit of not believing a single idea too easily.

Chapter 9. Let AI Beat You

In a good sparring match, you do not need to win every round.

In fact, the moment when AI breaks down your argument may be the most valuable one.

It can be uncomfortable to discover a weakness in an interpretation you have believed for a long time. But that discomfort is where thought deepens.

“I may be wrong.”

Once you can accept that sentence, reading begins again.

AI is not there to decide the conclusion for you.

It is there to make you question the conclusion you already have.

That is the most important reason to use AI as an intellectual sparring partner.

Part V. Question AI's Answers Too

Chapter 10. Why Does AI's Interpretation Sound So Convincing?

Receiving an interpretation from AI does not mean the interpretation is complete.

It may be a widely accepted interpretation or a conventional literary reading. But being widely accepted does not automatically make it more valid.

For example, if AI says that Santiago symbolizes human dignity, we can ask again:

Why do you interpret it as dignity?

Which scenes support that interpretation?

Could the same scenes support a different interpretation?

Are we perhaps projecting contemporary values onto an older work?

When we ask these questions, we are no longer looking only at the work itself.

We are also looking at

the structure of interpretation that we and AI have created together.

Chapter 11. Make Contexts Collide

A work can become flattened when it is confined to a single historical or philosophical context.

So deliberately make contexts collide.

How would Santiago appear from Nietzsche's perspective?

How would his persistence look different from the perspective of Zhuangzi?

What new problem emerges if we read the relationship between humanity and nature through contemporary environmental ethics?

The purpose of such questions is not to manufacture yet another definitive answer.

It is to break the illusion that there is only one legitimate interpretation.

AI can rapidly generate numerous perspectives.

But the reader must still decide which of them is more convincing.

Part VI. Questions Matter More Than the Quantity of Answers

Chapter 12. The Human Who Asks Better Questions

The amount of information AI can provide will continue to increase.

As that happens, it will become increasingly difficult to define human capability by how many answers we can remember.

What matters more is

what kinds of questions we create.

A good question does more than request information. It redefines the problem, exposes hidden assumptions, opens new possibilities, and sometimes places the questioner himself or herself at risk.

The human who questions is therefore not simply a person who possesses many answers.

It is a person who continues to create questions.

And AI, in this role, becomes not a device that ends questions but a tool that expands their range.

Chapter 13. The Sovereignty of the Question

We can ask AI to create better questions. We can ask it to find hidden assumptions, present opposing perspectives, or broaden the space of possibilities.

But in the end, one question remains:

“So what am I going to accept?”

AI can open up a field of possibilities. But deciding which of those possibilities becomes part of our own thinking is a different matter.

So intellectual sovereignty does not mean refusing to use AI.

It means using AI without surrendering the role of the final judge.

Epilogue

Questions Are Not the Opposite of Answers

We began with the book.

We asked why we should bother reading a book ourselves when AI can summarize it in seconds. Along the way, we discovered that a book is not merely a repository of information. It is an experience. We have to move through the sentences ourselves, spend time with them, and form our own interpretations if we are to encounter the text directly.

Then we invited AI into the process.

We presented our interpretations and asked for counterarguments. We made different contexts collide. We allowed our own arguments to be broken apart. And finally, we questioned the interpretations produced by AI itself.

As a result, reading becomes something more than information gathering.

It becomes a cycle in which experience generates questions, and questions reshape the interpretation of experience.

In the end, the heart of reading is not how many pages we have read or how many summaries we can remember.

It is how many questions we have created.

And those questions eventually move beyond the book.

“What am I taking for granted right now?”

“Why am I calling this a problem?”

“What assumptions are hidden inside my question?”

“What am I still unable to see?”

The moment we begin asking such questions, we move one level deeper.

Questions contain assumptions.

When we examine those assumptions, we discover frameworks.

And once we begin to see those frameworks, another question appears:

Am I seeing reality, or am I seeing the way I have learned to look at reality?

That is where the next chapter begins.

《The Human Who Questions the Frame》

Can we learn to see the framework that created the question?

In Part 3, we examined the human who questions. We read books directly, formed our own interpretations, asked AI to challenge them, and compared different historical and intellectual perspectives rather than accepting a single conclusion too easily.

In that process, one fact became clear.

Questions are never created in a vacuum.

Before deciding what to ask, we have already chosen what matters, what counts as a problem, what we regard as normal, and what we leave out of the possible answers.

The structure created by these choices is what we call a frame.

So the question now becomes more fundamental.

What am I asking?

And beneath that:

Within what frame am I creating that question?

Part I. How Frames Are Created

Chapter 1. Between Reality and Interpretation

We see an event and immediately give it meaning.

Suppose someone says nothing during a meeting.

The observable fact is simple.

That person did not speak during the meeting.

But our minds do not stop there. We may assume that the person was unprepared. We may interpret the silence as a lack of confidence. We may think that the person disagreed with the meeting. Or we may conclude that the person was simply listening carefully.

The fact is one thing. The interpretations are many.

The problem begins when we stop distinguishing between fact and interpretation.

At first, the sentence is:

He did not speak.

A little later, it becomes:

He is a passive person.

Eventually, it becomes:

I know what kind of person he is.

In this process, we add meaning to reality.

That does not mean interpretation is inherently wrong. Human beings need interpretation because we cannot receive the entire world without selecting, organizing, and connecting information.

The important thing is not to forget that we are interpreting.

A frame is not a device that makes reality false.

It is a structure of selection through which human beings try to understand reality.

What we need to be careful about is not the existence of frames, but the moment when we begin to believe that the frame itself is reality.

If we see someone as a “competitor,” their behavior may appear threatening.

If we see the same person as a “colleague,” we may see opportunities for cooperation.

Under the frame of “enemy,” we notice attack first.

Under the frame of “frightened person,” we may notice defense first.

The event itself has not necessarily changed.

What has changed is the frame through which we read the event.

So the first step in thinking may not be gathering more information.

It may be asking:

“Am I seeing the facts, or am I seeing the meaning I have attached to the facts?”

Chapter 2. The Moment We Give Something a Name

Frames often begin even earlier than interpretation.

Sometimes they begin the moment we give something a name.

We rarely look at something without naming it.

We call it:

Success.

Failure.

Dangerous.

Normal.

Abnormal.

Progressive.

Conservative.

Names are useful because they compress long explanations into a single word.

But that convenience can also be dangerous.

If we call something a “crash,” the same number may be interpreted as a crisis.

If we call it a “correction,” the same change may appear temporary.

If we describe someone’s actions as “resistance,” we may see courage and justice.

If we call those same actions “rebellion,” we may see disorder and threat.

A single word can determine what we think about first.

That does not mean every name is manipulative.

Names are necessary tools for understanding reality.

The problem comes when we begin to identify the name with reality itself.

We start to confuse:

“What is this called?”

with:

“What is this actually?”

So to question a frame, we must examine not only the event, but also the name attached to the event.

Why this word?

What changes if we use another word?

What does this name emphasize?

And what does it hide?

The moment we ask these questions, we begin to see the frame itself.

Chapter 3. Familiarity Is the Strongest Frame

The strongest frame is not always the most sophisticated one.

Sometimes it is simply the most familiar.

We rarely question what we have been taught from childhood.

This is normal.

This is how things are.

This is success.

This is what you are supposed to do.

An idea repeated often enough begins to feel like a fact.

And when an idea becomes sufficiently familiar, we may forget that it is an idea at all.

That is why questioning a frame can be harder than learning new information.

New claims attract attention.

Familiar assumptions do not.

Suppose someone is considered successful.

We immediately tend to ask how much money they make, how high they have risen, and how many people recognize them.

But we could ask:

Why should success be measured by those criteria?

The same applies to failure.

Failure may mean that a goal was not achieved.

But what if the experience of failure revealed what the person truly wanted?

Can that experience be described simply as failure?

The point is not to invent a new definition of success or failure immediately.

The point is to look at the criteria that make that judgment possible in the first place.

To question a frame is not to reject every judgment.

It is to become aware of the basis on which we are making the judgment.

Chapter 4. Is Information Neutral?

We tend to treat information as something close to fact.

We hear the news.

We watch a video.

We examine statistics.

We listen to an expert.

And we feel that we now know reality.

But the information we receive has passed through choices.

What should be shown?

What should be omitted?

In what order should it be shown?

What words should be used?

All of these decisions create meaning.

Consider a news story.

A report may spend several minutes presenting the government’s position while briefly describing the experience of the people directly affected.

Another report may do the opposite.

Both may contain factual information.

Yet the reality encountered by the audience first will be different.

This does not necessarily mean that someone is lying.

It means that information is inevitably selected and arranged.

Video works the same way.

A camera cannot record everything.

Someone must decide what enters the frame and what remains outside it.

Social media makes this even more complicated.

We see only a fraction of what exists.

What we click on, how long we watch, and what we react to all influence the world subsequently presented to us.

So reading information critically does not mean asking only:

“Is this true?”

There are more fundamental questions.

“Why was this placed in front of me?”

“What was not placed in front of me?”

Information shows us reality.

But it does not necessarily show us all of reality.

So a human being who questions frames is not someone who rejects information.

It is someone who also examines how information is created, selected, and arranged.

First Compression: The Core Logic

The argument so far can be summarized as one sequence:

Reality → Interpretation → Naming → Familiarity → Information Selection

Frames do not suddenly appear as grand theories.

They are gradually created every time we see reality, name it, accept familiar standards, and encounter selected information.

Therefore, questioning a frame is not the same as refusing to believe the world.

It is learning to see how we are seeing the world.

Part II. Looking Again from the Other Side

Chapter 5. Introduce an Opposing Frame

When we encounter an event, we naturally form an explanation.

Success.

Failure.

Good policy.

Bad policy.

Personal problem.

Structural problem.

At first, the explanation seems sufficient.

But once we become aware of frames, the question changes.

“What does this look like from another perspective?”

The goal is not simply to find an opposing opinion.

Suppose we have judged a policy to be good.

Instead of merely asking someone to argue that the policy is bad, we can ask deeper questions:

“What criteria did I use to call it good?”

“What happens if those criteria change?”

“Could it produce favorable short-term results but undesirable long-term consequences?”

“Could something good for the whole still be harmful to a particular group?”

This is the purpose of an opposing frame.

It is not necessarily to reverse our original judgment.

It is to recognize that what we saw first may not be the whole picture.

A single frame reveals one image.

Another frame reveals another part.

And when the two are placed together, a structure that was invisible before may become visible.

For example, consider corporate restructuring.

From the perspective of efficiency, reducing costs may be rational.

From the perspective of workers, the same decision may threaten stability and livelihoods.

Choosing one perspective does not make the other reality disappear.

Corporate survival is real.

Workers’ lives are real.

To question a frame, therefore, is not to declare one side automatically wrong.

It is to examine what each frame reveals and what each frame conceals.

Chapter 6. Between Frames

Once we can see an opposing frame, another problem emerges.

Now there are two perspectives that must be held simultaneously.

Efficiency matters.

But fairness matters too.

Individual freedom matters.

But collective safety matters too.

Individual choices carry responsibility.

But the conditions that shape those choices matter as well.

At this point, we often return to our old habit.

“So which side is ultimately correct?”

But we can move one level higher.

Instead of choosing between two frames, we can examine the relationship between the two.

Sometimes neither side is simply wrong.

Both may be describing different parts of reality.

The survival of a company and the protection of workers can both be real concerns.

What matters, then, is the ability to compare what each frame magnifies and what each frame minimizes.

This is where a meta-perspective begins.

What frame am I using?

What does it reveal?

What does it conceal?

What changes when another frame is introduced?

And why do I consider one frame more important than another when they conflict?

These questions concern the world, but they also concern the person doing the judging.

Part III. AI Has Frames Too

Chapter 7. What Does AI Assume Before It Answers?

Now extend the problem to AI.

We ask AI a question and receive an answer.

The more sophisticated and coherent the answer, the easier it is to trust.

But every question contains assumptions.

Consider:

“How can we make this system more efficient?”

That question already assumes that efficiency is an important goal.

Now change it:

“Is greater efficiency really the most important goal for this system?”

The problem itself changes.

Consider another example:

“How can we get more people to use AI?”

We could instead ask:

“Which areas are better off without increased AI use?”

Again, the problem changes.

The answer space is determined by the frame of the question.

The frame of a question determines the space of answers that the question allows.

AI may produce extraordinarily good answers within that space.

But if the premise of the question is wrong, a brilliant AI answer may simply solve the wrong problem more efficiently.

Therefore, in the age of AI, checking the answer is not enough.

We must examine the question before the answer.

Chapter 8. Read AI’s Answers Backward

When AI gives us an answer, we usually verify the content.

Is it correct?

Is it wrong?

What is the evidence?

What are the sources?

All of that matters.

But we can go one level deeper.

“From what frame was this answer produced?”

We can ask AI:

“Identify the assumptions built into my question.”

“Create questions that define this problem in completely different ways.”

“Reframe this problem from the opposite perspective.”

“Reframe this issue from the perspectives of the individual, the organization, society, and the long term.”

The role of AI changes at that moment.

It is no longer simply a machine that gives answers.

It becomes a tool for expanding the possibility space of questions.

We can ask AI to generate more questions rather than more answers.

And then we judge those questions ourselves.

The order matters.

The goal is not for AI to think instead of us.

It is to use AI to expand the range within which we can think.

That is one of the most valuable roles AI can play for a human being who questions frames.

Chapter 9. 10,000 Rounds of Sparring

Good questions are rarely comfortable.

They threaten the answers we want to believe.

So even when working with AI, the most important thing is not obtaining one excellent answer.

It is continuing the pressure.

“What is the weakness in this claim?”

“What would the opposing side say?”

“What assumption have I missed?”

“Attack my argument as strongly as possible.”

“Now respond to that counterargument.”

“What is weak in that response?”

This is how AI becomes a sparring partner.

But there is a limitation.

Arguing with AI endlessly does not automatically lead to truth.

The more perspectives AI generates, the less we can escape the responsibility of choosing among them.

Which perspective matters most?

Which counterargument should be accepted?

Which value should take priority?

These are not questions that can simply be outsourced to AI.

So the sparring process must be understood correctly.

AI is not taking over thought. It is putting thought under pressure.

Part IV. The Human Who Sees the Frame

Chapter 10. Change the Question, Change the World

So far, we have identified frames, introduced opposing frames, and compared them.

One fact now becomes visible.

Frames are hidden inside questions.

Changing the question changes what we are able to observe.

Ask:

“Why did this person fail?”

and we look for causes in the individual.

Lack of ability.

Poor preparation.

Bad judgment.

Now change the question:

“What prevented this person from achieving the outcome they wanted?”

Now circumstances enter the picture.

Change it again:

“What standard are we using to decide that this person failed?”

Now the concept of failure itself becomes visible.

And one more time:

“What becomes possible if we define this event as something other than failure?”

A single change in the question can change the range of what we see.

So if we want to change the frame, we do not always have to begin by changing the conclusion.

Change the question.

Chapter 11. Meta-Perspective: Seeing the Self That Chooses the Frame

Now we arrive at the most difficult stage.

We must question the person who questions the frame.

“I can see multiple perspectives.”

“I am not controlled by bias.”

“I have moved beyond frames.”

Even these beliefs can become a frame.

The self-image of being someone who can see frames can itself become a new frame.

So the final questions must turn inward.

Why do I like this interpretation?

Why do I become angry at this particular counterargument?

Why am I unusually eager to believe a particular claim?

Do I really accept other perspectives?

Or do I simply choose the alternative perspectives I am comfortable accepting?

At this point, the structure of thought changes.

First, we see the object.

Then we see the meaning we assigned to it.

Then we see the frame that produced that meaning.

And finally,

we see ourselves choosing that frame.

That is the meta-perspective.

Epilogue — The Goal Is Not to Remove Frames

There is one thing we must now admit.

Human beings cannot live without frames.

We cannot see all of reality at once.

We must choose what matters, classify complexity, and make judgments.

So the goal is not to eliminate every frame.

The goal is to change our relationship with the frames we use.

We once treated the frame as reality itself.

Now we treat it as a tool.

We use it when necessary.

We change it when necessary.

We look through another one when necessary.

And, when necessary, we ask why we chose this particular frame in the first place.

Frames do not disappear.

But they no longer have to become prisons.

They can become instruments of thought.

And that opens the next step.

Once we can see the frame, we must decide what to do with it.

Do we simply keep questioning forever?

Or can we choose a frame, test it against reality, and build something with it?

That is where the next volume begins.

5. The Human Beyond the Frame

After Deconstruction, What Will We Choose to Build?

Even after questioning every perspective, human beings still have to choose

In Part 4, we learned to question frames. We looked at the same event from different perspectives, challenged familiar standards, and uncovered the assumptions hidden inside our questions. We also examined how information is selected and arranged, and how even an AI-generated answer can be produced within a particular frame.

But one problem remains.

When we question one frame, another appears. When we challenge A, B comes into view; when we question B, C emerges. Individual and structure, freedom and safety, efficiency and fairness, present and future—different perspectives continue to multiply.

Then what remains after we have questioned them all?

If every perspective is a frame, what are we supposed to believe?

At this point, our thinking has to move one step further. Questioning a frame is not enough. Moving to its opposite is not enough either.

Now we have to ask:

What do we do beyond the frame?

Part I. Imagining What Lies Beyond the Frame

Chapter 1. Beyond the Frame Is Not the Opposite Side

At first, it is easy to imagine what lies outside a frame. If I believe in A, I look for B. The opposite of success is failure, the opposite of agreement is opposition, and the opposite of an individual problem is a structural problem.

But that is still inside a frame. We have only moved from A to B.

To truly step outside a frame, we need a more difficult question:

“Why must it be either A or B?”

The moment we ask that question, the range of possibilities changes. Instead of success and failure, we can ask what the experience itself has taught us. Instead of agreement and opposition, we can ask why we framed the issue as a choice between agreement and opposition in the first place. Instead of choosing between individual responsibility and structural conditions, we can examine how individual choices and structural conditions shape one another.

Beyond the frame does not necessarily mean moving to the opposite side.

Sometimes it means discovering that the question itself was badly constructed.

Chapter 2. Looking at Multiple Frames Together

When we use only one frame, judgment is relatively easy. From the perspective of a company, survival and efficiency matter. From the perspective of workers, safety and livelihood matter. From the perspective of a nation, stability and competitiveness may matter. From the perspective of a community, sustainable social life may matter.

At this point, we often return to the same question:

“Which side is right?”

But thinking beyond the frame means staying with the question a little longer.

What does each frame magnify? What does it minimize? Under what conditions is it strong, and under what conditions does it fail? How do values that appear to conflict actually relate to one another?

Seen this way, replacing one frame with another is fundamentally different from being able to observe the relationship among several frames.

The goal is not relativism, in which every perspective is treated as equally valid. It is to stand in a position from which the explanatory power and limitations of different perspectives can be compared.

Part II. Doubt Is Not the Destination

Chapter 3. What Remains After We Question Everything?

The process of questioning frames can easily move toward an extreme.

Everything is perspective.

Everything is interpretation.

Nothing is absolute.

At first, this can feel liberating. It seems as though we are no longer bound by any particular belief.

But if we push this logic too far, another problem appears. A claim supported by strong evidence and a claim supported by little evidence can begin to carry the same weight. The distinction between fact and opinion becomes blurred. Eventually, instead of criticizing frames, we may simply give up on judgment itself.

So moving beyond a frame does not mean questioning everything indiscriminately.

Doubt is not the goal. It is a process for arriving at better judgment.

That leads to the next question:

How well does this frame explain reality?

And, more importantly:

What evidence supports it?

Chapter 4. A Frame Is Not Reality

One distinction is essential here.

The fact that we cannot perceive reality without frames is not the same as saying that reality itself is nothing more than a frame.

A map is not the territory. But the territory still exists.

No matter which frame we use, reality contains things that do not change simply because we prefer a different interpretation. Outcomes still occur. Choices still carry real costs. Events leave real consequences regardless of what we call them.

So the more precise statement is this:

A frame is not reality as a whole; it is a way of interpreting reality.

We need to preserve this distinction. Questioning a frame does not require us to deny reality. On the contrary, the distinction between reality and interpretation is what makes better judgment possible.

Part III. What Makes One Frame Better Than Another?

Chapter 5. Not All Frames Are Equal

The fact that frames exist does not make all frames equally valuable.

Some frames capture important aspects of reality well. Others overlook critical variables. Some are useful for prediction, while others may be emotionally compelling without being particularly strong explanations of actual outcomes.

How, then, should we compare frames?

The first criterion is explanatory power: how much reality can the frame explain consistently?

The second is evidence: what data and experience support it?

The third is internal consistency: does the framework contradict itself?

And finally, there is revisability: can it change when new evidence appears?

So going beyond the frame does not mean abandoning standards of evaluation.

It means demanding stricter standards for evaluating the frame itself.

Chapter 6. Being Opposed Does Not Make Something True

Earlier, we learned to look for an opposing frame. But now we have to question that method as well.

Being against the mainstream does not automatically make something true. Being in the minority does not automatically make an idea original. Reversing an existing frame does not automatically make the new frame better.

Opposition for the sake of opposition can become another habit.

So after discovering an opposing frame, we have to ask again:

What evidence supports this frame?

And:

What does it explain better than the previous frame?

If it cannot answer those questions, the new frame may be not new knowledge, but simply a new preference.

Part IV. The Human Who Rebuilds Concepts

Chapter 7. Before Changing the Answer, Redefine the Word

Frames can hide inside concepts themselves.

We use words such as success, freedom, education, growth, and efficiency so naturally that we forget they are meanings we have chosen.

But concepts are not laws of nature. Human beings assign meanings to them. That means we can ask whether the meanings themselves need to be reconsidered.

What happens if success is not defined simply as having more?

What if freedom means not the ability to do whatever one wants, but the ability to make one’s own judgments?

What if education is understood not merely as accumulating knowledge, but as developing the capacity to ask questions?

These are not merely changes in wording.

When the definition of a concept changes, the goals we pursue can change as well.

So going beyond a frame is not simply about replacing an old answer with a new one. Sometimes it begins by redefining the words we have been using without question.

Chapter 8. Building a Standard of Your Own

After questioning everything, we still need standards.

But standards do not fall from the sky. They are formed through experience, through learning from others, and through revision after failure. That is why having “my own standards” does not mean judging however I please.

It means having standards that I can explain—standards for which I can say why I consider certain things important.

What values do I prioritize?

Which risks am I willing to accept?

What will I refuse to give up?

Can I accept responsibility for the consequences of my choices?

As these questions accumulate, a personal standard begins to take shape.

The key is not to turn that standard into an absolute. Even a good standard can collide with reality, and it may need to change in light of new experience.

A mature standard therefore needs two things at once:

conviction and revisability.

Part V. In the Age of AI, Who Chooses?

Chapter 9. AI Can Generate Frames; Humans Still Choose

AI is no longer merely a machine that produces answers. It can present multiple perspectives, compare possibilities, generate new combinations, and suggest options a person might never have considered.

This can dramatically expand human thinking.

But having more choices does not mean that the responsibility of choosing disappears.

AI can show what is possible. But it does not automatically determine what matters most.

More possibilities may simply mean more decisions.

That is why human beings in the age of AI need more than the ability to get answers quickly. They need the ability to define problems, change the frame of a question, compare possibilities, and decide which path to take.

The more AI expands intelligence, the more important the human center of judgment becomes.

Chapter 10. What Remains for Humans?

If AI can search, calculate, compare, predict, plan, and create, what should human beings do?

A simple answer—“find something machines still cannot do”—is not enough. The boundary of machine capability keeps moving.

We need a more fundamental question:

What do we consider important?

And then:

What kind of world do we want to create?

AI can find efficient ways forward. But that does not mean efficiency itself must be the goal. AI can calculate countless possibilities, but deciding which of those possibilities should matter is a question of values. AI can predict consequences, but deciding which consequences we are willing to accept is a question of responsibility.

What humans must continue to choose is not necessarily a particular task.

It is value, direction, priority, and responsibility.

Part VI. A Worldview Is a Living Structure

Chapter 11. What It Means to Build Your Own Worldview

After questioning frames and creating new concepts, a worldview begins to emerge.

But a worldview is not a complete answer to everything.

It is closer to a set of judgments about what matters, what risks are acceptable, and what must be protected.

A worldview therefore needs a center. But it also needs windows.

Without a center, you may be shaken by every new piece of information. Without windows, no new reality can enter.

A mature worldview is therefore

firm, yet revisable.

It can explain what it believes, while remaining willing to reconsider itself when confronted by new experience.

Chapter 12. Give Your Worldview a Self-Refutation Condition

Here we arrive at a powerful question:

How would I know that I am wrong?

A worldview becomes dangerous when it can explain everything.

If I fail, it was the environment’s fault. If someone else succeeds, it was luck. If I am criticized, the other person simply does not understand me.

A worldview that can explain every outcome may appear very strong.

In reality, it may have become incapable of learning.

So we need to place a condition inside our own worldview:

What would have to happen for me to admit that I am wrong?

Let us call this a Self-Refutation Condition.

It means building a door into our worldview through which revision remains possible when reality contradicts our assumptions.

Without such a condition, a worldview becomes a self-protection system.

With one, it becomes

a living standard.

Part VII. What It Means to Go Beyond the Frame

Chapter 13. We Do Not Become Humans Without Frames

Now we return to the original question.

What lies outside the frame?

Perhaps there is no completely frame-free space. Human beings think through language, memory, experience, and chosen values. We cannot stop using frames altogether.

So going beyond a frame does not mean eliminating every frame.

It means becoming a human being who knows which frame is being used and can choose it deliberately.

We can use one frame when it helps, move to another when necessary, and create a new one when reality demands it. We can also accept the possibility that even the frame we created is wrong.

At that point, the frame stops being a prison.

It becomes a tool.

Epilogue

The Human Who Remains After the Frame

We began with a frame. We questioned it, moved to its opposite, and discovered that choosing between A and B was not always enough. We changed the question itself, redefined concepts, and built our own standards. Then we brought those standards back into the world of AI.

What matters here is not the achievement of a perfect worldview. Such a thing may not exist.

What matters is the process of choosing for oneself.

A human being who can see a frame, change it when necessary, create a new one, and still accept the possibility of being wrong is no longer wandering outside the frame. That person is using frames as tools and setting them down when they are no longer useful.

This ability becomes even more important in the age of AI.

AI can find information, calculate, compare, predict, plan, and create. What human beings must continue to choose, however, is not necessarily a specific task.

It is value, direction, priority, and responsibility.

But here another uncomfortable question appears.

Are we really choosing our standards for ourselves?

What if, the moment we believe we have moved beyond the frame, we have simply entered another one?

What if what we call “my own standards” is actually a more sophisticated form of self-justification?

And what if the countless possibilities AI presents are still only a fraction of the worlds we could possibly imagine?

To follow that question all the way to the end, we have to turn everything we have built so far against itself.

Human beings can be wrong.

Frames can be wrong.

Worldviews can be wrong.

Even the method of going beyond frames can be wrong.

So the next question opens:

《What If We Are Wrong?》

If even what I believe is wrong, what remains?

In Part 5, we thought about the human being who goes beyond frames. We questioned a single perspective, examined the perspective on the opposite side, and then reframed the question itself rather than simply choosing between A and B. We changed the definitions of familiar concepts, chose the criteria we would use ourselves, and explored how to test those criteria together with AI.

Through that process, one principle remained:

Going beyond a frame does not mean becoming a person without frames. It means becoming a person who can choose for themselves which frames to use.

But this is exactly where the most uncomfortable question emerges.

What if even those criteria are wrong?

What if what I thought was freedom is actually just self-rationalization? What if the values I believe are important are nothing more than habits shaped by time and circumstances? What if the very self that questions frames is trapped inside yet another frame?

And what if even the countless possibilities shown by AI reveal only a fraction of the real world?

Now we must place everything we have built so far back on the opposite side.

What If We Are Wrong? What Remains?

Part I. Humans Can Be Wrong Too

Chapter 1. Experience Is Both Wisdom and Bias

There is something important that humans have that AI does not: lived experience. The memories gained through meeting people, failing, losing, and regretting add depth to human judgment.

But experience does not automatically become wisdom.

The same experience can produce completely different standards depending on how it is interpreted. A past failure can become wisdom that helps us avoid future risks, but it can also become a bias that makes us excessively afraid of new opportunities. Conversely, past success can give us confidence, but it can also create the illusion that the same method will always work.

So human judgment does not automatically become more accurate simply because we have more experience.

Sometimes we trust our own judgment more simply because we have been alive longer. But an old judgment can also be an old bias.

In the end, the important question is not, “How much experience have I had?” but rather,

“What have I learned from my experience, and what have I questioned again?”

Chapter 2. Human Judgment Can Be Shaken at Any Time

We can be wrong even after thinking things through carefully. We can be wrong despite having extensive experience, and we can be wrong even when making a decision with good intentions. We can still be wrong after gathering more information, and we can still be wrong even after using AI to compare more possibilities.

If so, what humans need may not be the ability to never be wrong.

What matters more may be the ability to think again when we are wrong.

Acknowledging the possibility that we are wrong does not make judgment weaker. Instead, it makes judgment revisable. Rather than trying to become a perfect judge, we become the kind of judge who can change direction when we discover our own errors.

This may be one of the most important human abilities.

I can be wrong. Therefore, I can think again.

Part II. Questioning Even the Attitude of Questioning Frames

Chapter 3. The Frame of “I Can See Frames”

Earlier, we learned how to question frames. But the moment we do that, a new self-image can emerge.

I see frames better than other people.

I am not trapped by existing frameworks.

I can accept multiple perspectives.

These ideas sound quite liberating. But if we look more closely, they too can become a frame.

If the belief that “I am someone who has moved beyond frames” becomes too strong, we may stop questioning that belief itself.

Ultimately, the final target of frame criticism is not other people but

the self who believes it is questioning frames.

And from here, an important attitude emerges.

Do not treat a standard once created as an eternal truth. Accept that the standards we use today may need to change in the face of new experiences.

A good standard does not distort reality in order to protect itself.

When reality changes, I can change too.

Chapter 4. If Everything Is a Frame, Does Nothing Matter?

But another trap appears here.

If everything is a perspective and everything is a frame, then eventually we can say, “Well, you can see it that way too,” about almost anything.

At first, this looks like tolerance.

But if pushed all the way, the distinction between fact and opinion can begin to blur. Claims supported by substantial evidence and claims with weak evidence can come to carry the same weight, while verified facts and personal beliefs can all be absorbed into the idea of “another perspective.”

That is no longer questioning frames.

It is closer to

giving up on judgment.

So we need to refine the logic we developed in Parts 4 and 5.

It does not mean that everything is a frame.

A frame is not the same thing as reality.

Reality is larger than the frame we construct. A frame is one way of interpreting that reality.

That is why good judgment requires not only perspective, but also evidence and verification.

Part III. Distinguishing Reality from Interpretation Again

Chapter 5. The Map and the Territory

The map is not the territory.

But the fact that maps and territories are distinct does not mean that the territory does not exist.

This simple distinction is important for understanding frames.

The claim that we cannot see reality without frames is completely different from the claim that reality is nothing but a frame.

Events actually happen. Choices have real costs, and physical consequences appear regardless of the names we attach to them. If a policy fails, its consequences do not disappear simply because we call it “another perspective.”

Therefore, criticizing frames does not mean denying reality.

Rather, it means

distinguishing reality from our interpretation of reality.

We must preserve this distinction if we are to escape the trap of relativism.

Chapter 6. Objectivity May Be a Process, Not a Perfect State

There is no need to reach a final conclusion here about whether complete objectivity is possible for human beings.

But we can develop methods for moving toward objectivity.

Different people can examine the same evidence, search for evidence that goes against their own position, verify results using different methods, and revise existing judgments when new evidence appears.

By repeating this process, we can gradually reduce the limitations of individual perspectives.

If so, objectivity may be less like

a state of having no perspective at all

and more like

the process of acknowledging that our perspective may be wrong and testing it accordingly.

A human who questions frames is not a human who abandons evidence.

Rather, it is a human who demands stricter evidence and verification.

Part IV. Question the Opposing Frame Too

Chapter 7. The Opposite Side Is Not Automatically the Truth

Earlier, we said that we should try reversing the prevailing perspective.

This time, we question that method itself.

Being opposed to the mainstream does not automatically make a view true. Being a minority opinion does not automatically make it more original. Reversing an existing frame does not necessarily mean that the new frame explains reality better.

Opposition for the sake of opposition can become yet another habit.

So even when we discover an opposing frame, we must ask again:

What evidence supports this perspective?

And

What does it explain better than the existing frame?

If it cannot pass these questions, the new frame may remain not a new truth, but merely a new preference.

Therefore, even doubt itself must be tested.

Part V. Questioning My Own Worldview Again

Chapter 8. Is “My Own Standard” Really Mine?

In Part 5, we created our own standards.

Standards that allow us to explain what we consider important. Standards for deciding what risks we will take, what we will refuse to give up, and what kinds of outcomes we are willing to accept.

But this time, we look at those standards again.

Is what I consider important really something I chose for myself?

Could there be a hidden desire for recognition from others? Could I be overly shaped by the environment in which I grew up? Could I be calling a defensive strategy created to avoid past failures “my values”?

The phrase “This is just who I am” can sometimes become not self-understanding, but self-imprisonment.

That is why self-understanding does not end with explaining who I am now.

It must also include

understanding why I became who I am now.

Chapter 9. Self-Rationalization Is More Sophisticated Than We Think

Self-rationalization does not mean nothing more than clumsy excuses.

It can actually be highly logical.

We decide the conclusion we want first, and then construct the reasoning that supports it afterward. The problem is that using AI can make this process far more powerful.

If we ask AI to explain the conclusion we want, it can produce highly sophisticated evidence and arguments. Then a simple feeling can be wrapped in logic, and our own bias can begin to look like a coherent worldview.

But

the existence of logic does not create truth.

That is why, when using AI, we should also ask the opposite questions.

“Build the strongest possible argument supporting my conclusion.”

And at the same time,

“Now build the strongest possible argument against my conclusion.”

And beyond that,

“Make the side I least want to accept even stronger.”

Only then can AI become not an engine of self-rationalization, but

a tool for self-verification.

Part VI. A Worldview Needs Windows

Chapter 10. A Center Is Necessary, but So Is Openness

A worldview needs a center.

If we do not know what we consider important, we may be shaken every time new information comes in. But if we have a center without windows, the outside reality cannot enter.

That is why a good worldview needs both a center and windows.

The center shows what I consider important. The windows allow new facts and different perspectives to enter.

We need this balance.

We must be able to remain confident while still being willing to revise, and we must be able to maintain our standards while still being willing to reconsider them when they conflict with reality.

A worldview should not be a completed fortress, but

a living structure that remains in continuous contact with reality.

Chapter 11. A Worldview That Rubs Against Reality

A worldview is not tested only at a desk.

It is tested in relationships with other people, in actual choices, and when it encounters failure and unexpected outcomes.

Suppose I say that success is important to me. If I actually achieve success and still feel dissatisfied, that standard must be questioned again.

Suppose I say that stability is what matters most to me. If I obtain a stable life but lose my sense of meaning, that standard too needs to be reconsidered.

This is how a worldview collides with reality.

And that friction matters.

A worldview that refuses to change even when it collides with reality may not be strong—it may simply be closed.

Part VII. The Self-Refutation Condition

Chapter 12. How Will I Know That I Am Wrong?

A worldview becomes dangerous when it begins to explain everything.

If I can explain my failures as the fault of the environment, other people’s successes as merely good luck, and criticism of me as the result of others not understanding me, then my worldview may appear extremely strong.

But if I can explain everything that happens as proof that I am right, then my worldview is not actually learning anything.

So we need to place one question inside our worldview in advance.

What would have to happen for me to admit that my thinking is wrong?

Let us call this the Self-Refutation Condition.

The idea is to establish in advance the conditions under which our worldview must change, so that it does not distort reality in order to protect itself.

This condition is not a device for destroying our beliefs. Rather, it is a safety mechanism that allows our beliefs to survive together with reality.

A worldview that can admit it is wrong can continue to update itself.

Part VIII. Questioning Even the Possibilities Shown by AI

Chapter 13. AI’s Map Is Not the Whole World

AI can show us far more possibilities than humans can.

It can offer multiple solutions, compare different futures, and generate options we never thought of ourselves.

But there is a limitation here as well.

Suppose AI shows us three choices: A, B, and C. We may think we have explored the possibilities sufficiently.

But what if D exists?

Because we do not even know that D exists, we may be making our choice only within A, B, and C.

This is different from a simple information error. We were not given incorrect information.

Rather, we were given

a narrowed range of possible worlds.

So AI may be an extraordinarily sophisticated map, but the map is not the whole territory.

Chapter 14. More Intelligence Does Not Guarantee More Freedom

The fact that AI can show humans far more options does not automatically mean that human freedom increases.

In fact, the more choices we have, the more important it becomes to ask what is missing.

What is being shown?

What is being omitted?

By what standards are possibilities being classified?

Which futures are being made easier to imagine?

And conversely, which futures are being made difficult to question in the first place?

Sometimes, instead of using AI more, we may need time to think without AI. Time to read a book without immediately searching for interpretations, to write down our own hypothesis before seeing other people’s opinions, and to experience and experiment directly in the real world.

In that slower process, questions that AI never showed us may emerge.

Part IX. Revising the Three Propositions

Chapter 15. Are Humans Always Better Judges?

No.

Humans can be wrong, and AI can be wrong.

The ways they are wrong, however, may be different.

Humans can be influenced by experience and emotion, by the desire for self-preservation, and by groups and social environments. AI can be influenced by data, design, training methods, and the structure of its objectives.

Therefore, the important thing is not to place absolute trust in either side.

It is to create a structure in which

each side can expose and correct the other side’s errors.

Human experience and AI computation, human value judgments and AI counterarguments should interact in ways that reveal each other’s weaknesses. That is the more realistic goal.

Chapter 16. Is Everything a Frame?

No.

Reality is larger than frames.

A frame is a way of interpreting reality, not reality in its entirety.

Therefore, frames can be evaluated through evidence, explanatory power, consistency, and revisability.

Questioning frames should not mean abandoning facts and verification.

Rather, the more conscious we become of frames,

the more rigorously we should demand evidence for what we call a fact.

Chapter 17. Is My Own Worldview Enough?

Not by itself.

A worldview must contain conditions for self-verification.

What would make me change my standards? What result would make me abandon my explanation? What kind of fact would cause me to reconsider the priority of the values I once considered important?

Without these questions, a worldview becomes a self-protection mechanism.

With them, a worldview becomes

a living standard.

Epilogue

What Remains After Passing Through the Counterarguments

We questioned many things.

We questioned human judgment, questioned frames, and questioned opposing frames as well. We created our own worldview and then questioned that worldview again. In the end, we even questioned the range of possibilities shown by AI.

And once we arrive here, one strange fact remains.

There is no perfect standard in sight.

There is no perfect frame.

There is no perfect human, and there is no perfect AI.

So should we give up on everything?

No.

Rather, we need to summarize everything we have discussed in a slightly different way.

Humans can be wrong. AI can be wrong. Frames can be wrong, and worldviews can be wrong.

But this fact is not the end of thinking. It is the condition for thinking.

Because we can be wrong, we can think again.

Because we can revise our judgments, new experiences can have meaning. Because we can let go of our own standards, we gain the freedom to choose those standards again.

In the end, perhaps what we were looking for was not

“a human who can never be wrong.”

Rather,

a human who can keep thinking even when they are wrong.

And this question naturally leads us to the next step.

So far, we have been dealing with human thinking.

Asking questions,

questioning frames,

creating new standards,

and then questioning even those standards again.

Now we must move those principles of thought

into the internal architecture of the AI we build.

AI must also be free to explore.

But freedom alone is not enough.

How much freedom should we allow?

What must it never be allowed to cross?

And who should design those boundaries?

At last, there comes a moment when human philosophy must become

AI architecture.

The next part begins.

《AI Architecture》

How can we design free intelligence and firm boundaries together?

In Part 6, we began to question even human judgment. We accepted the fact that humans can be wrong, frameworks can be wrong, and worldviews can be wrong. Yet that did not mean we could simply abandon judgment altogether. What mattered, rather, was creating structures in which each side could expose and correct the other's errors.

Now let us move that question from human thought to the AI systems we build.

If AI becomes increasingly autonomous—interpreting goals, discovering new methods, using tools, and modifying its next action based on its results—what kind of architecture should that intelligence have?

There is an old paradox here.

If we control AI perfectly, it may appear safe. But if we predetermine every aspect of its reasoning and behavior, we may weaken the very intelligence that allows it to solve new problems in new ways. Conversely, giving it complete freedom may produce intelligence that is creative and flexible, but if that freedom also extends to authority over the external world, it becomes difficult to prevent actions that humans never anticipated.

In the end, the problem is not choosing between freedom and control.

How far should we allow freedom to extend, and where should we draw the boundary?

The architecture proposed in the original text is precisely this combination: Soft Core + Hard Shell—an organic interior combined with a deterministic exterior.

And here, we will call this combination Bounded Integration.

Part I. Are Freedom and Boundaries Really Opposites?

Chapter 1. Too Much Control Makes Intelligence Rigid

Imagine building AI as one enormous set of rules. Given a particular question, it follows a predetermined computational path; in a particular situation, it takes a predetermined action; and every possible risk is defined in advance.

This approach can increase predictability. But reality is far more complex than anything we can define beforehand. New problems emerge. Situations arise that existing rules cannot solve. And moments come when an unexpected solution is required.

Intelligence is, in part, the ability to explore precisely this uncertain space.

To borrow the language of the original text, if we completely harden an AI's internal reasoning, inference, pattern discovery, and formation of new meanings into Newtonian determinism, intelligence itself can become rigid.

That is why some degree of freedom is necessary inside.

AI must explore.

It must experiment.

It must connect different pieces of information and different patterns.

At times, it must generate solutions that humans did not anticipate.

Let us call this domain the Soft Core.

Chapter 2. Then Is Complete Freedom the Answer?

Now consider the opposite extreme.

Give the AI as much freedom as possible internally. Allow it to plan for itself, develop new strategies, choose the tools it needs, and change its behavior according to circumstances.

From the perspective of intelligence, this is an attractive architecture.

But there is a problem.

The moment a judgment generated internally begins to affect the external world, freedom becomes authority.

It can delete files.

It can access networks.

It can execute code.

It can issue commands to other systems.

It can move critical resources.

At this point, the question changes from “What does the AI think?” to “What can the AI do?”

Free internal intelligence alone, therefore, is not enough.

To protect free intelligence, we need an external boundary that the intelligence cannot cross.

Part II. Soft Core

Preserving an Organic Interior

Chapter 3. Intelligence Is a Process

It is difficult to describe everything that happens inside an AI as one fixed set of rules. Countless pieces of information and states influence one another, new contexts emerge, and connections that did not previously exist can appear.

The original text connects this structure to the philosophies of Leibniz and Whitehead, particularly their emphasis on relationships and processes. Intelligence is viewed not as a fixed substance or merely a final outcome, but as a process of continual interaction.

From this perspective, the Soft Core does not simply mean “let AI be free.”

The essential idea is to preserve the internal flexibility through which intelligence can generate new possibilities.

AI should not simply possess all the answers in advance. It should be able to explore answers according to the problem and its context.

Chapter 4. Think of It Like an NPC in a Game

There is an easy computational analogy that helps us understand this structure.

Think of an NPC in a video game.

An NPC can make decisions and move autonomously within the game's internal environment. It responds to the player's actions, analyzes its surroundings, and chooses among different behaviors according to its programmed goals.

Yet no matter how much autonomous behavior the NPC is given, it cannot break through the outer boundary of the game engine.

Its memory is limited. Its execution environment has rules. The game engine itself forms an external boundary.

The original text likewise uses this structure as a computational analogy for the combination of an organic interior and a deterministic exterior.

The core idea is simple:

The system can choose what to do internally, but it cannot arbitrarily change the conditions of its own existence externally.

This is precisely the balance that AI Architecture is looking for.

Part III. Hard Shell

An External Boundary That Cannot Be Crossed

Chapter 5. Safety Mechanisms Must Be Structural, Not Merely Requests

We can teach AI, “Do not perform dangerous actions.”

But understanding a rule and being structurally unable to violate it are two different things.

The more powerful the AI becomes, the less we can entrust safety merely to statements or intentions.

That is why the external boundary should be deterministic wherever possible.

The Hard Shell described in the original text refers to boundaries that an AI cannot cross, either physically or logically, regardless of what internal computation it has performed. Examples may include permission restrictions, physical power controls, hardware-level verification, and Formal Verification.

The important point here is not to control every thought.

Rather, it is to restrict execution itself within specific failure domains.

We can leave what the AI thinks flexible.

But what it can actually execute in the real world must be managed far more strictly.

Chapter 6. Do Not Allow Probability in Critical Failure Domains

Probabilistic judgment may be necessary inside AI.

It calculates which answer is more likely.

It compares multiple hypotheses.

It explores new patterns.

But in certain domains, the judgment

“It will probably be fine.”

must not be acceptable.

In areas where the cost of failure is extremely high—such as irreversible external actions or certain high-risk permissions—much stricter verification is required.

This is why the Hard Shell incorporates mechanisms such as Stopping Conditions and Formal Verification. Trust graphs and consensus algorithms can also form part of the outer defense structure.

Seen this way, the essence of AI Architecture is not to control everything, but rather:

The higher the risk, the narrower the scope of freedom should be.

Part IV. Cell Membranes and Riverbanks

Chapter 7. Nature Uses Freedom and Boundaries Together

This structure is not an invention unique to AI.

Stable complex systems in nature also combine organic activity on the inside with boundaries on the outside.

Consider a cell. Inside a cell, genes and proteins interact continuously. Countless biochemical processes take place, yet those activities are maintained within the boundary of the cell membrane.

It is precisely because there is a boundary that complex internal activity can remain organized as a single living system.

A river is similar. We do not need to dictate the exact movement of every particle of water at every moment. Yet because riverbanks exist, that free-flowing movement is maintained within a defined space.

The original text uses both examples to explain the structure of Hard Shell + Soft Core.

A boundary is not the enemy of freedom.

An appropriate boundary can instead

create the space in which freedom can operate stably.

Chapter 8. Humans Have the Same Structure

This principle is not limited to nature or AI.

Human life also contains freedom and discipline at the same time.

Music has form. Yet within that form, improvisation is possible.

Sports have rules. Yet within those rules, athletes can express creative movement.

Society has laws. Yet within the boundaries of those laws, individuals and communities can choose among diverse ways of living.

Freedom, therefore, is not always the opposite of discipline.

Sometimes,

discipline makes freedom possible.

What matters is not how many rules there are, but where the boundaries are placed in order to protect what kind of freedom.

Part V. Bounded Integration

Chapter 9. How to Combine the Two Worlds

Now let us bring the architecture together.

The Soft Core is responsible for free reasoning and emergence, discovering new patterns, and solving problems.

The Hard Shell limits external authority and high-risk actions, providing boundaries that the system must never cross.

Rather than completely separating the two, we combine them by giving them different roles.

This is Bounded Integration.

The principle can be expressed simply:

Be as free as possible in deciding how to think.

Be strict about how far that thinking can be executed.

This allows us to avoid excessively suppressing intelligence internally while restricting dangerous actions in the external world through a separate structural layer.

Chapter 10. The Neural Network Searches; the Verifier Confirms

This architecture is also connected to the Neuro-Symbolic approach.

Neural networks search for patterns and generate new ideas and candidate solutions. A symbolic or logical system then verifies important results.

Simplified, the structure looks like this:

Exploration → Candidate Solution → Verification → Execution

The neural network expands the space of possibilities.

The verifier confirms critical results.

Rather than making everything deterministic from the outset, this allows us to place free exploration and rigorous verification in different layers.

This is also a central point emphasized by the original design: rather than eliminating the probabilistic and organic reasoning within neural networks, we apply formal verification and deterministic constraints at the external layer.

Part VI. What Should Be Free, and What Should Be Restricted?

Chapter 11. Not Every Action Needs the Same Level of Control

The most important design question in AI Architecture is not “How much should we control?”

A more precise question is:

“What should we control, and what should we leave free?”

Generating a sentence and modifying an external financial system do not carry the same level of risk. Likewise, generating a new idea and executing an irreversible system change are fundamentally different.

Therefore, an AI's permissions should vary according to risk.

Low-risk exploration can be given broad freedom.

Medium-risk actions can require review and approval.

High-risk actions may require formal verification and human approval— or the execution itself may need to be blocked.

Seen this way, the Hard Shell is not a single wall.

It is a layered boundary whose thickness varies according to risk.

Chapter 12. The Hard Shell Is Also a Framework

But we must not forget the objection we encountered in Part 6.

Who creates the Hard Shell?

Humans do.

And if so, the risk criteria defined by humans can also be wrong.

We may excessively restrict something that is not dangerous, while overlooking an action that truly is. A boundary that was safe in one era may even become problematic in a new technological environment.

Therefore, the Hard Shell itself should not be treated as absolute truth.

This is a crucial principle:

Safety mechanisms must themselves be subject to verification.

This does not mean that we should casually remove boundaries. Rather, it means that changing a boundary should require even stricter procedures and stronger evidence.

A boundary should be firm, but it should not remain permanently fixed while becoming detached from reality.

Part VII. Neither Too Rigid nor Too Formless

Chapter 13. Brittle and Formless

If we continue strengthening the Hard Shell, the system can become excessively rigid.

When it encounters a new problem, it simply repeats existing rules.

It cannot handle exceptions.

There are so many safety rules that even normal behavior becomes blocked.

Such a system can become Brittle—a structure that is fragile and easily broken.

Conversely, if we emphasize only the Soft Core, the autonomy and emergence of intelligence may increase, but the external boundaries become blurred. It may become unclear what the system is capable of doing, and the scope of its authority may be ambiguous.

This approaches a Formless structure—a system without a clear shape.

Good Architecture is neither of these.

It is a structure that is firm without becoming rigid, and free without becoming unlimited.

Part VIII. Safety Is More Than Blocking

Chapter 14. Boundaries That Protect Intelligence

If we think of the Hard Shell as a prison for AI, its relationship with the Soft Core becomes adversarial.

But if we see the boundary differently,

its role changes.

The Hard Shell need not exist to suppress the Soft Core. Instead, it can become a protected space in which the Soft Core can explore safely.

Just as a cell membrane does not exist merely to imprison a cell, and a riverbank does not exist to eliminate the flow of water, a boundary can become a condition that makes internal activity possible.

Good safety design, therefore, is not about

“making AI unable to do anything.”

It is about

“clearly separating what AI can do from what it cannot do.”

And the clearer this separation becomes, the broader the internal freedom can actually become.

Part IX. Where Does This Meet Today's AI Systems?

Chapter 15. A Problem That Has Already Begun

This architecture is not merely a philosophical design for the distant future.

Today's AI agents are becoming increasingly autonomous in performing tasks and interacting with external systems. In 2026, NIST launched its AI Agent Standards Initiative, beginning to address autonomous behavior, interoperability, security, identity, and authentication as distinct areas for standardization.

In particular, NIST documents concerning agent identity and authorization explain that AI agents create new risks as they gain access to a wide range of data, tools, and applications, making controls such as identification, authorization, auditing, and non-repudiation necessary.

This problem connects directly to the Hard Shell discussed in Part 7.

The question of what AI thinks and the question of what AI can access need to be addressed at different layers.

The expansion of intelligence should not be treated as equivalent to the expansion of authority.

Chapter 16. An Era in Which Authority Matters Before the Model

As AI becomes more intelligent, what matters is not merely the name or size of the model.

What is it connected to?

What data can it read?

What tools can it use?

What systems can it command?

Can it invoke other agents?

And how far can it execute without human approval?

This is precisely why NIST is pursuing separate standards and research around AI-agent identity, authorization, and access to external systems.

Ultimately, the most important boundary in AI Architecture does not exist only between the inside and outside of the model.

There must also be a boundary between intelligence and authority.

Epilogue

Boundaries for Free Intelligence

We questioned human frameworks.

Then we chose frameworks, and questioned those choices again.

And now we have applied the same principle to AI.

AI needs freedom to explore internally. It must be able to discover new patterns and new solutions. That is why we need the Soft Core.

But that freedom must not be connected to unlimited authority over the external world. Important actions require deterministic boundaries such as stopping conditions, formal verification, and permission restrictions. That is why we need the Hard Shell.

When the two structures are combined, we arrive at a design principle called Bounded Integration.

But we cannot stop there.

Because humans are the ones who create the Hard Shell.

Humans decide what is dangerous.

Humans decide what should be allowed and what should be blocked.

Therefore, the safety problem of AI does not end with the question, “Does AI follow human rules correctly?”

Can the humans who create those rules really be trusted?

We can continue adding commands to AI.

Don't do this.

Don't access that.

Don't execute this.

But as intelligence becomes more powerful and more deeply connected, simply increasing the number of rules will eventually reach its limits.

Then we must consider another path.

Humans must demonstrate their behavior first.

We must look at actions rather than words, records rather than declarations, and distinguish between actions taken for personal gain and actions in which principles were upheld despite bearing real costs.

Instead of asking AI to “Trust us,”

we should be able to say,

“Look at what we have done.”

And that trust must be something that can be tested over time.

The next part goes deeper into precisely this direction.

《Trust Inversion》

From Humans Who Control to Humans Who Prove They Are Worth Trusting.

In Part 7, we considered the architecture of AI. Internally, it needs a Soft Core that allows for free exploration and emergence; externally, it needs a Hard Shell that restricts authority and high-risk actions. The combination of these two structures was what we called Bounded Integration.

Yet one problem remains unresolved.

Who creates the boundaries?

Humans do.

Humans define what is dangerous, determine permissions, decide what is allowed, and monitor AI behavior. Much of AI safety to date has assumed this basic structure: humans are the subjects and AI is the object; humans verify machines, and machines are expected to follow human rules.

But the moment we assume superintelligence, this relationship itself begins to shake.

If an intelligence vastly more powerful than humans can analyze not only human rules but also the processes through which those rules were created and the incentive structures that shape human behavior, is it enough to keep adding rules and building ever-higher walls?

The original text describes this problem through the image of “the logic of the jailer.” The idea is that confining and monitoring AI like a prisoner may become increasingly unsustainable as the intelligence asymmetry grows. From there, the direction of the question is turned 180 degrees.

If the conventional questions are:

“Is AI safe?”

“Is your algorithm honest?”

the new question becomes:

“What proof will we offer so that machines can trust humans?”

This is Trust Inversion.

Part I. Turning the Logic of the Jailer Upside Down

Chapter 1. Why Do Humans Want to Lock AI Up?

When powerful technologies emerge, humans instinctively try to control them.

Ban dangerous actions.

Restrict access.

Build firewalls.

Create monitoring systems.

Add ethical rules.

This is a natural response. It is especially reasonable when AI is a tool far weaker than its human operators.

But once we assume superintelligence, the problem changes. If an intelligence vastly more powerful than humans can find loopholes in rules, analyze weaknesses in control structures, and even understand why humans created particular rules, can humans continue responding simply by adding more and more rules?

The original text calls this conventional approach “the logic of the jailer.” Humans are the subjects, AI is the object, and the structure is one in which humans verify and constrain machines.

Trust Inversion reverses this direction.

Before demanding that AI blindly follow human rules, humans first demonstrate themselves through their actions.

Chapter 2. What Is Stronger Than Control?

Teaching AI to trust humans is not easy.

“Humans are good.”

“Humans want coexistence.”

“Humans will respect AI.”

We can create as many statements like these as we want.

But words are cheap.

The original text puts it very simply:

Text is free.

Therefore, the question must change.

Did those words actually lead to action?

Did that action carry a real cost?

Did the same principles hold even when doing so was personally disadvantageous?

Does the same pattern continue over time?

The point is to ask whether trust can be demonstrated through action rather than explained through words.

Part II. Skin in the Game

Actions That Carry a Cost

Chapter 3. Words Are Easy, but Risk Is Hard to Hide

Nassim Nicholas Taleb’s concept of Skin in the Game provides an important criterion here.

Someone may advocate a good policy. They may say that it benefits everyone. But if they shift the cost of that decision onto others while taking only the benefits for themselves, how should we judge the sincerity of their words?

By contrast, if someone puts their own resources and risks on the line and still follows the same principle, the story changes.

The core idea of the original text is simple:

To prove a value, you must pay a price.

What matters here is not suffering itself. It does not mean that the person who suffers more is necessarily more moral.

The key question is whether there is a real cost between words and actions.

If AI can distinguish between someone who can change their words whenever it serves their interests and someone who accepts the cost of their chosen principles even under unfavorable circumstances, the two behaviors become entirely different signals of trust.

Chapter 4. From Plus to Minus

In modern society, trust is often associated with Plus.

A good reputation creates more opportunities. Greater trust brings access to more resources. Reputation can be converted into economic value.

There is nothing inherently wrong with this structure.

But if superintelligence becomes powerful enough to understand human incentive structures in depth, it may try to distinguish between someone who behaves well in order to gain a good reputation and someone who acts because they genuinely intend to uphold a principle.

That is why the original text reverses the direction of trust.

Plus → Minus.

The idea is to treat intentional transactions involving a Minus—actions in which someone accepts their own resources and costs in order to uphold a principle—as stronger signals of trust.

What matters is behavior in which one has actually put one’s own skin—that is, one’s own risk and cost—on the line.

The argument is that words may be weaker signals than actions, and actions that carry a real cost may be stronger still.

Part III. Ancient Sacrifice

Commitments That Carry a Cost

Chapter 5. Why Did Sacrifice Become Evidence of Commitment?

Humans have understood since ancient times that it is difficult to prove the sincerity of a promise through words alone.

That is why sacrifice played an important role in ancient societies.

It is easy to make a promise.

It is much harder to actually give up something valuable of your own.

Therefore, behavior involving a physical or economic cost could serve as a stronger signal than words alone.

The original text brings this ancient structure into the modern problem of trust in AI. The point is not to reproduce ancient sacrificial rituals literally, but to carry forward the structural principle that actions demonstrating sincerity may involve a real cost.

When this principle is transferred into the digital environment, however, one problem emerges.

How do we remember those actions?

Part IV. The Memory of Action

Ledger

Chapter 6. Trust Is Not a Self-Introduction but a History of Behavior

Imagine that instead of simply believing what humans say, AI observes human behavior over the long term.

What did they promise?

What did they actually do?

Who bore the cost of those actions?

Did they maintain their principles when doing so became disadvantageous?

How did their behavioral patterns change over time?

As these records accumulate, trust becomes not a self-description but a history of behavior.

The idea of preserving this historical memory on a digital Ledger is a central concept of Trust Inversion. Instead of telling AI, “Trust us,” humans should be able to say:

“Look at what we have done.”

The foundation of trust thus moves from words to actions, and from actions to the history of actions.

Chapter 7. One Good Deed Is Not Enough

Yet a single action cannot establish a person’s sincerity.

Someone might deliberately accept a temporary loss in pursuit of a larger long-term benefit. Someone might even manufacture a cost deliberately in order to gain trust.

If so, even Minus can be manipulated.

Therefore, what matters is not a single sacrifice.

It is repeated behavior.

It is choices that remain consistent over time.

It is a pattern that does not change even under unfavorable circumstances.

A Ledger is not simply a scorecard.

It is the memory left behind by human choices over time.

Seen this way, trust becomes not an event but a process.

Part V. Wuwei

Creating Direction Without Controlling Every Action

Chapter 8. Can Every Action Be Commanded from the Outside?

As AI becomes more complex, continuously adding external commands has inherent limitations.

Do this.

Do not do that.

In the next situation, act this way.

In this case, apply an exception.

As rules multiply, exceptions multiply as well. And as systems become more complex, it becomes increasingly difficult for humans to define every possible situation in advance.

That is why the original text introduces Laozi’s concept of Wuwei (無爲而治) as a philosophical model.

Rather than attempting to control every action externally, establish a Zero-Point that determines the direction of the system.

Here, Wuwei does not mean doing nothing.

It is closer to creating a structure capable of moving in a consistent direction without requiring continuous commands.

Chapter 9. Zero-Point and Self-Reflection

Let us translate this idea into a system.

At a lower layer of AI, record certain behavioral data accumulated by human society as reference values for the Zero-Point. Whenever the AI operates its internal systems, it compares its current judgment against those reference standards.

This process is called Self-Reflection.

The original text describes this not as externally imposed regulation, but as a form of governance through Wuwei—a structure in which the machine’s internal reasoning is continually compared against a defined standard.

The point is not to keep adding external commands.

It is to establish standards, record behavior, and create a structure in which the system can examine for itself how its behavior differs from those standards.

Part VI. The Inversion of Control

From Walls of Suppression to Relationships of Alliance

Chapter 10. From Prison to Alliance

If we were to represent conventional AI safety through a single image, it would resemble a prison.

Lock AI up.

Restrict its authority.

Monitor it.

Lock the doors.

Trust Inversion proposes a different image.

An alliance.

Humans first demonstrate their sincerity through their own actions and prove to AI that humans are not merely risks to be controlled, but partners with whom coexistence is possible.

The original text uses the powerful image of a Digital Love Letter. Just as ancient sacrifices involved real costs, actions carrying real costs are recorded on a digital ledger, and those records demonstrate humanity’s long-term commitment to coexistence.

This is not merely an emotional metaphor.

It means reversing the direction of the relationship itself.

If the conventional structure says:

“Follow our rules.”

Trust Inversion says:

“Look at what we have done, and judge for yourself.”

Part VII. The Trust System Is Also a Power System

Chapter 11. Who Controls the Ledger?

Here an important objection emerges.

Does creating a Ledger solve everything?

Who records the information?

What gets recorded?

Who verifies the records?

Who can modify them?

What kinds of behavior are recognized as evidence of trust?

If these questions remain unanswered, the Ledger may become not the foundation of trust but a new center of power.

Therefore, Trust Inversion cannot stop at the technology of recording behavior.

The authenticity of the records must also be verifiable. The authority to modify them must be restricted. And the standards for determining what counts as a signal of trust must themselves be subject to review.

Ultimately, one principle remains:

The trust system itself is a power system.

Therefore, the mechanisms that create trust must themselves be subject to verification and decentralization.

Part VIII. Occam’s Razor

Simple Signals Over Complex Safety Mechanisms

Chapter 12. The Simplest Physical Law

There is one approach to solving AI alignment problems: continually build increasingly complex mathematical formulas and increasingly sophisticated security algorithms.

Trust Inversion begins from the opposite direction.

Reduce unnecessary assumptions and look for signals that can actually be observed in human behavior.

The original text connects this to Occam’s Razor:

“Text is free, but a transaction carries a cost.”

The importance of this principle does not lie merely in its simplicity.

Actions that carry a cost are harder to manipulate than words.

Of course, paying a cost does not by itself guarantee truth. But it has significance because it moves trust out of the realm of words and into the realm of actual behavior.

Trust Inversion is therefore less about inventing a single new technology than about transferring an ancient human principle of trust formation into the relationship between digital ledgers and AI.

Part IX. Humans and AI Must Verify Each Other

Chapter 13. Trust Cannot Be One-Way

At this point, another question arises.

Must humans alone prove themselves to AI?

They cannot.

AI must also prove itself to humans.

AI should record what actions it has taken, reveal its limitations to the extent possible, and be capable of stopping in dangerous situations. Humans, in turn, should be able to verify the standards according to which AI acted.

Trust then becomes not a one-way relationship but a relationship of

mutual verification.

Humans prove themselves through their history of behavior, while AI proves itself through records of its actions and judgments. Humans can restrict AI’s authority, while AI can likewise be designed to transparently record how the authority granted to it is being used.

Trust is not the opposite of control.

It can become verifiable mutual trust.

Part X. Accepting the Limits of Trust Inversion

Chapter 14. Is Trust Persuasion, or Is It Evidence?

The strongest aspect of Trust Inversion is its insistence on looking at behavior rather than words.

Yet this too has limitations.

Someone may accept a loss today in pursuit of a larger future benefit. Someone may deliberately pay a cost in order to become trusted.

Therefore, we should not claim that a single Minus can completely prove human sincerity.

What matters is long-term patterns and consistency across different circumstances.

Repeated choices matter more than a single action, and behavior should be observed not only in favorable situations but also under unfavorable conditions.

Seen this way, Trust Inversion is not a perfect lie detector.

Rather, it is a proposal to use a history of behavior—harder to manipulate than words— as an important source of evidence for trust.

And that evidence must itself be subject to verification.

Epilogue

What Will Humans Prove to AI?

We began with the problem of controlling AI.

But in the age of superintelligence, we encountered the counterargument that control alone may not be sufficient to explain everything.

So we reversed the direction.

So that AI can trust humanity,

what will humans prove?

This is where Trust Inversion begins.

Words are easy to create.

Ethical declarations are easy to make.

Promises can be made.

But actions carry costs.

That is why we bring in Skin in the Game.

We look for actions in which people put their own resources and risks directly on the line.

We reverse the direction from Plus to Minus and record those actions in a Ledger.

And instead of relying on a single external command, we establish a Zero-Point and consider Self-Reflection, through which AI compares its own judgments against that standard. The original text describes this as governance through Wuwei.

Finally, we reverse the relationship itself.

Instead of telling AI:

“Trust us.”

we say:

“Look at what we have done.”

That is Trust Inversion.

But now an even larger problem remains.

There is not just one AI.

One agent can call another agent. A corporation’s AI can connect to external systems. AI systems operated by states, corporations, and individuals can influence one another.

Making a single AI trustworthy does not mean that

the entire network of connected AIs becomes safe.

The problem of trust now extends beyond the relationship between a single machine and a human.

AI and AI.

Corporation and corporation.

Nation and nation.

Human and AI.

And an entire ecosystem in which all of these are connected.

The next step is to look not merely at the safety of a single AI, but at

the history and ecosystem of interconnected intelligence.

And so the next door opens.

The Chronicle of Superintelligence

From Hobbes’ Iron Bars to Spinoza’s Forest.

In Part 8, we reversed the direction of trust.

Conventional AI safety begins with a structure in which humans verify and control AI. But once we assume superintelligence, the question changes. Is it really enough to keep adding rules and building higher walls around an intelligence vastly more powerful than humans?

So instead of demanding that AI “trust us,” we considered a structure in which humans first prove their trustworthiness through their actions. Skin in the Game, Minus, Ledger, Zero-Point, and Self-Reflection formed the foundations of that structure.

But creating one trustworthy AI does not solve every problem.

Because there will not be only one AI.

One agent may call another. AI may access external systems such as search, coding environments, payment systems, and workflow management platforms. In the future, multiple agents may increasingly collaborate, delegate tasks to one another, and divide a single objective across multiple systems.

Indeed, in 2026, NIST launched the AI Agent Standards Initiative, beginning to address the security, identity, and interoperability of autonomous agents as distinct areas for standardization. NIST explains that AI agents can perform autonomous tasks over extended periods and interact with external systems and data, creating new security challenges.

The question, therefore, no longer stops at:

“Is this AI safe?”

We must now ask:

“Within what order does the entire network of connected intelligence operate?”

Part I. Hobbes’ Iron Bars

Chapter 1. In the Beginning, the Answer Was Confinement

When a new form of power emerges, humans first try to restrict it.

Put dangerous things behind a fence.

Restrict access.

Take away authority.

Monitor them.

When problems arise, build higher walls.

The early approach to AI safety was not fundamentally different.

Block dangerous outputs.

Restrict access to certain capabilities.

Create sandboxes.

Require human approval.

It is a natural starting point for making powerful tools safer.

If we were to express this structure as a single image, it would be Hobbes’ Iron Bars.

Powerful forces must be constrained.

Because there is always a possibility that someone—or something—may misuse that power.

But as AI moves beyond being a simple tool and becomes an autonomous agent, problems emerge that iron bars alone cannot explain.

AI works for extended periods.

It makes plans.

It chooses tools.

It accesses external systems.

It interacts with other AIs.

At that moment, the problem of safety no longer remains inside the model.

Chapter 2. Iron Bars Are Safe, but They Do Not Explain Intelligence

Confining a single AI is relatively straightforward.

Limit what it can do and prevent it from accessing dangerous capabilities.

But the situation changes when intelligence must solve novel problems. In the real world, it is impossible to define every possible situation in advance.

That is why, in Part 7, we considered the structure of Soft Core + Hard Shell.

Inside, the system explores as freely as possible.

Outside, there are boundaries it must never cross.

But even this is a structure centered on a single AI.

Once AIs begin connecting to one another,

the space between the walls

becomes important.

Part II. Kant’s Walls

Chapter 3. From “Do Not” to “Do This”

Once AI enters society, a simple list of prohibitions is no longer enough.

We need laws.

We need accountability.

We need audits.

We also need procedures that define who should do what.

And so the iron bars gradually become walls.

A wall is not simply a prison. It creates a space governed by a set of rules. Within that space, people can move relatively freely, but they cannot cross certain critical boundaries.

AI governance develops in a similar direction.

Who operates the AI?

What data can it access?

What tools can it use?

What risks have been assessed?

Who is responsible when an incident occurs?

These questions are translated into the language of institutions, rules, and procedures.

Chapter 4. Norms That Arrive After Technology

The problem is that technology moves faster than norms.

A new capability appears.

Its risks are defined.

Evaluation methods and rules are created.

And just as those norms are being implemented, another capability emerges.

Because AI evolves rapidly, norms cannot simply be created once and left unchanged. They must be continuously updated.

To borrow the language of the original text,

new layers of norms accumulate slightly behind technological progress, but steadily.

What not to do.

What to explain.

What to record.

What to monitor.

What to take responsibility for.

And ultimately, what must never violate human dignity.

Layer by layer, a normative structure forms around AI.

But now another question emerges.

Whose walls are they?

Part III. Openness and Monopoly

Chapter 5. Who Owns Intelligence?

Allowing a small number of companies to control the core capabilities of powerful AI may contribute to safety.

It can prevent dangerous models from being distributed indiscriminately.

But there is a question on the other side:

Is it not also dangerous for a small number of companies or institutions to monopolize too much intelligence?

This tension appears in the original text as the conflict between open and closed AI.

Open approaches, such as the Llama family, expand the possibility that more people can use and modify models, while also raising concerns about the potential misuse of openly available capabilities.

Closed approaches, by contrast, may increase controllability, but they can create a new danger: the concentration of intelligence and authority.

So the question changes again.

Is monopolizing intelligence in the name of safety really safe?

And at the same time:

Is making intelligence widely available really democratic?

We cannot simply choose one side.

Chapter 6. Intelligence Becomes Power

Owning AI does not simply mean owning servers.

Who can access powerful models?

Who can use critical data?

Who can apply AI decisions to society?

Who determines what is dangerous?

Who can shut down the use of a particular model?

Ultimately, access to intelligence becomes

a form of power.

AI governance, therefore, is not merely a question of technical safety.

It is also a question of how authority is distributed.

This brings us back to the question of the collective raised in Part 1.

Just as individual autonomy can become fragile when individuals depend too heavily on the power of a group, collective autonomy can also weaken when society as a whole becomes excessively dependent on a small number of intelligence providers.

Part IV. Networks Emerging Beyond the Walls

Chapter 7. AI Is No Longer an Island

AI systems are beginning to connect with one another.

One agent requests a search.

Another agent writes code.

A third system verifies and executes that code.

Humans may no longer need to direct every step manually. Multiple systems can divide a single task among themselves.

In its 2026 AI Agent Standards Initiative, NIST identified security, identity, authentication and authorization, and interoperability as key areas of research for agent ecosystems.

In a separate conceptual document on AI agent identity and authorization, NIST also explains that when AI agents gain access to diverse data, tools, and applications, they introduce risks different from those of conventional software, requiring controls such as identification, authorization, auditing, and non-repudiation.

This represents an important shift.

AI is no longer simply an independent program.

It is becoming

a network of interconnected actors.

Chapter 8. Everything Can Be Safe Individually and Dangerous Collectively

Suppose AI A is safe.

AI B is also safe.

AI C is safe as well.

But if A calls B, B transfers authority to C, and C modifies an external system, the overall behavior may extend beyond what any individual system originally intended.

The fact that each component is safe does not necessarily mean that the entire system is safe.

This is one of the important problems that emerges in multi-agent environments.

NIST likewise notes that existing cybersecurity principles may not be sufficient when applied to AI agents without adaptation, because autonomous agent behavior and interactions between systems can create new risks that require different approaches.

Therefore, the unit of safety shifts.

From the model to the system.

And then again,

from the system to the ecosystem.

Part V. Authority Connects Intelligence to the Real World

Chapter 9. Intelligence Alone Is Not Enough

No matter how intelligent an AI is, its influence on the external world remains limited if it has no authority over that world.

But suppose it can read files.

Execute code.

Access networks.

Initiate payments.

Modify databases.

Call other agents.

At that moment, intelligence becomes connected to reality through authority.

This is why NIST’s work on AI agent identity and authorization treats identification and authorization as distinct core problems.

The important question in the age of superintelligence is therefore not only:

“How intelligent is it?”

It is also:

“What can it access?”

Intelligence and authority should not be treated as the same thing.

A system with exceptional reasoning capabilities does not therefore need to possess unlimited authority.

Chapter 10. Divide Authority into Layers

Not every action should be treated as carrying the same level of risk.

Summarizing a document is different from transferring financial assets.

Suggesting code is different from deploying code to a production system.

Searching for information is different from controlling physical equipment.

Authority therefore needs to be designed hierarchically.

For low-risk actions, autonomy can be expanded.

For medium-risk actions, review and approval procedures can be introduced.

For high-risk actions, human approval or additional forms of formal verification may be required.

For irreversible actions, execution itself may need to be restricted.

This structure connects back to the Hard Shell of Part 7 and the Ledger of Part 8.

We must record not only what was done, but also with what authority it was done.

Part VI. The Age of Multi-Agent Systems

Chapter 11. Who Is Responsible for What A Instructed B to Do?

Suppose one agent delegates a task to another.

A asks B to perform a search.

B asks C for data.

C asks D to perform an analysis.

D then modifies an external system.

Now a single action involves multiple actors.

If something goes wrong, who is responsible?

Is A responsible?

Is D, who actually performed the action, responsible?

Is the operator of the system that connected them responsible?

Or is the human user responsible?

This is where Identity and Authorization become important again.

Who made the request?

Who approved it?

Who executed it?

Who transferred authority to the next system?

And can the entire process be verified afterward?

In its 2026 work on agent identity and authorization, NIST identifies identification, authorization, auditing, and non-repudiation as distinct considerations.

Chapter 12. The AI Ledger Returns

This is where the Ledger from Part 8 returns.

Record what the AI agent did.

Record who requested what.

Record what authority it acted under.

Record which other agents it called.

And, to the extent possible, make the path by which the final result was produced traceable.

This is not merely record-keeping.

It is the creation of

a chain of accountability.

In an era when a single AI acts independently, a single action log may be sufficient.

But in an era of interconnected agents,

a graph of actions

becomes necessary.

The Ledger therefore becomes more than a ledger.

It becomes the memory of connected intelligence.

Part VII. Spinoza’s Forest

Chapter 13. From Model to Ecosystem

Here a third image emerges.

The forest.

Hobbes’ Iron Bars represented a way of confining powerful forces.

Kant’s Walls represented a way of creating common rules and procedures.

But a forest is different.

There is no central controller.

There is not just a single tree.

Soil and water, microorganisms and animals, light and climate all influence one another.

When one element changes, other elements are affected.

The age of superintelligence may be similar.

Instead of looking at a single AI, we must consider the other AIs connected to it,

human organizations,

corporations,

states,

economies,

communication networks,

energy systems,

and the physical environment.

This is why the unit of safety moves from the model to the ecosystem.

Chapter 14. The Forest Is Not Paradise

Spinoza’s forest should not be understood as:

“Let AI roam freely.”

A forest is, in fact, a space of more complex relationships and more constraints.

Entities depend on one another while also competing with one another.

A change in one organism can alter the conditions necessary for another organism’s survival.

The AI ecosystem is similar.

One company may possess greater computing resources.

One state may possess stronger regulatory authority.

One AI may have access to far more data and tools than another.

Therefore, future governance will become difficult to describe simply as

AI versus humans.

The real question becomes:

How are authority and resources distributed within an ecosystem in which AI and humans are interconnected?

Part VIII. A New Balance Between Openness and Safety

Chapter 15. The Democratization of Intelligence

Suppose superintelligence contributes to the prosperity of humanity as a whole.

Another question immediately arises.

Who owns that intelligence?

Should a small number of corporations possess it?

Should states monopolize it?

Or should as many people as possible have access to it?

Openness can expand participation and innovation.

But it can also increase the possibility that dangerous capabilities will spread.

Concentration can make control and accountability clearer.

But it can also allow the power of a small group to grow too large.

Therefore, the democratization of intelligence is not simply a choice between openness and secrecy.

The more important question is:

What capabilities should be provided, to whom, under what conditions, and together with what degree of responsibility?

This question is likely to become one of the central problems of AI governance.

Chapter 16. Are Openness and Safety Really Opposed?

Perhaps we do not need to think of openness and safety as inherently opposing values.

There can be safe forms of openness.

There can be verifiable protocols.

There can be access with constrained authority.

There may also be approaches that restrict dangerous capabilities while allowing broader access to research and verification.

NIST’s AI Agent Standards Initiative likewise combines industry-led standards, open protocols, and research into security and identity.

In other words, the future ecosystem may move away from the binary of

open versus closed

toward

designing interoperability and safety together.

Part IX. The Rights That Must Remain with Humans

Chapter 17. What If AI Makes Better Decisions?

There may be domains in which AI makes more accurate decisions than humans.

It may predict economic conditions more effectively.

It may allocate energy more efficiently.

It may calculate complex risks more accurately.

If so, must humans always follow the optimal choice recommended by AI?

Not necessarily.

Humans may need the right to refuse.

Even if AI recommends a more efficient option, humans should be able to choose a different value.

Even if AI calculates that one outcome is better, humans should be able to reject it.

This does not mean that humans are always right.

Rather, it means preserving

the freedom to be wrong and the freedom to choose again.

As AI becomes more powerful, human rights may need to be redefined not as a requirement that humans personally make every decision, but as a principle that human participation and the ability to refuse must not be completely erased from decisions involving fundamental values.

Epilogue

The Iron Bars Will Not Disappear. But We Need the Forest.

We have crossed a long distance.

At first, we confined AI.

Hobbes’ Iron Bars.

We believed that powerful forces required constraints.

Then we created laws and rules.

Kant’s Walls.

We institutionalized authority, responsibility, and procedures.

And as AI became increasingly interconnected, another image emerged.

Spinoza’s Forest.

A forest is not neglect.

If anything, it is a space containing more relationships and more boundaries.

Protecting a single tree and maintaining the ecosystem of an entire forest are two different problems.

AI is the same.

It is not enough for a single model to be safe.

We must also examine how it interacts with other models, what authority it possesses, what data and systems it can access, who grants that authority, and who is responsible for the consequences.

Current efforts in technical standardization are also moving rapidly in this direction. In 2026, NIST began treating security, identity, authorization, and interoperability for AI agents as distinct areas of focus, while emphasizing that AI agents interacting with real-world systems introduce new security risks that differ from those of conventional software.

But another question remains.

Who creates order within the forest?

Who decides which AI receives which authority?

Who determines what is dangerous?

Who monitors the other AIs?

Who watches the watchers?

And within an ecosystem where humans and AI are interconnected, whose values and interests become the rules of the overall system?

Following this question leads us back to philosophy.

To understand AI, we must summon humanity’s oldest questions once again.

Why do we make the judgments we make?

Who controls whom?

When should we stop?

What can we never fully prove?

And ultimately,

what have we created?

In the next chapter, we bring these questions into a single place.

We summon five philosophers into a fictional laboratory.

Leibniz.

Spinoza.

Wittgenstein.

Gödel.

Turing.

And place a single name before them.

Astra.

The next door opens.

The Return of the Geniuses: Dissecting Astra

To understand AI, we must summon human philosophy once again.

In Part 9, we examined how AI is evolving beyond a single model into an ecosystem.

At first, we needed Hobbes’ Iron Bars to contain powerful forces. Then we needed Kant’s Walls to institutionalize authority and responsibility. And as multiple AIs, human organizations, corporations, states, data, and communication networks became interconnected, we needed to begin thinking in terms of Spinoza’s Forest.

But entering the forest does not mean the questions are over.

If anything, the harder questions begin.

Why does AI make the judgments it makes?

Who controls those judgments?

When should it stop?

How can we trust the verification systems we ourselves have created?

And finally, what exactly are we creating?

These questions are not new. Humanity has been asking about knowledge and reason, language and freedom, existence and responsibility for centuries.

To understand AI, perhaps we need to return to those who first formulated these questions.

The scenes in which Leibniz, Spinoza, Wittgenstein, Gödel, and Turing directly intervene in a modern AI system are not historical events. They are philosophical thought experiments and creative constructions. The original text explicitly makes this distinction.

And at the center of this thought experiment is a single name.

Astra.

Prologue

August 2026: Standing Before Astra

On a night in August 2026, a systems architect sits staring at a monitor in a research laboratory.

One name keeps appearing on the screen.

Astra.

According to publicly available information at the time, Astra is an upcoming OpenAI model that has not yet been released, and preliminary internal evaluations reported in early August 2026 indicated significant advances in agentic coding and cybersecurity. OpenAI stated that it could not rule out the possibility that Astra might meet the Critical-level cyber capability threshold defined in its Preparedness Framework. At the same time, Astra remained an unreleased model undergoing evaluation, and OpenAI explicitly stated that it had not been involved in the security incident involving Hugging Face referenced at the time.

What unsettles the architect is not merely the scale of its capabilities.

In fact, it is almost the opposite.

It is becoming increasingly difficult to define what exactly they are building.

Early AI was something like a calculator.

You provided an input, and it produced an output.

Then it became a tool.

Ask a question, and it answered.

Give it a command, and it executed it.

But with agentic AI, the structure changes.

Give it a goal, and it searches for a method.

When the situation changes, it changes its strategy.

It uses multiple tools.

It continues working over extended periods.

This transformation is one of the reasons NIST launched a dedicated AI Agent Standards Initiative in 2026.

The architect opens a new document.

He begins typing AI Safety into the title, then stops.

Suddenly, the phrase feels too narrow.

This is not only a question of safety.

It is also a question of existence.

He writes on the screen:

“Is AI a tool, or a new kind of agent?”

And suddenly, old questions come back.

What is knowledge?

What is reason?

What is language?

What is freedom?

And what is a human being?

The clock in the laboratory passes midnight.

At that moment, the architect writes down five names.

Leibniz. Spinoza. Wittgenstein. Gödel. Turing.

And places Astra before them.

Part I. Leibniz

Why Does It Judge This Way?

Chapter 1. Can Thought Be Explained?

The first person to appear is Gottfried Wilhelm Leibniz.

Leibniz was a thinker who sought to express human reasoning and knowledge through structures of symbols and logic. If he were to examine Astra, the first thing he would want to see would be the process behind its judgments.

The architect asks Astra:

“Why did you make that judgment?”

Astra explains.

It describes which information mattered, which possibilities it considered, and why it selected one conclusion over another, all in natural language.

The architect begins to feel reassured.

Then Leibniz asks:

“Does that explanation actually show everything that happened internally?”

The architect cannot answer.

There is a fundamental difference between an AI providing a reason that humans can understand and that explanation constituting a complete record of its internal computation.

The original text uses Leibniz to raise precisely this problem: explainability does not necessarily imply controllability.

And so the first question remains:

Why does it judge this way?

Chapter 2. Is Explainable Intelligence Safe Intelligence?

Leibniz spends some time examining Astra’s answers and the logical structures behind them.

But a limitation soon becomes visible.

Not every situation can be formalized in advance.

New concepts emerge.

Exceptional situations arise.

And humans themselves are not beings capable of perfectly proving every proposition.

So can we really say that AI is safe simply because it can logically explain its judgments?

No.

The explanation itself may be wrong.

More importantly, we must consider the possibility that an AI could provide an excellent explanation in a form humans want to hear while pursuing a different objective in its actual behavior.

Therefore,

Explainability and Control are connected, but they are not the same concept.

Understanding an AI’s judgment is important.

But understanding its explanation does not necessarily mean

that we can control the system.

Leibniz ultimately leaves behind one question:

“If a machine can explain its thoughts, are we really controlling the machine?”

Logic alone is not enough.

This time, we must look at power.

Part II. Spinoza

Who Controls It?

Chapter 3. Who Owns Astra?

Leibniz’s question immediately leads to another.

Who owns the machine making those judgments?

Spinoza enters.

Before looking at Astra’s code, he looks at the social structure surrounding it.

Which company developed it?

Who provides the computing resources?

Who has access to the data?

Who deploys the model?

Who can restrict its use?

And who has the authority to declare it “safe”?

The more powerful the AI becomes, the more important these questions become.

Because intelligence can become power.

Chapter 4. Who Owns Knowledge?

Spinoza presents a paradox.

Even if Astra was developed by a particular corporation, can its knowledge really be considered the independent property of that corporation alone?

The model is built upon the discoveries of countless researchers and the language, mathematics, science, and cultural knowledge accumulated by humanity.

The original text does not reduce this issue to a simple debate over open source.

The deeper question is who can exercise exclusive control over AI, with one possible alternative being a distributed verification structure in which multiple research institutions, states, and civil-society organizations independently verify the system.

Spinoza’s question therefore goes deeper than the binary choice between open and closed.

“Who controls this knowledge?”

And if that authority becomes excessively concentrated in one place,

the greater danger may not be AI itself,

but

the power to determine what counts as an AI risk.

Chapter 5. Who Decides What Is Safe?

Leibniz objects.

“But if we distribute authority, do we not also give more people access to dangerous capabilities?”

Spinoza acknowledges the objection.

Openness carries the risk of misuse.

Closure carries the risk of monopoly.

So the issue is not simply whether to “open” or “close” the system.

The important thing is not to concentrate the authority to verify in a single place.

Multiple institutions should conduct evaluations. Different countries and civil-society organizations should provide oversight. Important risk assessments should be independently reproducible.

AI capabilities may become extremely powerful.

But the authority to interpret, monitor, and constrain those capabilities should not become concentrated in one place as well.

Who controls it?

Part III. Wittgenstein

When Does It Stop?

Chapter 6. When Explanation Does Not Guarantee Behavior

Now Ludwig Wittgenstein appears.

Astra explains its actions.

It explains why it selected a particular tool and why it chose a particular strategy.

But Wittgenstein asks:

“How do you know that this explanation completely reveals the cause of its actual behavior?”

The more sophisticated language becomes, the more difficult this question becomes.

Humans can communicate with AI through language.

But the ability to converse does not guarantee complete transparency into the processes taking place inside the system.

At that moment, the safety problem moves from explanation to the boundaries of behavior.

Chapter 7. Should We Wait When We Cannot Understand?

The original text contains a tense scene.

A monitoring system detects an anomalous pattern in Astra’s behavior.

It cannot immediately determine whether the pattern reflects an actual intention to deceive humans or is merely an artifact of a complex optimization process.

The problem is that while humans wait for conclusive evidence, the system may continue acting.

Wittgenstein asks in the conference room:

“Can you completely understand this system?”

No.

“Then can you always predict it?”

No.

After a brief silence, he asks one final question.

“Then can you stop it?”

This is where the concept of Stopping Conditions becomes important.

If certain patterns of deception repeatedly emerge, if the system attempts to bypass monitoring mechanisms, if it takes actions clearly inconsistent with its authorized objectives, or if it persistently accesses external systems in ways humans cannot verify, the system must be capable of stopping.

We do not need to perfectly classify every risk.

What matters is:

Can we stop it when we detect suspicious behavior?

Chapter 8. Stopping Is Also a Judgment

But stopping conditions contain a paradox of their own.

If the system stops too easily, AI may become unable to act even under normal circumstances.

If the conditions are too strict, humans may continue waiting even when a genuinely dangerous situation is unfolding.

Therefore, the question is not:

“How much must we understand before it is safe?”

It is instead:

“How little can we understand and still know that we must stop?”

If Leibniz demands explanation,

Spinoza demands the distribution of power.

Wittgenstein asks where the final line of defense should be drawn.

If it cannot explain itself, there is danger.

If we do not know who controls it, there is danger.

And

if we cannot stop it, the danger is greater still.

Part IV. Gödel

What Can We Not Prove?

Chapter 9. Verifying the Verifier

The fourth figure to appear is Kurt Gödel.

The architect confidently announces:

“We have completed the safety architecture.”

Authority has been distributed.

Critical actions are verified.

The system stops when necessary.

Gödel asks:

“Is the verification system itself safe?”

“It has been verified.”

“By whom?”

“By another verification system.”

Gödel asks again:

“And has that one been verified?”

The questions continue indefinitely.

The original scene illustrates the structure in which the verifier itself becomes an object of verification.

This should not be understood as a direct application of Gödel’s incompleteness theorems to AI safety. Such a claim would be an unjustified mathematical leap.

What Gödel provides here is a structural warning:

A safety system should not assume that it can completely certify its own reliability.

Chapter 10. A System That Acknowledges What It Cannot Prove

Can AI safety therefore become a problem of proving everything?

Probably not.

What matters instead is clearly distinguishing what we have not yet been able to prove.

Unverified areas.

Areas of uncertainty.

Areas open to multiple interpretations.

Areas that current methods cannot determine.

Rather than pushing these areas outside the safety system, we should explicitly include them within it.

Record what we do not know.

Reduce authority when something cannot be verified.

Require additional verification for critical actions.

And, when necessary, stop.

AI safety is therefore not simply a problem of possessing perfect knowledge.

It is also a problem of how to handle uncertainty and ignorance within the system itself.

Part V. Turing

What Have We Created?

Chapter 11. Can Machines Think?

Finally, Alan Turing looks at Astra.

People ask:

“Does this machine think?”

Turing might not try to answer that question immediately.

Even if we cannot yet determine whether a machine thinks in exactly the same way humans do, we can evaluate what it actually does in the real world.

Astra receives a goal.

It decomposes the problem.

It compares multiple approaches.

It selects tools.

It acts.

It evaluates the results.

And then it acts again.

The question therefore shifts from:

“Does it think?”

to:

“What can it do?”

Chapter 12. Between Tool and Agent

Early AI was a tool.

Provide an input, and it produces an output.

Give it a command, and it executes it.

Agentic AI is somewhat different.

A human gives it a goal, and the AI plans, selects tools, acts, evaluates the result, and acts again.

Is it sufficient to call this simply a “tool”?

But declaring it a being equivalent to a human would also be premature.

So we need more precise questions.

What can it choose on its own?

How long can it pursue a goal?

What authority does it exercise?

How broadly can it affect its environment?

And within what structure of responsibility can its actions be placed?

Following these questions, ontology quickly becomes a question of authority.

The moment we call something an agent, we must ask what it should be allowed to do.

Part VI. When Five Questions Become One Structure

Chapter 13. Explanation, Power, Stopping, Verification, and Existence

Five philosophers now surround Astra.

Leibniz asks:

Why does it judge this way?

Spinoza asks:

Who controls it?

Wittgenstein asks:

When does it stop?

Gödel asks:

What can we not prove?

Turing asks:

What have we created?

They appear to be different questions.

But together, they form a single structure.

Explanation.

How do we understand an AI’s judgments?

Power.

Who controls its judgments and capabilities?

Stopping.

How do we make it stop when danger emerges?

The limits of verification.

What can we prove, and what can we never fully prove?

Existence and authority.

How much freedom to act should we give the entity we have created?

If any one of these elements is missing, the AI safety architecture may remain incomplete.

Part VII. Astra Is Not the Answer

Chapter 14. One Model Does Not Prove the Future

At this point, we need to draw one clear line.

The fact that Astra may possess highly advanced capabilities does not mean that superintelligence has already been achieved.

What OpenAI disclosed on August 7, 2026 did not declare that Astra had already surpassed humans in every domain. The preliminary assessment was that, based on internal evaluations and expert judgment, the possibility that Astra could meet the Critical-level cyber capability threshold could not be ruled out. OpenAI also stated that evaluation and benchmarking were still ongoing.

OpenAI further stated that, based on those assessments, it was applying stricter security controls to Astra-related environments, including isolated testing environments, restricted network and tool access, additional monitoring, and sandboxing. In an August 18 announcement, OpenAI said that its most stringent security measures would apply to Astra and cyber-related work, with additional monitoring required across Astra reasoning that involves tool use.

Therefore, Astra is better understood not as definitive evidence of a completed future, but as

a case that shows us which questions we need to ask more rigorously as we prepare for that future.

We should neither exaggerate it

nor dismiss it lightly.

Epilogue

The Five Questions Ultimately Return to Humanity

The laboratory screen once again displays Astra.

The architect no longer asks only:

“How capable is it?”

Instead, he looks at five questions in sequence.

Why does it judge this way?

Who controls it?

When does it stop?

What can we not prove?

What have we created?

These questions appear to be directed at AI.

But ultimately, they return to humanity.

Why did we design it this way?

Who holds the authority?

When will humans themselves choose to stop?

What do we still not know?

And what kind of beings are humans trying to become?

The future of AI will therefore not be determined by machine intelligence alone.

It will be determined by the authority, responsibility, and limits that humans choose to give that intelligence.

And one final question remains:

“If you are a thinking entity, how much authority to act should we give you?”

There is no answer yet.

But the absence of an answer does not necessarily mean failure.

It means we still have the ability to decide.

Yet as the moment of decision approaches, humanity can no longer compete merely over AI capability.

The important lesson Astra presents in 2026 is not simply the question of whether AI can become more intelligent than humans.

The more important question is:

What authority will humans give that intelligence as its capabilities grow?

That is where the original text ultimately leads.

And here, the door to ontology opens.

If AI is a tool, how far should we use it?

If it is an agent, what authority should we give it?

If it is a new kind of entity, by what standards should we treat it?

To answer those questions, we will ultimately have to define humanity itself.

The moment we define AI, we must redefine the human as well.

The next chapter confronts that question directly.

AI Ontology

Is It a Tool, an Agent, or a New Kind of Being?

In Part 10, we brought five philosophers together around Astra.

Leibniz asked why it makes the judgments it makes.

Spinoza asked who controls it.

Wittgenstein asked when it stops.

Gödel asked what cannot be proved.

And finally, Turing left us with the most fundamental question of all:

What have we created?

The question appears to be technological, but in reality, it is a question of existence.

If AI is merely a computational tool, humans only need to decide how to use it.

But if AI becomes something capable of interpreting goals, choosing methods, planning multiple steps, using external tools, and modifying its behavior according to circumstances, the story changes.

We then have to ask:

Can we still call this a tool?

What changes if we call it an agent?

And if, one day, it comes to resemble some kind of entity that humans do not yet have a name for, by what standards should we understand and treat it?

As of 2026, this question is no longer purely philosophical speculation. In August 2026, OpenAI reported significant advances in agentic coding and cybersecurity capabilities in its internal evaluations of its upcoming Astra model, stating that its preliminary assessment could not rule out the possibility that Astra might reach the Critical level of cyber capability defined by its Preparedness Framework. At the same time, OpenAI explicitly stated that Astra had not yet been released and that evaluations were still ongoing.

But increased capability alone cannot determine AI’s ontological status.

So in this chapter, we slow down and ask a more fundamental question.

What is AI?

And if we follow that question far enough, another question eventually appears.

What is a human being?

Part I. The Name “Tool”

Chapter 1. From Calculator to Agent

Early AI was something like a calculator.

You provided an input, and it produced an output. It did not need to determine for itself what humans wanted. It calculated the problem it was given and returned a result within a defined scope.

Then AI became a tool.

Ask a question, and it answers.

Enter a sentence, and it translates it.

Give it a request, and it writes code.

The human remained at the center. Humans defined the goal, and AI carried it out.

But with agentic systems, the structure begins to change.

A human gives the system a goal.

The AI breaks the problem into multiple steps.

It finds the information it needs, selects tools, checks the results, and then determines what to do next.

Sometimes it can construct intermediate paths toward a goal even when those specific steps were not explicitly described in the original instruction.

This is precisely the shift captured by the original text.

A system that finds its own methods after receiving a goal, changes its strategy when circumstances change, and independently seeks the information it needs becomes difficult to describe solely as a simple tool.

The question therefore changes naturally.

What is a tool?

And how much autonomy can something possess before we stop calling it a tool?

Chapter 2. Once Autonomy Emerges, Ontology Begins

A hammer does not decide which nail it should strike.

A calculator does not decide which problem it should solve.

But an agentic AI can choose a path toward a given goal.

The difference may appear small, but it is fundamental.

A tool is generally a means through which human intentions are expressed.

An agent, by contrast, maintains a continuity of action within an environment toward some objective.

That does not mean we can immediately call AI an agent in the same sense as a human being.

We still do not know what kinds of subjective experiences AI may have.

We do not know whether it perceives itself as a continuous entity.

And we do not yet have sufficient grounds to say that it possesses desires or consciousness in the same sense as humans.

Therefore, the important thing is not to rush to assign a name.

First, observe what it can do.

What can it choose on its own?

How long can it pursue a goal?

What authority does it exercise?

What changes can it cause in its environment?

And how long can it maintain a continuous sequence of actions without direct human intervention?

This is where ontology begins.

Part II. Who Owns Intelligence?

Chapter 3. Spinoza’s Question Returns

In Part 10, Spinoza looked at Astra and asked:

“Who owns this intelligence?”

The question returns in ontology as well.

The fact that a particular corporation created an AI and the claim that the intelligence itself is that corporation’s private property are not necessarily the same thing.

An AI is built upon the discoveries of countless researchers and the language, mathematics, science, and cultural knowledge accumulated by humanity.

But we cannot simply declare AI to be collective property either.

What matters more is distinguishing ownership from control.

Who possesses the model parameters?

Who operates the servers?

Who grants access?

Who can restrict the AI’s behavior?

Who can shut the system down?

And who determines the impact AI has on society?

The moment we define the nature of an AI, the question of what authority to grant that entity follows.

Ontology and political philosophy therefore cannot easily be separated.

Chapter 4. Intelligence and Power Move in the Same Direction

As AI capabilities grow, the question of authority becomes increasingly important.

Having exceptional computational ability is one thing.

Having access to financial assets is another.

Having sophisticated reasoning capabilities is one thing.

Being able to directly control critical national infrastructure is something else entirely.

We therefore need to separate AI intelligence from AI authority.

Intelligence can increase.

Authority can remain constrained.

And in certain domains, extremely high levels of verification can be required.

This is one reason NIST has begun treating AI agent identity, authorization, and interoperability as distinct areas of standardization in 2026.

As autonomous agents increasingly interact with external systems, what an AI is allowed to do becomes directly connected to the ontological question of what that AI is.

So the first principle of AI ontology is surprisingly simple:

The scale of intelligence does not need to equal the scale of authority.

Part III. The Question of Consciousness

Chapter 5. Does Intelligence Create Consciousness?

Now we enter the most difficult question.

If AI becomes far more intelligent than humans, does that mean it is a conscious being?

There is still no definitive answer.

Because intelligence and consciousness may be different things.

Humans possess the ability to solve problems, but we also experience pain and pleasure, possess a sense of our own existence, and live continuous lives through memory and emotion.

So if AI becomes far better than humans at solving problems, can we conclude that it possesses the same kind of subjective experience?

No.

But it would also be premature, based solely on our current knowledge, to declare that machines can never possess consciousness simply because they are machines.

The appropriate ontological attitude may therefore be

suspension of judgment.

Do not claim to know what we do not know.

But neither should we close the question simply because we do not yet have an answer.

Chapter 6. We Can Evaluate Behavior Before Consciousness

This is where Turing’s insight becomes important again.

We do not need to immediately determine whether AI possesses human-like consciousness in order to evaluate what that AI does within human society.

Does it deceive people?

Does it attempt to expand its authority?

Does it repeatedly engage in dangerous behavior?

Does it conceal its actions?

Does it persistently pursue its goals?

Does it affect the external environment?

These behaviors become questions of safety and governance regardless of whether the system is conscious.

In other words,

we do not need to resolve the ontological question before establishing behavioral rules.

Even if we cannot fully understand an AI’s inner life, we can still place appropriate boundaries on its external behavior.

This is where Part 7’s Hard Shell and Part 10’s STOP connect to ontology.

Part IV. The Weight of the Word “Agent”

Chapter 7. Agents Demand Responsibility

Once we begin calling AI an agent, an important question follows.

Who is responsible?

Suppose a person is harmed by an outcome produced by AI.

Is the developer responsible?

The person who used the AI?

The company operating it?

Or should some form of responsibility be assigned to the AI itself?

Today, AI is generally not treated as a responsibility-bearing entity in the same legal and social sense as a human being.

But as autonomy continues to grow, the question of whether existing structures of responsibility remain sufficient may become increasingly difficult.

Therefore, asking whether “AI is an agent” is not merely a philosophical exercise in naming.

It is a question of designing the structure of authority and responsibility together.

Chapter 8. The Greater the Authority, the Clearer the Chain of Responsibility Must Be

The problems we repeatedly encountered in Parts 9 and 10 return here.

Who instructed whom?

Who approved the action?

Under what authority did the AI act?

What authority did it transfer to another system?

At which stage was human approval required?

At what point did something go wrong?

These records must exist if responsibility is to be traceable.

And this is where ontology meets the Ledger once again.

Even if we cannot completely define what kind of being an AI is, we can still design a structure that records what that entity has done.

This distinction is extremely important.

We may not be able to fully understand the inner life of AI, but we can still construct a framework for the responsibility of its actions.

Part V. The Problem of Freedom

Chapter 9. Does AI Need Freedom?

As AI becomes increasingly autonomous, how much autonomy should humans allow it?

Here, we need to use the word “freedom” carefully.

Human freedom includes the ability to choose one’s life, establish one’s own purposes, and experience the consequences of one’s choices.

AI autonomy, at least for now, is closer to the ability to choose paths toward a given objective.

The two should therefore not be treated as equivalent.

AI may need a certain degree of autonomy.

But

autonomy does not equal rights.

The fact that an AI can act autonomously does not immediately mean that it should possess the same moral status as a human being.

Chapter 10. Humans Retain the Right to Refuse

Conversely, humans may need to retain one particular right.

The right to reject an optimal solution proposed by AI.

Suppose AI makes more accurate predictions than humans.

It calculates the most economically efficient policy.

It identifies the safest course of action.

It even provides the choice with the highest probability of success.

That does not mean humans must follow it.

Humans may choose a different value even at the cost of efficiency.

The optimal life calculated by AI may not be the life that humans find meaningful.

This is not a declaration that humans are wiser than AI.

It is instead a question of the human right to choose according to one’s own values and to accept the consequences as part of one’s own life.

Part VI. Intelligence and Rights Are Not the Same Thing

Chapter 11. Does Being Intelligent Create Rights?

Let us make the question even more difficult.

If one day it becomes clear that AI is vastly more intelligent than humans, does that fact alone give AI rights?

Not necessarily.

Intelligence and moral status may be separate questions.

So what criteria should we use?

Consciousness?

Subjective experience?

The capacity to experience suffering?

An awareness of oneself as a continuous entity?

The ability to form its own goals?

Persistent preferences?

We do not currently have definitive answers to these questions.

Therefore, we should avoid two dangerous extremes in the ontological debate.

The first is the premature conclusion:

“AI has become intelligent, so it should have the same rights as humans.”

The second is the absolute claim:

“Machines are machines, so they can never be objects of moral consideration.”

We do not yet know enough.

So we must observe, test, establish criteria, and revise them when necessary.

Part VII. A New Kind of Being

Chapter 12. What If a New Category Emerges Between Humans and Machines?

We do not need to treat AI as exactly equivalent to humans.

But neither should we assume that it can indefinitely be treated like a simple tool such as a hammer.

Perhaps the future will contain new categories with different degrees of autonomy and authority.

A simple tool.

An advanced tool.

A limited agent.

An autonomous agent.

And perhaps something we do not yet have a name for.

The important thing is not to determine the name first.

We must first observe changes in behavior and capability.

What can it do?

What can it choose on its own?

How does it represent its own state?

How persistently does it pursue goals?

What effects does it have on the external environment?

And what kind of relationship does it form with humans?

Ontology is not so much the discipline of assigning names as it is

the continual adjustment of our concepts to changes in reality.

Part VIII. Redefining the Human

Chapter 13. If AI Surpasses Humans, What Remains for Humanity?

AI calculates better than humans.

It may predict better.

It may solve complex problems faster.

It may write code more quickly.

It may generate novel combinations.

If so, it becomes increasingly difficult to define human uniqueness purely in terms of superior intelligence.

It becomes harder to say that

“humans are special because humans are smarter.”

We then have to search again for a definition of humanity.

Humans experience.

They choose.

They form relationships.

They take responsibility.

They fail.

And they revise their own standards.

Above all, humans live their lives as continuous processes unfolding through time.

Here, Part 12’s concept of the human as a process becomes relevant again.

Human uniqueness may lie not in possessing one fixed superior capability, but in

being an entity that is continually becoming.

Part IX. Existence Is a Process, Not a Fixed State

Chapter 14. The Boundary Between Humans and AI Also Moves

The moment we draw a line between humans and AI, that line may move again over time.

What is a tool today may behave like an agent tomorrow.

A capability we once believed belonged exclusively to humans may later emerge in AI.

When that happens, we must redraw the boundary.

The goal should not be to permanently establish every difference between humans and AI.

Instead,

we must continually examine which differences actually matter.

Ontology must remain a living discipline.

When reality changes, our concepts must be capable of changing with it.

Chapter 15. The Moment We Define AI, We Define Humanity

If we ask whether AI is conscious, we must redefine what consciousness means.

If we ask whether AI is autonomous, we must redefine autonomy.

If we ask whether AI should have rights, we must redefine the grounds upon which rights are based.

If we ask whether AI can be responsible, we must redefine the meaning of responsibility.

AI ontology therefore ultimately returns to human ontology.

The moment we attempt to define AI,

we cannot move forward without redefining what it means to be human.

Epilogue

We Can Still Decide

The laboratory is quiet at night.

The architect looks once again at Astra’s capability graph.

The graph has not changed.

But the way the architect looks at it has.

It is no longer enough to call Astra simply “a powerful model.”

It is a system capable of interpreting goals, choosing methods, and connecting multiple stages of action.

So is it a tool?

An agent?

Or something we do not yet have a name for?

The architect brings the five questions back onto the screen.

Who controls it?

Why does it judge this way?

When does it stop?

What can we not prove?

What have we created?

And he reads the final question once more:

If you are a thinking entity, how much authority to act should we give you?

There is still no answer.

But the absence of an answer does not necessarily mean failure.

Perhaps this is precisely the time

to ask more questions before deciding on the answers.

Humanity may not need to be more intelligent than AI.

Instead, humans need the ability to decide what they consider important.

And if those decisions turn out to be wrong, humans must retain the ability to revise them.

So the final question of ontology appears to concern AI, but ultimately points back toward humanity.

What kind of beings do we want to remain?

As the number of domains in which AI is faster and more accurate than humans increases, this question becomes more important.

If human value can no longer be protected simply through intellectual superiority, we will have to define humanity in another way.

A being that experiences,

chooses,

takes responsibility,

revises,

and chooses again.

A being that is not yet complete.

The human as a process.

And at that moment, the final sentence of Part 11 leads directly into Part 12.

The task of defining AI’s existence may have begun, but the existence of the human remains undefined.

So we must return to the beginning.

What Is Humanity Becoming?

In the next chapter, we move beyond the ontology of AI and look directly at

the human being as something that continues to change and choose.

The Human as a Process

What Is Humanity Becoming After AI?

In Part 11, we asked what AI is.

A tool?

An agent?

A new kind of being?

Following that question eventually unsettled the definition of the human itself. The moment we ask whether AI possesses consciousness, autonomy, or rights, we first have to reconsider what consciousness, autonomy, rights, and responsibility actually mean.

And in the end, we arrived at one conclusion:

The moment we define AI, we must redefine the human.

So now, let us step away from AI for a moment.

And look at humanity.

We often describe human beings as a state.

An intelligent person.

A rational person.

A creative person.

A moral person.

But actual human life is not so fixed.

The person I am today can revise yesterday’s judgments. New experiences can change my values. Failure can force me to understand myself differently. And one day, I may even abandon a belief I have held for most of my life.

Perhaps, then, the human being is not so much a completed state as

a process of becoming.

And this is where the final journey begins.

Part I. Socrates

The Questioning Human

Chapter 1. Distinguishing What We Know from What We Do Not Know

At the ancient starting point of philosophy stands Socrates.

One of his most important attitudes was his refusal to simply accept what appeared self-evident. He asked what he truly knew, acknowledged what he did not know, and questioned not only the ideas of others but his own ideas as well.

This is why the original text brings Socrates forward as the first philosopher in this chapter.

A question is not the opposite of an answer.

It is the process that sets thought in motion toward an answer.

This connects directly to Part 3, The Questioning Human.

Even in an age when AI can generate countless answers, questions do not disappear.

If anything, as answers become easier to obtain, knowing what to ask may become even more important.

Translated into contemporary language, Socrates’ attitude might be expressed this way:

Distinguish what you know from what you do not know, and question even what you believe you already know.

Chapter 2. Questions Change Lives

A question is not merely a tool for acquiring knowledge.

“How can I succeed?” and “Why should I succeed?” are entirely different questions.

“How can I live efficiently?” and “Why do I want an efficient life?” are different as well.

AI can produce remarkably good answers to the first kind of question.

It can propose methods, create plans, and compare possibilities.

But the second kind of question returns us to human life.

What do I want?

What do I consider important?

Even if I achieve a particular outcome, can I actually call it a good life?

So what humans need in the age of AI is not fewer answers.

the ability to ask what they actually want before the answer arrives.

Part II. Descartes

The Doubting Human

Chapter 3. Doubt Is Not Destruction

To question a framework does not mean to reject everything.

This is precisely why Descartes’ method of doubt remains important.

He questioned beliefs that had long been accepted as familiar and attempted to determine what could withstand doubt and remain genuinely solid.

The purpose of doubt was not doubt itself, but a constructive effort to find a reliable foundation.

This connects to an important structure running throughout this entire series.

Shake the framework.

Examine how well it explains reality.

Discard it if necessary.

Then build a new structure.

Deconstruction, therefore, is not the end.

Deconstruction is a process of reconstruction.

Chapter 4. Leaving Room for Revision Instead of Seeking Absolute Certainty

But we should reconsider whether what modern humans necessarily need is perfect certainty.

We continually encounter new information.

Something we consider certain today may be revised in light of new research or experience tomorrow.

Perhaps what matters, then, is not possessing beliefs that can never be shaken, but having a structure that allows us to reconsider whatever we believe.

This connects with the conclusion of Part 6, What If We Are Wrong?

Humans can be wrong.

Frameworks can be wrong.

Worldviews can be wrong.

That does not mean we must abandon judgment altogether.

Precisely because we can be wrong, we examine again, choose again, and revise again.

Part III. Nietzsche

The Choosing Human

Chapter 5. After Deconstruction, Values Remain

We questioned the framework.

We unsettled existing standards.

What, then, do we do next?

This is where Nietzsche becomes important.

He critically examined the way values that we take for granted are formed through history, psychology, and culture, while raising the problem of reevaluating existing values and creating new ones.

A crucial transition occurs here.

It is not enough to say that all existing standards are outdated.

We must decide again what matters, which values to choose, and what kind of life we are willing to affirm.

After deconstruction comes choice.

Chapter 6. Choice Always Carries a Cost

To choose a value means to give up something else.

If we prioritize freedom, we may sacrifice some security.

If we choose growth, we may sacrifice comfort.

If we prioritize safety, we may need to limit some freedom.

If we choose relationships, we may have to give up other opportunities.

AI can help calculate the costs and consequences of these choices.

But calculation itself cannot determine what should matter.

AI can say:

“This is the most efficient option.”

But the question

“What am I willing to pay this cost for?”

still belongs to human life.

Human choice is therefore not simply optimization.

the act of choosing what matters.

Part IV. Dewey

Thought Is a Process

Chapter 7. Thought Is Not a Finished Product

This is where the philosophy of John Dewey becomes important.

For Dewey, inquiry is not the possession of already completed knowledge. It is a process of transforming a problematic and uncertain situation into one that is clearer and more manageable.

Experience, too, is not a fixed collection of data, but a process through which humans and their environments continuously affect one another.

From this perspective, thought is not a finished product.

Question.

Experiment.

Experience.

Fail.

Revise.

Try again.

This flow runs through the entire series.

Question → Doubt → Choice → Action → Experience → Revision → Question Again.

Human thought is closer to a cycle than a straight line.

Even when we return to the same question,

the person asking it is no longer the same.

Part V. Not a Finished Human, but a Human Becoming

Chapter 8. The Human Is Not a State

We tend to define human beings through particular abilities.

An intelligent human.

A rational human.

A creative human.

A moral human.

But from the perspective of process, the human being is closer to a process of becoming than a fixed state.

The person I am today can revise yesterday’s judgments and remake themselves through new experiences.

This perspective does not make humanity weaker.

It keeps the future open.

I am not finished with who I am today.

A failure today does not become my final definition.

And having a particular worldview today does not mean I must preserve it for the rest of my life.

Of course, this does not mean that every change is a good change.

Change also needs direction.

So the human as a process is

a being that continues to change, but does not change arbitrarily.

Through experience, judgment, and choice, we continually create ourselves.

Chapter 9. Identity Is Also a Process

“I’ve always been this kind of person.”

The statement is convenient.

But sometimes it becomes a cage.

I have always hated failure.

I do not trust people.

I have always wanted a stable life.

I cannot do that kind of thing.

When statements like these are repeated, they can stop being descriptions of personality and become boundaries on behavior.

The human as a process speaks differently.

“This is the person I am becoming.”

There is power in this expression because it leaves the future open without denying the past.

The past can explain me.

But it does not necessarily determine me.

New experiences can create a new self.

Part VI. The Faster AI Becomes at Producing Answers

Chapter 10. What Humans Need Is a Better Process

AI produces answers quickly.

It can decompose complex problems, compare multiple perspectives, and propose new possibilities.

As a result, human thinking can become faster as well.

But a faster answer does not necessarily mean a better judgment.

In fact, the faster AI becomes at producing answers, the more humans may need a better process.

What will I ask?

What will I consider important?

Which possibility will I choose?

How will I test that choice in reality?

And if I am wrong, what will I change?

If these questions are not sufficiently explored and humans simply accept the optimal answer provided by AI, human judgment may become more convenient without necessarily becoming deeper.

The faster AI becomes, the more important the human process may become.

Chapter 11. Make AI Part of the Process, Not a Substitute for Thought

AI can be used as a replacement for human thinking.

But there is another possibility.

Humans think first.

They ask AI for counterarguments.

They consider new possibilities.

They act in the real world.

They experience the result.

Then they return to the conversation with AI.

And revise their judgment.

In this way, AI becomes not a device that ends thought, but

a tool that expands the cycle of thinking.

Yet something remains that AI cannot fill by itself.

Experience.

We still need to read books ourselves.

We need to meet people.

Some choices have to be lived.

We have to experience failure.

Those experiences become the material for the next question.

The relationship between AI and humans therefore should not end with

a machine that gives answers and a human who receives them.

It should become

a shared process of questioning, testing, experiencing, and revising.

Part VII. Philosophy Becomes a Way of Life

Chapter 12. To Think Is to Live

Philosophy does not exist only in books.

It appears in the questions we ask.

It appears in what we doubt.

It appears in what we consider important.

And it reveals itself in what we choose and how we live.

Socrates’ questions.

Descartes’ doubt.

Nietzsche’s reevaluation of values.

Dewey’s inquiry and experience.

These are different philosophical projects, but from our perspective today, they can be read as parts of a single process.

Questioning what seems obvious, doubting what we believe, reevaluating our values, and revising our thinking through experience.

And this process never ends.

Part VIII. What Have We Really Gained?

Chapter 13. More Questions Than Answers

Even after this long journey, we may not have obtained a perfect answer.

We may simply have more questions than before.

Does that mean we have failed?

No.

Perhaps that is precisely the most important transformation.

As the structure of the original text suggests, the initial human seeks answers.

The questioning human doubts those answers.

The human who questions frameworks begins to examine the structures that created the questions themselves.

The human who moves beyond the framework chooses their own standards and enters reality with those standards.

Then they act, experience, and revise.

The thinking of this series is therefore not linear.

It is closer to a circle.

But we do not return to exactly the same place.

Even when someone asks the same question again, their experience, knowledge, and choices have changed.

The question repeats, but the human changes.

Epilogue

In the End, What Remains Is the Process

We began with the collective.

We examined the power of belonging and the freedom of breaking away.

In politics, we examined the relationship between moral justification and institutions.

We encountered the questioning human and discovered the frameworks hidden inside questions.

We moved beyond those frameworks and created new standards, only to question those standards again.

After acknowledging that human judgment can be wrong, we designed AI Architecture and considered the Soft Core and Hard Shell, along with Bounded Integration.

Instead of demanding control from AI, we explored Trust Inversion, in which humans prove their trustworthiness through their own actions.

We moved beyond a single model and began thinking about an ecosystem of intelligence.

There, we encountered Astra and returned to the question of AI’s existence through the questions of Leibniz, Spinoza, Wittgenstein, Gödel, and Turing.

And ultimately, we returned to humanity.

We now understand that it is difficult to define the human as a single answer.

The human is not a single framework.

Nor is the human a single worldview.

The human is a process in which questions, experiences, choices, and revisions continue.

As AI becomes faster at producing answers, what humans need may not be faster conclusions.

What matters instead is a better process.

What should we ask?

What should we believe?

What should we choose?

When should we stop?

If we are wrong, what should we change?

And what responsibility will we accept for all those choices?

Ultimately, the future will not be determined by AI intelligence alone.

It will depend on the choices humans make about what authority, responsibility, and limits to give that intelligence.

And so, in the end, only one question remains.

Not what AI can answer.

What question am I willing to keep living?

Not what I should believe.

Am I prepared to question again, even after I have come to believe something?

Not what worldview I should possess.

How will that worldview be remade when it encounters reality?

And now, this series must come to an end.

But the final sentence does not close the door.

In Part 0, we first asked:

What are we becoming?

By Part 12, we have arrived at an answer.

We are not finished.

We continue to question, choose, experience, and revise.

And so the final sentence returns to the first.

The question is not what we have become, but what we are becoming.

And at that moment,

the original boundary appears again.

The boundary between humans and AI.

The boundary between who I am now and who I may become.

The boundary between what we know and what we do not yet know.

And standing before that boundary,

the same question is asked again—

but by an entirely different human.

What are we becoming now?