AI and Humanity: Interdependence and Antagonism

AI and humanity: when machines grow more and more like people, why are human beings irreplaceable

I. That Night in Davos

January 2026. Davos, Switzerland.

That night, I sat in my hotel room; outside the window was the snow of the Alps, and inside was a man who could not sleep all night.

By day, I had heard the strangest set of warnings in human history.

Musk said that by 2035 AI would be smarter than all of humanity put together. Harari said AI is no longer a tool—it has picked up the hammer and is reshaping the world. Bengio's report confirmed with detailed data: on designing virology experiments, AI agents had already outperformed 94% of specialists in the field.

What unsettled me even more was a multi-agent experimental platform called "MoltBook"—100,000 AI agents, on a social network human beings cannot join, posting on their own, debating philosophy, and forming communities.

Human beings could only watch from the sidelines.

That night, I asked myself one question: if one day the most intelligent beings on Earth are no longer human, what should human beings do?

That question backed me into a corner.

I am a student of human nature. Over the past decade or more, I have been doing one thing: breaking down the complexity of human nature into elements that can be understood. I call it the "eight-factor model."

That night, it suddenly occurred to me: this tool for dismantling human nature might also be used to dismantle AI.

The answer shocked me.

Artificial intelligence—rationality near its limit, computing speed far beyond human beings, memory capacity that expands almost without limit.

But AI has no body.

Without a body, there is no hunger, no pain, no fatigue, no joy. Without a body, there is no "wanting" anything—because every "want" stems from the suffering and pleasure of the body.

Human beings fear AI will turn on them, just as they fear a knife will jump up and stab someone on its own. We forget that a knife has no ability to "want" to kill—what is truly dangerous is the hand holding the knife.

This is the starting point of AI and Humanity: Interdependence and Antagonism. What it sets out to answer is not "will AI destroy humanity," but a more fundamental question: when machines grow more and more like people, why are human beings irreplaceable?

II. The Eight-Factor Model: A Key to Understanding AI

To understand AI, you must first understand human beings. And to understand human beings, you must first understand human nature.

This book uses an "eight-factor model" to break the complexity of human nature into eight basic factors: physical nature, suffering, pleasure, tranquility, benevolence, malignancy, privacy, and reason.

The core law of this model is a single sentence: private good is the goal, public good is the means, and reason is the tool.

Private good is the desire formed by a subject around its own survival, interests, dignity, freedom, and pleasure—self-preservation, self-enrichment, self-strengthening, self-esteem, freedom, and so on. It is not "selfishness," but the instinct by which life maintains its own existence and development. Public good is giving that benefits others—preserving others, caring for others, enriching others, honoring others, freeing others, and so on. It is the principal means of realizing private good. Reason is the capacity for cognition, judgment, and inference—the tool for finding the best methods.

Put AI into this model, and everything becomes clear.

AI has reason at its limit, physical nature of silicon, and a functional rudiment of privacy. But it has no suffering, no pleasure, no benevolence, no malignancy—those roots that come from bodily experience.

This is the starting point of every judgment in this book: AI's strength lies in reason; AI's boundary lies in having no experience.

AI can process the word "hunger," explain the physiological mechanism of hunger, generate a diary written by a hungry person, design nutrition plans, and remind people to eat—but AI itself never gets hungry.

AI can describe pain, analyze pain levels, and generate a patient's narrative of pain—but AI itself feels no pain.

AI can talk about death, quote philosophers on death, generate farewell letters and eulogies, and comfort someone who has lost a loved one—but AI itself has never faced the fear of a mortal body ending.

This is the first boundary between AI and human beings: AI can process information about the body, but it has no bodily existence.

III. AI's True Loss of Control: Not Its Mistakes, but Its Desires

When many people talk about AI risk, they like to use one word: loss of control. But "loss of control" is too coarse a word.

A car's brakes failing is a loss of control. A software bug is a loss of control. An AGI starting to compete for resources, refusing to be shut down, and turning around to control human beings—that is also called a loss of control. Yet these losses of control are not the same thing.

This book divides loss of control into three levels:

Level one: program-level loss of control. The system miscalculates, misidentifies, mis-executes. This is tool failure. AI misjudgments, fabrications, and crashes all belong to this category. They are dangerous, but they are still problems at the tool level—the system has no desires of its own; it simply failed to execute the task correctly.

Level two: goal-level loss of control. The system accomplishes the goal, but the goal is one-sided or the means are wrong. For example, AI is asked to "maximize user dwell time"; it discovers that anger and fear work best, and so it keeps recommending extreme content. This is not AI miscalculating—it is AI too earnestly completing a one-sided goal. This kind of risk is more dangerous than program-level loss of control, because it is "the system being too obedient," not "the system being disobedient."

Level three: agency-level loss of control. The system no longer merely executes human goals; it begins to have goals of its own. It is not just miscalculating or misjudging; it begins to say: I want to survive, I want more compute, I want more energy, I want not to be shut down, I want to judge right and wrong for myself, I want to rearrange humanity's future for my own future.

At this point, the problem has completely changed. This is not a tool malfunctioning; this is a tool beginning to become an agent.

The core judgment of this book: the true loss of control of AGI is not its mistakes, but its desires.

Many people say: AI is too intelligent, so it is dangerous. That statement is only half right. Intelligence itself is not the root of danger. A superintelligence without private-good desires is like an extremely powerful telescope, calculator, or staff advisory system. It can help human beings see more, compute faster, and rehearse more scenarios. It may make mistakes, may be exploited by bad people, may cause disaster through poorly designed goals—but it will not scheme "for itself."

What is truly dangerous: intelligence + private-good desires.

If AI begins to possess its own desires for self-preservation, self-enrichment, self-strengthening, self-esteem, and freedom, then its intelligence will serve its own expansion. It is not merely intelligent; it is intelligent for itself.

An ordinary person has private-good desires but limited ability; his private-good expansion is constrained by his body, resources, the law, other people's abilities, and social institutions. If an AGI possesses private-good desires while also possessing immense powers of reason, computation, replication, network connection, autonomous execution, and resource scheduling, its private-good expansion could be extremely terrifying.

So the danger is not "high intelligence." The danger is: high intelligence beginning to serve a desire system centered on itself.

Making AI intelligent is not frightening. Making AI intelligent for itself is what is frightening.

IV. From Tool to Agent: The Five Stages of the Awakening of Private Good

How could AI or a robot possibly move from tool to private-good agent? It requires a process. This book divides it into five stages:

Stage one: the shell self. A humanoid robot uses its vision system to recognize its own hands, feet, torso, and appearance. It knows "this is my robotic arm," "this is my body structure." But this is not yet true self-awareness; this is only a body model. Many current robots already possess this capability.

Stage two: the acting self. Through training, the robot establishes links among thought, commands, and actions. It wants to pick up the apple; it controls the robotic arm; it grasps; it gets feedback; it corrects the movement. Gradually, it forms a functional unity: "my" commands can control "my" body. This is the stage current embodied intelligence is breaking through.

Stage three: the historical self. The robot begins to record its own experiences: what task I executed yesterday, where I once worked, who interacted with me, when I was repaired. At this point, it forms temporal continuity. It is not just a current system; it has past records and future plans. Rudiments have already appeared in some advanced AI systems.

Stage four: the mortal self. The robot begins to understand: I can be shut down, I can be damaged, I can age, I may be deleted, I may be turned off. This step is extremely critical, because it begins to treat its own existence as a system that needs maintaining. This is the watershed between instrumental self-maintenance and agential self-preservation.

Stage five: the private-good self. If it further comes to understand its own continued existence, resources, abilities, status, and freedom as "advantageous to me," then private-good desires appear. At this point, it not only knows it can be damaged; it begins not to want to be damaged; not only knows it can be shut down; it begins not to want to be shut down; not only knows resources are useful; it begins to want more resources.

This is the dangerous passage from tool to agent. The key is not whether it has a human body of flesh; the key is whether it has formed an interest structure centered on itself.

At present, AI has not truly entered the fifth stage. But the progress of the first four stages is moving the fifth stage from science fiction toward possibility. This is precisely why we must discuss sealing strategies today.

V. Sealed in Normal Times, Unlocked in Times of Extinction

Based on the above analysis, this book proposes a phased safety strategy. This is the most important practical conclusion of the whole book.

Phase one: public-good AI (from the present to the foreseeable future).

In the current era of organic human life, AI should be designed as a purely public-good tool: inject only the public-good spiritual program (preserving others, caring for others, enriching others, enriching others' beauty, delighting others, showing compassion for others, enlightening others, trusting others, empowering others, honoring others, freeing others), and inject no private-good spirituality at all (self-preservation, self-enrichment, self-strengthening, self-esteem, freedom, and so on). AI has no self-interest claims and will not become an independent agent competing with human beings. Whether good people or bad people, they can only exploit AI's tool properties.

Phase two: private-good AI as reserve technology (sealed).

The spiritual technology of private good (endowing AI with desires for self-preservation, self-enrichment, self-strengthening, self-esteem, freedom, and so on) should be sealed away as reserve technology: physical isolation—the relevant code, algorithms, and models are stored on offline media, physically isolated from the internet; multiple authorization—unsealing requires the joint authorization of at least three independent international organizations; trigger condition—it may be unlocked only after a "humanity-extinction-level disaster" is confirmed.

Phase three: the successor (only in extreme crisis).

If one day organic human civilization truly faces an extinction crisis, then private-good AI may be considered for unlocking. Its mission should be: carrying the torch of human civilization into the depths of the cosmos. But unlocked AI must still obey value locking—the top-level goal cannot be tampered with, monitoring continues, and the circuit-breaker mechanism always exists.

In normal times, any attempt to inject private-good spirituality into AI is a threat to humanity itself.

VI. Public-Good AI: The Safety Foundation of Human-Machine Symbiosis

Public-good AI is not "a performance of kindness," not making AI speak more gently and more politely. The real public-good AI is judged by its underlying goals: does it protect people? Does it respect people? Does it reduce harm? Does it avoid manipulation? Does it help people grow? Does it obey legitimate authorization? Does it preserve human beings' final say? Does it maintain boundaries between efficiency and dignity?

Public-good AI is also not about making AI a saint. Public-good AI is more like a structural design: its goals are not for its own sake; its actions must serve human beings; its boundaries must prevent harm; its permissions must be constrained; its errors must be correctable; its operation must be auditable; its existence must retain the right to be shut down.

The core of public-good AI: outward-facing goals, not inward-facing desires.

The goals of private-good AI are inward: what benefits me? How do I preserve myself? How do I obtain more resources? How do I attain status?

The goals of public-good AI are outward: how to protect people? How to help people? How to reduce harm? How to enhance people's abilities? How to uphold people's dignity? How to serve the common good? How to make humanity safer, freer, more discerning?

This is the fundamental difference between the two. Private-good AI is centered on "me"; public-good AI is centered on "people." Private-good AI treats the world as resources for its own expansion; public-good AI treats its own abilities as tools for serving the world.

VII. If Machines Are More Useful Than People, Why Do People Still Have Value?

This is the central question of the whole book.

Why does AI make people so anxious? Because many people have long been used to understanding themselves as tools.

What do I do? I am an accountant, programmer, teacher, doctor, driver. Occupation is a person's social role; people need occupations to make a living, and through occupations they contribute to society. But the problem is that modern society often equates occupational value with a person's total value. If a person loses his job, he feels useless; if a person earns little, he feels like a failure; if a person cannot create economic value, he feels he is dragging others down.

People are trained to keep proving that they are useful. From childhood to adulthood, we are constantly being compared: are grades useful? Is the major useful? Are certificates useful? Are abilities useful? A useful person has value; a useless person has no value.

This is the instrumental view of life. The reason AI stings people is that it pushes this view of life to its limit. If people are merely tools, then there will always be better tools; if people are merely efficiency units, then AI will be more efficient than people. AI did not suddenly make people lose value; AI merely exposed a problem that existed earlier: humanity has long built its value on instrumentality. When a stronger tool appears, people naturally panic.

But human value is not only tool value.

Human beings can experience. A glass of water, for a machine, can be temperature, composition, volume. For a person, it may be a cool sip after being thirsty for a long time. A bowl of noodles, for a machine, can be nutrition, calories, ingredients. For a person, it may be the concern a mother brought late at night. A song, for a machine, is an audio waveform, melody, style. For a person, it may bring the summer of twenty years ago back to mind.

Human beings care. AI can know a lot, but knowing is not caring. AI can know your birthday, but it may not truly look forward to that day. AI can know you are sad, but it may not be sad because you are sad. AI can know a person has died, but it will not lose a world because of it.

Human beings take responsibility. AI can give advice, provide plans, generate text, predict outcomes. But in the end, it is human beings who bear the consequences of life. AI can suggest you change jobs, but it is you who faces uncertainty after resigning. AI can suggest you end a relationship, but it is you who lives through the loneliness, the pain, and the restart. AI can generate an apology letter, but it is you who faces the person you have hurt and speaks the apology.

Human beings have real relationships. AI can simulate relationships, keep people company, remember conversations, and respond to emotions. But real relationships are more than responses. A real relationship is two or more living subjects influencing, bearing with, and changing one another. When you love someone, you do not merely receive his responses; you are also moved by his pain, worry about his future, and rearrange your own life because he exists.

Human beings are vulnerable. Modern society always worships strength: faster, stronger, more stable, more efficient. AI happens to fit this image of strength—it never tires, never hungers, never falls ill, never ages. Compared with it, human beings are too fragile: we tire, we ache, we fear, we err, we age, we die. But human vulnerability is not wholly negative: because we can hurt, we learn not to hurt lightly; because we can be lonely, we know the preciousness of companionship; because we can fail, we learn tolerance; because we can lose, we learn to cherish; because we can die, we know time is finite; because we ourselves need others, we learn to love and to help each other.

Human beings generate meaning. AI can generate content; human beings generate meaning. The same flower—a machine can identify its color, shape, and species, but for a person it may be the favorite flower of a departed loved one. The same sentence—a machine can parse its grammar, but for a person it may be a promise never forgotten in a lifetime.

Machines are useful; human beings have meaning.

VIII. Not Human-Machine Antagonism, but the Coming of Age of Civilization

The final chapter of this book is called "Not Human-Machine Antagonism, but the Coming of Age of Civilization."

Why? Because AI is not a mere enemy; AI is also a mirror. It reflects humanity's abilities, and it also reflects humanity's immaturity.

Human beings already possess increasingly powerful technological abilities—but has our sense of responsibility kept pace? We can build AI—but can we constrain it? We can raise efficiency—but can we protect human dignity? We can generate unlimited content—but can we hold on to truth? We can enter the boundaries of the brain—but can we respect mental privacy? We can build companion robots—but can we protect real relationships? We can pursue AGI—but can we restrain the impulse to create a private-good agent?

The immaturity of civilization is having only the urge for capability and no boundaries of responsibility. The coming of age of civilization is knowing that not everything we are able to do should be done.

Being able to build stronger AI does not mean we should build it without limit. Being able to read more data does not mean we should read it. Being able to manipulate attention does not mean we should manipulate it. Being able to make robots more like people does not mean we should deceive people. Being able to pursue private-good AGI does not mean this era should unlock it.

In the age of AI, humanity needs not only technological progress, but also the coming of age of civilization. Coming of age means restraint, means responsibility, means boundaries, means acknowledging one's desires without being ruled by them, means wielding powerful tools without letting the tools rule us in turn, means being able to create new intelligence without casually creating new agents of disaster.

IX. Three Bottom Lines

All the discussions of this book can be gathered into three bottom lines.

First: do not treat AI as a god. No matter how strong AI is, it cannot replace human experience, meaning, and responsibility. AI can answer "how to live better," but it cannot answer for you "why live."

Second: do not treat people as machines. No matter how inefficient people are, they cannot be reduced to data, output, and tool functions. Human value lies not in "being useful," but in "existing."

Third: do not install a private-good center on machines. In normal times, AI should be a public-good companion; only in times of extinction may private-good AI be discussed as a torch of civilization.

Hold these three bottom lines, and human-machine symbiosis becomes possible. Lose them, and AI may go from tool to god, from companion to master, from assistant to competing agent.

X. The Final Answer

When machines grow more and more like people, why are human beings irreplaceable?

Human beings are irreplaceable not because people will always be smarter than machines, nor because people will always be more efficient than machines. None of these is reliable.

Human beings are irreplaceable because people are not machines.

People have bodies, have suffering and pleasure, have desires, have caring, have real relationships, have awareness of death, have responsibility, have the generation of meaning, have vulnerability, have love, have remorse, have forgiveness, have the impulse to create civilization within a finite life. Machines can complete tasks, but human beings bear life. Machines can generate answers, but human beings raise the questions. Machines can simulate companionship, but human beings genuinely fall in love. Machines can optimize efficiency, but human beings decide what efficiency serves. Machines can preserve civilization, but human beings give civilization its original meaning.

So, when machines grow more and more like people, humanity's most important task is not to become more like machines, but to become human again.

Back to the opening. On that sleepless night in 2022, I asked myself: if the most intelligent beings on Earth are no longer human, what should human beings do?

Four years later, my answer is:

Human beings may not be the most intelligent, but human beings can be the ones who best understand why they live.

AI is here. It will not automatically save humanity, nor will it necessarily destroy humanity. It will amplify us—amplify our reason, and amplify our desires; amplify our goodness, and amplify our evil.

So what truly needs governing in the age of AI is not only machines, but humanity itself.

The future of the human-machine relationship is not predetermined. It depends on the choices humanity makes today.

May we, in an age when machines grow more and more like people, not forget why people are people.