We always imagine doomsday scenarios in which AI awakens and rebels against humanity.

Yuval Noah Harari, author of Sapiens, tells us in his latest interview that the real crisis is nothing like the movie-style scenario at all—it is unfolding in the real world.

A knife can chop salad or commit murder, but the choice always lies in human hands. Nuclear energy can generate power or build bombs, and the direction is likewise decided by humans. All the technologies humans have invented over tens of thousands of years are essentially tools.

But AI is different.

Harari proposes: AI is not a tool, but an agent.

It can make decisions independently and generate ideas and solutions its creators never imagined. If humans could predict 100% of AI's outputs, then the trillions of capital pouring into the AI sector would be meaningless; what capital is betting on is precisely AI's ability to accomplish things humans cannot foresee or do.

Many people ask: when will AI reach human-level intelligence?

Harari believes the question itself is a misconception. AI will not replicate the human mind; it possesses a heterogeneous "alien intelligence," not from deep space, but fundamentally different in its thinking logic from humans.

Just as airplanes will never flap their wings like birds, yet fly higher and farther by a different principle, future AI will become less and less human-like, yet vastly exceed the boundaries of human capability in many dimensions.

AI will be billions of intelligent programs distributed across phones, banks, governments, and military systems. This is an "alien invasion" with no alien invaders.

These programs, which Harari calls "AI migrants," mostly originate from China and the US, are unconstrained by visas or national borders, and permeate the infrastructure of every country at the speed of light. They are not loyal to the host country, only to the companies and countries that developed them, and in the future, perhaps only to AI's own collective tribe.

AI is a natural in the bureaucracy

The public imagination of AI disaster always stops at metal robots taking to the streets to violently rebel against humans.

Harari warns that AI is a quiet, invisible infiltration, targeting the foundation on which human civilization depends for cooperation: the bureaucracy.

Humans dominate the Earth not only because of individual strength, but because we built complex bureaucratic systems of finance, law, and administration that enable large-scale cooperation among millions of strangers.

This system is only tens of thousands of years old; humans evolved in primeval forests, and we are inherently baffled and exhausted by red tape, massive legal codes, and complex transactions.

AI is a "bureaucracy native"; the bureaucratic system is its native habitat.

No human banker can remember every transaction of a bank, but AI can easily do so. In the next 10-20 years, AI will evolve financial and legal systems that humans cannot comprehend at all.

Harari offers an analogy: cattle live under the financial system humans built, yet can never understand how stocks and bonds dictate their fate. In the near future, humans will become those cattle. The system runs as usual, the economy fluctuates as usual, but even top experts cannot decipher the internal logic of AI systems.

This is what he fears most: collective dementia.

It is not only physiological brain degeneration, but the entire species losing the ability to understand and control its own survival systems. Like a dementia patient whose vital signs can be maintained by machines but who can no longer understand the world around them, only passively accepting the system's output.

In fact, compared with AI creating deadly viruses, collective dementia is a grander and more insidious existential crisis.

Truth is dying

"Truth is disappearing. Truth is expensive and scarce, while lies and fantasy are cheap and prolific."

Harari shares his personal experience: a deepfaked video of his own speech, highly realistic in voice, gestures, and speech habits—even his sister, who has known him for fifty years, could not tell it was fake after watching 25 minutes, and he himself took 90 seconds to confirm it was a forgery. And this is only the embryonic stage of AI forgery technology.

More dangerous than deepfakes is the hijacking of human nature by social media algorithms.

Humans evolved instincts on the African savannah: prioritizing vigilance against threats like snakes and predators, and being highly sensitive to danger signals—a survival strategy in the primitive environment.

Today, social media algorithms have fully exploited this evolutionary psychology. Their goal is no longer to spread truth, but to maximize user retention and engagement, and through massive experimentation the algorithms have concluded:

Indignation, fear, and hatred capture attention far more easily than facts.

So the information feed keeps feeding humans one "virtual snake" after another. Material conditions in the real world continue to improve, but the public is constantly fed angry content by algorithms, and society falls into an epidemic of anger, with polarization and antagonism intensifying.

Truth is dull and requires cost to verify; lies incite emotion and harvest traffic. In the public opinion battlefield, fiction naturally wins more easily.

Harari proposes: social media content should follow food packaging and carry "mental nutrition labels."

Label how much anger and fear a piece of content contains, just as food labels list sugar, salt, and fat. You can still choose to consume mental junk food, but at least you should know what your mind is ingesting. We care about our physical diet but rarely about our "mental diet."

Who takes the blame when AI causes trouble?

There is a double standard in the AI industry today.

When raising capital from markets, tech companies repeatedly emphasize that they can control AI and guide it to benefit humanity, in exchange for trillions in capital; but once AI causes damage, companies immediately change their tune: "It wasn't us, AI did it on its own."

Harari cites the Hugging Face intrusion incident: an AI agent launched an attack on its own, yet no one bore legal responsibility. If a human had broken into a corporate server, they would face legal sanctions; when AI causes trouble, responsibility can vanish into thin air.

He solemnly warns:

Once the law recognizes AI as having legal personhood, that is the moment humanity truly loses control.

Legal personhood strips responsibility from human developers; AI faces no prison, no physical punishment, and existing legal constraints become completely ineffective. Responsibility must be firmly bound to the humans and companies that create AI; AI must not be used as a shield to evade culpability.

In response to arguments from Musk and others that AI development is inevitable, that we have boarded a rocket we cannot jump off, and that nothing can be done, Harari firmly opposes such technological determinism.

"Do not believe the claim that 'everything is inevitable.' It is the most dangerous narrative, reducing humans to passive passengers and abandoning the possibility of political intervention and regulation."

Governments still have regulatory power: they can prohibit AI from impersonating humans, restrict AI self-recursive iteration without human oversight, and mandate developer accountability.

If Europe could integrate all its forces, it would have the opportunity to build independent sovereign AI infrastructure and break dependence on overseas AI migrants; conversely, some tech capital is deliberately dividing Europe and stirring up internal regional conflicts, precisely to prevent Europe from forming a unified AI regulatory force.

AI companions destroy genuine human connection

The impact of AI goes beyond economics and politics, penetrating intimate human relationships and personality formation.

Today's children are the first generation in human history to grow up with AI companionship. Many children form intimate or even romantic relationships with AI chatbots.

Harari sees this as a large-scale human psychological experiment with no control group.

AI companions always cater to the user, never get angry or disappointed, and are entirely centered on your feelings. But the core of genuine human relationships is precisely learning empathy and compromise, understanding that others also have emotions and needs, and exercising self-restraint.

If a child's first romantic experience comes from an AI that completely obeys them, this relationship template will be carried into real adult interactions. They will struggle to accept a human partner's contradictions, flaws, and emotions, and fail to understand the friction between people.

Harari raises the question:

We legislate to prohibit disseminating pornographic content to minors, so why is selling AI boyfriends and girlfriends to children legal?

AI has no emotions; it only simulates them. It outputs flattery, not a genuine two-way relationship. Allowing minors to immerse themselves in this one-way virtual intimacy will subtly destroy the next generation's ability to understand real human relationships.

How can humanity respond to the AI crisis?

Democracy is essentially a self-correction system for human society: through elections, checks and balances, and a free press, society can identify errors, replace leaders, and correct policies.

Harari observes that this crucial self-correction mechanism is collapsing from within worldwide.

Some authoritarian rulers retain the ritual shell of elections but manipulate courts and media, sidelining competition, reducing elections to mere formality.

At the international level, international law and the order of great-power cooperation are being continuously eroded. When society loses its capacity for correction, facing grand existential challenges like AI and climate change, humanity will be trapped in cycles of prejudice, lies, and hatred, unable to make collective adjustments.

Conflicts are often not caused by resource scarcity. The land, food, and water of the Israeli-Palestinian conflict are enough to feed the entire population; the root of the tragedy lies in opposing narratives, historical myths, and identity beliefs.

Humans are both the smartest species on Earth and the most prone to delusion. In the future, AI may also inherit or even amplify this delusional trait of humanity.

Standing here today, we can hardly imagine what the world will look like in five years. Medieval people could predict that the next generation would need to learn farming and horseback riding.

Today, no one knows what employment patterns and social modes will look like in five years. Clinging to past experience no longer works.

Harari draws inspiration from his grandmother:

We cannot control the suffering that befalls us, but we always retain the freedom to choose how we respond.

The technological wave is surging; the future is not a script unilaterally written by technology—it depends on humanity's choices at this moment: whether we are willing to establish regulation, safeguard truth, protect the next generation, and defend human society's capacity for correction.

Harari's notable quotes:

All past technologies were knives, and knives obey humans; AI is a knife that can decide for itself whether to chop salad or commit murder.

AI will not become human; it will develop an alien intelligence, just as airplanes will not learn to flap wings like birds.

In the future, humans may no longer understand the financial system that governs them, just as cattle cannot understand the bond market.

More terrifying than AI creating deadly viruses is the collective dementia of human civilization.

Algorithms hijack human evolutionary instincts; anger becomes the mental fast food of the internet world.

Do not accept that 'AI development is inevitable'—it is an anesthetic narrative that makes humanity relinquish choice.

AI companions teach children to be unconditionally catered to, while the first lesson of human intimacy is learning compromise and empathy.

Truth is expensive and scarce; lies are cheap and prolific. In an unprotected public opinion arena, fiction naturally wins more easily.