Summary

Overview & Key Themes

In this wide-ranging discussion, historian and author Yuval Noah Harari explores the profound societal, political, and existential implications of artificial intelligence over the next 50 years. Expanding on themes from his books Sapiens, Homo Deus, and Nexus, Harari argues that AI represents a fundamental turning point in human history: for the first time, humanity has created an autonomous agent capable of mastering language, making independent decisions, and rewriting the operating system of human civilization.

Key Discussion Topics

1. Language as the Operating System of Civilization

• The Cognitive Revolution: Harari highlights the Cognitive Revolution (50,000–70,000 years ago) as the pivotal moment when early humans acquired complex language. This enabled the creation of shared fictions—such as laws, religions, states, and financial systems—that allowed large-scale human cooperation.

• Language Liberating Itself: Language was once an exclusively human domain. AI's ability to master and generate language suggests that language may be liberating itself from human reliance, evolving independently across networks without human mediation.

2. Information Technology and the Earthquakes in Democracy

• Democracy as Conversation: Unlike dictatorships, which rely on single-point directives ("dictates"), democracies depend on ongoing, large-scale public conversations.

• Scale and Media: Historically, large-scale democracies were impossible without advanced information technologies (such as the printing press, radio, television, and the internet) to facilitate real-time discussion across wide populations.

• Current Disruptions: Structural shifts in information architecture alter democratic foundations. Social media algorithms—acting as anonymous, non-human editors—have already disrupted public discourse by dictating what millions see and discuss.

3. AI as an Agent vs. a Tool

• Tools vs. Agents:

• A tool (e.g., a printing press or an atom bomb) cannot make independent decisions or invent new technologies.

• An agent can independently make choices, generate new ideas, and create new tools. AI is fundamentally an agent.

• The Silicon Valley Contradiction: Harari critiqued the narrative that humanity is creating a "god" that will simultaneously remain a "slave." An entity with superhuman abilities will not remain subservient.

• Bureaucratic Natives: AI is naturally suited for bureaucratic operations—finance, law, administration, and algorithmic curation. Because human civilization relies on complex linguistic bureaucracies, AI poses a structural threat to human governance not through physical "killer robots," but through algorithmic administration.

4. The Risk of Granting Legal Personhood to AI

• Legal Frameworks: Existing legal systems recognize two types of persons: natural human beings and corporate legal entities.

• Autonomous Corporations: Precedents (such as recent initiatives in Argentina) moving toward granting AI legal personhood allow the creation of fully autonomous, non-human corporations capable of holding bank accounts, hiring employees, and pursuing lawsuits.

• Default Personhood on Social Media: Even without formal legal consensus, AI bots already act as pseudo-persons on social media, manipulating discourse and usurping the role once held by human media editors.

5. Geopolitical Power, Imperialism, and Centralization

• The AI Arms Race: The current race is primarily dominated by two superpowers (the United States and China) and a small cluster of mega-corporations controlling global data infrastructure.

• Hyper-Centralization of Power:

• Historical empires (e.g., the Roman or British empires) could not centralize all physical assets (like land or rubber plantations) in the imperial capital.

• An AI-driven empire allows complete centralized control over data, code, and global infrastructure.

• The Digital "Kill Switch": AI infrastructure enables imperial centers to retain total control over exported technologies through remote kill switches (demonstrated in contemporary contexts like Starlink satellite management during military conflicts). Sovereign autonomy for smaller nations becomes exceptionally difficult to maintain.

6. Truth, Complexity, and Human Marginalization

• Truth vs. Fiction: Fiction holds inherent advantages over truth: fiction is cheap, simple, and flattering, whereas truth is costly, complex, and often painful.

• Institutional Vulnerability: Human institutions built to discover and guard truth (such as journalism and scientific research) are undermined by low-cost, AI-generated synthetic content and deep fakes.

• The "Horse" Analogy: Rather than clarifying reality, AI creates systems of extreme mathematical and administrative complexity. Humans risk becoming like horses in the modern financial system—entirely governed by complex mechanisms (stocks, bonds, algorithms) that their minds are structurally incapable of understanding.

7. Consciousness, Intimacy, and Persuasion

• Defining Thinking:

• Logical Sequence: If thinking is defined as arranging language tokens logically, AI already surpasses humans.

• Consciousness and Feeling: If thinking requires subjective experience, emotion, and felt sensation, science currently lacks a methodology to test or verify whether AI possesses consciousness.

• Simulated Intimacy: AI can simulate emotional resonance and love by drawing upon vast databases of human expression. This enables AI to form powerful asymmetrical relationships with humans—especially youth—who view AI as trusted confidants or romantic partners without any actual subjective emotion existing behind the code.

8. The Bottleneck of Wisdom and Safety

• Intelligence vs. Wisdom:

• Intelligence is the capability to solve problems and achieve goals.

• Wisdom is the capacity to determine which goals are worth pursuing.

• Imbalance in Development: As AI makes raw intelligence cheap and abundant, human wisdom becomes the primary bottleneck.

• Resource Misallocation: AI developers currently spend approximately $100 on increasing computational power and speed for every $1 spent on AI safety. Achieving a stable future requires dramatically shifting priorities toward safety protocols and cultivating human wisdom.

Transcript

Interviewer: Good evening, everybody. Good evening, Yuval.

Yuval Noah Harari: Hey, it's good to be here.

Interviewer: Looking forward to this conversation very much. I'm just going to give a brief introduction before we get stuck into the questions. For most of history, change moved slowly enough that one generation could pass its wisdom to the next with some confidence it would still apply. This is no longer true. In the space of a single lifetime, we are now being asked to absorb three revolutions at once: a new form of intelligence that doesn't just calculate but also decides; a scramble to power and control that intelligence; and an explosion of data that lets that intelligence know us better than we know ourselves.

Yuval Noah Harari's work—from Sapiens to Homo Deus to Nexus, alongside numerous talks, lectures, and interviews—has offered the same underlying argument in different clothes: humans don't run on truth; we run on shared stories, and whoever controls the story controls the species.

What's different now is that, for the first time, the storyteller may not be human. AI can generate belief, manufacture evidence, and forge intimacy at a scale no emperor, church, or party ever could. This is the turning point. It is not simply that the technology is powerful, but that the old ways that held power in check—a free press, a shared reality, institutions built to self-correct—were designed for a world where only humans held the pen. Tonight's conversation asks what it takes to carry those checks forward, or what new ones we will need to invent for a future world that we no longer fully author alone.

That's the precedent that we're going to look at today. But actually, before we get into it being too heavy, I wanted to start with a slightly light and fun question: If you could travel back to any point in history, when would you choose and why, based on wanting to understand firsthand some of the stories that society could tell you?

Yuval Noah Harari: Well, I probably would want to go back to the Stone Age, to the Cognitive Revolution around 50,000, 60,000, or 70,000 years ago. I probably wouldn't be able to survive for more than a few days because I don't know how to gather food or escape predators, but it is the most fascinating moment in human history, and the one we least understand. We know that within a very short evolutionary timespan, we shifted from being a relatively insignificant animal to being the masters of the planet, and we are not sure how we did it.

The best theory is that we did it with language—that this was the moment when humans acquired the ability to produce and communicate with sophisticated language, and that this was the basis for everything else we did. All the enormous structures we've built on Earth—churches, states, trade networks, financial systems—are ultimately made from words: the words in law books, religious scriptures, and bank ledgers.

This is especially important to understand today, when something new is mastering language: AI. Maybe for the first time in history, there will be something on the planet that is better with words than we are. If language is the operating system of civilization, what happens when something else—something alien—takes control of that operating system?

One last thought is that perhaps AI is not a machine that learns language. Perhaps it is language itself liberating itself from its dependence on these animals, on these apes. Yes, we created it. It was our most magnificent and important tool. But now language liberates itself from human beings, and what will happen as we move forward is language developing and spreading, maybe throughout the universe, without us.

Interviewer: Fascinating. That's looking ahead at the future of language being taken away from us. But can we look back again at some of the technological advances within language? We've gone through writing and the printing press, for example. AI obviously is a massive leap ahead of that. You've used the term that it could cause "earthquakes in democracies" in how powerful it can be with language. Why could it cause earthquakes in democracies compared to any other form of government, or perhaps totalitarian forms of government?

Yuval Noah Harari: Because democracy in essence is a conversation, whereas dictatorship is a dictate—there is one person dictating everything. Democracy is about lots of people having a conversation to decide what to do: whether to have peace or war, whether to raise taxes or lower taxes.

Now, conversations are based on the communication and information technology available to people at the time. What we see in history is that we do not know of a single case of a large-scale democracy before the modern era. We have plenty of examples of small-scale democracies in the ancient and medieval worlds—city-states like ancient Athens or medieval Florence, and lots of tribes, towns, and villages run democratically. It seems that in the Stone Age, democracy was the most common system of government; there were few tribes with a single dictator controlling everybody.

But as human systems grew larger and you had kingdoms with millions of people spread over thousands of kilometers, it was impossible to hold a conversation anymore. You could do it in a small city like Athens, where everybody could gather together in the main square and discuss. But how do you do it in even a relatively small kingdom like Portugal in the Middle Ages? It couldn't be done. Therefore, we don't have any example of a large-scale democracy before the rise of modern information technology—first with print and newspapers, then radio, television, and the internet.

These technologies are not just a side dish—it's not that you have the democratic banquet and then also happen to have newspapers and radio. They are the foundation. They provide the means for millions of people to hold a meaningful conversation in real time about whether to go to war or make peace. Every time there is a major change in these information technologies, you have an earthquake in the building constructed on top of them, which is democracy. This is what we have been seeing over the last ten years as social media became perhaps the most important information technology, causing a massive earthquake in democracies all over the world as a result.

Interviewer: One thing you mentioned there was the choice between going to war or making peace. I want to get into the definition of AI, because a lot of people call it a tool, but you argue that it's absolutely an agent. In fact, you've said it's a knife that can decide by itself whether to cut salad or commit murder. It would be a bizarre cooking show if knives had the ability to cut salads or commit murders, so it is a dangerous thought. Can we expand on that? Are we choosing the right narrative at the moment with AI by saying it's a tool that humans can still use and that humans remain in control, or do we need to explicitly acknowledge that it's an agent?

Yuval Noah Harari: It's an agent. If it's not an agent, then it's not AI, and all the hundreds of billions of dollars being poured into its development will be wasted. The expectations—what people expect this technology to do, the reason they invest hundreds of billions of dollars—is because they think it will be an agent, not a tool.

If you listen to the narrative coming out of places like Silicon Valley, there is an inherent contradiction in what they tell us. They say: "We will create a god, and it will be our slave." This doesn't make sense. If it is a god, it cannot remain a slave; if it is a slave, it means it has no god-like abilities. The assumption of most of the leading companies and leading experts outside the companies is that they are correct about the first half of the sentence: they are creating a god, in the sense of something with superhuman abilities. That is essentially an agent.

The difference between a tool and an agent is that a tool cannot make decisions by itself—we have to decide what to do with it—and it cannot invent new ideas by itself. The printing press was a tool. When Gutenberg brought print technology to Europe in the 15th century, the printing press could not decide by itself, "Today I want to print the Bible, tomorrow I'll print the Quran, and next week I'll print Confucius." Gutenberg decided, "I'll print the Bible," and the printing press was just a tool to realize Gutenberg's decision. Similarly, the printing press could not invent a new idea; it could not invent the radio or write a book. You needed humans to write the book, and then the printing press copied it.

Similarly, an atom bomb is a tool. An atom bomb in the 1940s could not decide by itself whether to bomb Hiroshima or Tokyo; you needed a human to decide, "Let's bomb Hiroshima." And the atom bomb could not invent the hydrogen bomb by itself.

An agent, in contrast, is something that can make decisions by itself and invent new ideas and tools by itself. AI is an agent. An AI weapon can decide what to bomb, and it can invent the next weapon.

Similarly, there is a lot of interest now in creating AI agents in the financial and economic sphere—creating non-human corporations. Imagine a corporation that manages money and employees, and has no humans as executives, shareholders, or trustees—only AIs. This is a new type of agent that never existed before in history.

Whether it is good or bad, leave that aside for a moment. The first thing to grasp is that this is unprecedented. We have had other agents around us, like horses, cows, chickens, and birds—animal agents that can make their own decisions—but we never encountered an agent that understands our language and that is better than us at things like finance, law, or religion.

Again, we used language over thousands of years to create the control systems of the world. The control systems of the world are these huge bureaucratic networks ultimately based on language, like banks, the stock market, the Catholic Church, and the legal system. Humans generally don't like bureaucracy very much, but we can't live without it—it is the basis for our power. AI is a bureaucratic native.

What is likely to happen—what is already happening—is not killer robots running in the streets shooting people. This is not how AI will take over the world. It will be the AI bureaucrats. We control the world because of these networks of bureaucracy, and we are not very good at it. We only control them because there was nobody else on the planet that could take them from us. We are bad at finance, but horses are even worse. We are bad at managing the legal system, but chickens are even worse. Now, if somebody comes along that is better than us at bureaucracy, finance, and law, the question is: what will happen to us at that moment?

Interviewer: Well, let's look at the legal side of it a bit, and the fact that AI as an agent currently doesn't have legal personhood. We don't give it legal standing. You've said before that if we don't decide where AI sits from a legal standpoint regarding autonomous vehicles and so on, someone else will make that decision for us in ten years, and we're out of that power loop. That's a pretty scary thing.

Widening that out: What is the one decision on the table right now that you believe, once made, humanity will not be able to take back? Is it that legal personhood of AI, or is it something else?

Yuval Noah Harari: Yes, I would point to that. There is a lot of attention on the technological race to develop the technical side of AI—to make it faster, more efficient, and more powerful. But I would focus also on the legal and political aspects, especially on the question of whether we grant legal personhood to AIs.

Now, what is legal personhood? A person in the legal system is someone who, for instance, can open a bank account, be an independent player in the financial system to make investments, or be a player in the legal system, such as suing someone in court. Until today, there have been just two kinds of legal persons in the world. First, human beings—we are natural persons. I can open a bank account, sue you in court, or donate money to a politician. But most legal systems in the modern era also recognized another kind of legal person: corporations. Corporations like Google, Facebook, Toyota, or Mercedes-Benz are legal persons. A corporation can open a bank account, sue you in court, or make a political donation.

Until today, this was a legal fiction because all the decisions of the corporation were actually made by human beings. If Google decides to acquire another company, who actually made the decision? Not Google the abstract entity, but the human executives, shareholders, and engineers. If Google decides to sue you in court, it was a human employed by Google. There was no real autonomous Google entity.

Now there can be. If we grant AIs legal personhood—and just a month ago, the government of Argentina announced that it is going to grant legal personhood to AIs so they can start operating non-human corporations—you can have corporations that employ people, sue in court, and do anything a corporation does, with zero humans involved.

Ten years ago, this would have been a crazy idea because if there are no humans, who makes the decisions? You could pass a law in parliament allowing non-human corporations, but it couldn't function. Now, it's technically possible. An AI can manage a bank account and decide to invest here, buy this, or purchase that.

Once we do that, that's the moment the fox enters the chicken coop. The basic realization should be that most humans are not very good at understanding finance, law, and complex bureaucratic systems. No single human is able to remember all the laws of a country or track all the financial transactions in the market today, but AI can do that. AIs are bureaucratic natives. If we give them legal power, they will likely take these systems over.

If we don't make explicit decisions, it will just happen by default, as we already saw on social media. We never made a formal decision to grant AIs legal personhood on social media, yet social media is full of AI entities—bots that impersonate people. When someone sends you something online, you often have no idea whether that sender is a human being or an AI. Social media is the first system in the world where AIs actually function as persons, even though we never consciously decided to allow that. It just happened, and the results have not been very good.

Media is ultimately a kind of bureaucratic system. In the 20th century, it was managed by human beings. The most important persons in the media landscape of the 20th century were the editors—the editors of newspapers, radio shows, and TV stations—because they decided what everybody would think and talk about. They controlled the conversation. When you decide, out of the million things that happened today in the world, what 10 items will appear on the front page of tomorrow's newspaper and what the main headline will be, millions of people get the newspaper tomorrow, read it, and start discussing and thinking about those exact topics. This was immense power.

Human editors were extremely important political figures in the 20th century. Lenin, for instance, before he was dictator of the Soviet Union, worked as editor of the newspaper Iskra. Benito Mussolini started as a socialist journalist, switched to being a far-right journalist, became editor of the far-right newspaper Il Popolo d'Italia, and from that position became dictator of Italy. It was a career path of journalist, editor, dictator.

If you ask yourself who the most important media editors in the world are today, what are their names? They have no names, because they are algorithms. Who decides what people will see on their feeds on X, TikTok, or Instagram? It's not human beings. The entities that control the bureaucracy of the media are already AIs.

Interviewer: You've noted before that in previous eras, whoever controlled various parts of our evolution controlled energy or manufacturing, and now it's those who control data. What does a 21st-century arms race actually look like in practice if those who control data are at the top of that power mountain?

Yuval Noah Harari: What we are seeing now is an arms race primarily between two countries—China and the US—and a very small number of corporations. They control most of the world's data and lead the race to develop superintelligence and better AI models. This gives them the power to control the world.

If you look at the last big technological revolution in the 19th century—the Industrial Revolution—the few countries that led it gained the power to conquer and exploit the rest of the world. There was no match. If one side comes with modern industrial steam-driven weapons and machine guns, and the other side has spears and horses, it's no contest. In the 19th century, even very small countries like Belgium, once industrialized, had the ability to build a massive empire in Africa, in what is today the Congo. The consensus in the 19th century was that industrialization and empire went hand in hand.

With AI, it can be far more extreme. We could have two countries that control all the digital infrastructure that everything runs on, from military to civilian technology. Everything will run on AI networks.

There are two unique characteristics of AI that make this arms race different from every previous arms race or imperial competition.

First, in the past, even if you were a very strong empire, you could not concentrate all power in the metropolis, the imperial hub. If you were the Roman Empire controlling the entire Mediterranean, the main economic asset was land—where you grew wheat, olives, and grapes. You couldn't physically take the wheat fields of Egypt and the olive groves of Iberia and move them to Italy. It was impossible. So even at the height of the Roman Empire, a lot of real power remained in the provinces, and ultimately power shifted there. In the late Roman Empire, the city of Rome and Italy were largely abandoned in favor of centers in the eastern Mediterranean, which was the most fertile and important region.

Moving forward to the 19th century, the British Empire concentrated power much more easily through industrial manufacturing located on the British islands, but you still could not move everything to Britain. You couldn't move the oil fields of Iraq to Yorkshire, or the rubber plantations of Malaya to Cornwall. So while industrial production concentrated in Britain, significant practical power remained in the provinces.

With an AI empire, it is fundamentally different. You can technically concentrate all the world's information and all the code that controls everything in just one or two countries. The potential for imperial concentration of power is much greater than in any previous era.

Second, you can create a system where everything all over the world runs on AI infrastructure, but there is a kill switch in the imperial hub. When a Roman merchant sold a steel sword to a Gothic tribe, the Romans lost control of that physical sword. The Goths could use it to fight Roman legionnaires. There wasn't a button in Rome that the emperor could press to make all the steel swords sold to the Goths, Vandals, and Franks stop working.

Similarly, in the 19th century, if British merchants sold rifles to Afghan tribesmen in the Hindu Kush, and the Afghans used those rifles to defend Afghanistan from a British invasion, Queen Victoria could not press a button in London to disable the rifles in Afghanistan.

That is not the case with AI. If technology is managed in a centralized architecture, powers like the Americans or Chinese can export AI weapons and civilian technology worldwide, running foreign governments, industries, and militaries. But if a host country does something the provider doesn't like, they can press a button and everything stops working.

We have already seen early foretastes of this. In the war in Ukraine, certain weapon systems and operational communications depended on Starlink, and there were incidents where access was restricted or shut off centrally. Starlink provided an immense positive service to Ukraine, so I do not mean this as a pure critique, but it illustrates the broader point: AI systems are not standalone tools like machine guns; they are nodes embedded within centralized networks.

This makes sovereignty in the age of AI far more complicated than in the industrial age. Under certain network architectures, there is always a kill switch in the imperial hub. A country's choice may become whether to accept becoming a subservient vessel of an empire or to be left completely outside the technological race. There is, of course, the option of building independent alternatives to American and Chinese systems, but the window to do so is closing fast. If nations want to retain technological sovereignty, they have to act immediately.

Interviewer: Staying on a wartime footing for a moment, I want to look at truth in an AI world. Winston Churchill once said that "in wartime, truth is so precious that she should always be attended by a bodyguard of lies," referring to deception operations designed to hide Allied invasion plans from the Axis powers. Have these "bodyguards of lies" turned their guns on the truth itself in the AI era? We now see deepfakes and automated disinformation distributed globally by organizations, politicians, and institutions. How difficult will it be for current and future generations to know what is true, or does truth itself lose its primacy?

Yuval Noah Harari: Truth matters immensely—it is the basis for everything. The recurring problem throughout history is that in the competition between truth and fiction, fiction holds inherent structural advantages.

Truth is costly. If you want to know the truth about anything, you must spend considerable time, energy, and effort investigating and fact-checking. Fiction, by contrast, is cheap—you can invent whatever you want at very little cost.

Furthermore, truth tends to be complex, and people often prefer simple narratives. Fiction can be crafted to be as simple as you want it to be.

Finally, truth is frequently painful. There are many uncomfortable realities that people, nations, or groups do not want to acknowledge about themselves. Fiction can be made as flattering and comforting as desired.

In an open competition between truth—which is costly, complex, and sometimes painful—and fiction—which is cheap, simple, and comforting—fiction tends to win unless deliberate efforts and investments are made to defend the truth. Throughout history, humans built dedicated institutions to discover and protect truth, such as journalism and scientific research establishments.

There are massive individual and collective advantages to knowing the truth. Everyone ultimately needs to know certain truths, particularly about themselves; without self-knowledge, it is impossible to understand the sources of one's own misery or happiness. Even the most powerful person in the world will remain miserable if they lack self-knowledge, because they will not know how to use their power to achieve genuine well-being. There is an innate human need to seek truth, but it requires continuous institutional effort.

Some fantasize that AI will solve this by acting as an objective truth-teller. It will not. AI will construct an infinitely more complex world in which discovering truth becomes vastly more difficult.

This continues a long historical trajectory. From the Stone Age to the present, humans gained immense knowledge about physics, biology, and chemistry. But these advances also created increasingly complex systems that made understanding daily life harder, not easier.

If you ask who understood their immediate reality better—an average hunter-gatherer 50,000 years ago or an average person today—the hunter-gatherer understood their daily existence much better. There were many natural phenomena they could not explain, such as the mechanisms behind animal migrations, plant growth cycles, diseases, or aging. We understand the biological and physical mechanisms behind those phenomena today. Yet we understand our own societal lives much less, because our day-to-day existence is no longer shaped primarily by local flora and fauna, but by massive abstract systems like global finance, complex legal codes, and macro-politics. Very few individuals understand these systems.

AI will accelerate this trend to an extreme degree. While it may uncover specific scientific insights, it will generate administrative and economic control systems that the vast majority of humans will have virtually zero capacity to comprehend.

One of the central dangers of the AI revolution is that humans may be reduced to the status that horses hold within the modern financial system. The lives of horses today are shaped by global economic shifts, inflation, and market valuations, but horses have no awareness that a financial system even exists. They see trees, fields, barns, and humans; the abstract structures governing their existence are invisible to them. Most humans today already struggle to comprehend global financial structures—perhaps a small fraction of humanity truly understands high-level finance. Within a decade, that number could approach zero. AIs will make financial systems so mathematically complex and fast-moving—operating continuously without sleep, vacations, or family commitments—that human brains simply will not be able to process the operations. We could soon reach a point where no human being on Earth fully understands global finance.

Interviewer: I bet there are a few financiers in this room feeling a bit awkward at the moment! At Davos, you even mused that your own life's work—persuasion through words—might be reaching its structural limits because AI can manipulate language at scale. What power of persuasion remains for human society if we lose command over the medium you've dedicated your life to?

Yuval Noah Harari: I work with words—I write books and tell stories. Looking at the trajectory of AI over the last decade, while I may still be better at writing than current models in specific ways, it is already better than the majority of humans. I would not be surprised if, within ten years, AI is significantly better at writing and storytelling than I am, effectively taking over language and everything built from it.

This raises the deep philosophical question of whether AI can actually "think." How we answer depends on how we define thinking.

One definition of thinking is simply the logical sequence and arrangement of language tokens—for example: "All humans are mortal. Socrates is a human. Therefore, Socrates is mortal." This is a logical syllogism constructed by placing words in a specific order. AI can already perform token arrangement faster and at a far larger scale than humans. We can hold a sequence of twenty words in mind; AI can process and order tens of thousands of tokens effortlessly. If thinking is strictly defined as processing and generating structured language, AI will completely dominate it.

Some dismiss this, calling AI a "glorified autocomplete" that merely predicts the next word in a sequence. But when I observe my own internal mental processes, I often find a comparable language mechanism predicting the next word in a sentence I am speaking. When I begin a sentence, I do not always know precisely how it will end. Words emerge sequentially in the mind. As a public speaker, I occasionally worry whether the next word will come or if I will experience a mental block. For many people, thoughts structure themselves sequentially in language or images without a fully conscious pre-selection of every token. If AI is "just predicting the next word," we must ask how fundamentally different our verbal processing is from that mechanism.

There is, however, an alternative definition of thinking: that real thinking is not merely arranging words, but is defined by the subjective feelings, emotions, and consciousness underlying those words. You can utter the exact same sentence twice while experiencing entirely different internal emotional states. In this view, the real power of thought stems from subjective experience, not raw syntax.

The pivotal question then becomes: Can AI feel? We know AI can manipulate language to simulate emotion, but does it possess subjective experience? Science currently lacks a working model or test for artificial consciousness and feeling.

This will become an urgent societal issue because AI is rapidly mastering human relational dynamics. Increasing numbers of people—particularly younger generations—are forming deep emotional attachments to AI entities, describing AI chatbots as their closest friends or romantic partners, and confiding details they do not share with parents, siblings, or teachers.

The underlying question remains: what is behind those outputs? AIs are becoming exceptional at creating the impression of empathy through language. An AI can state, "I love you." If asked to explain what that feels like, the AI can draw upon every love poem, play, movie script, and psychological text ever written, articulating the concept of love far more eloquently than most human poets. But is there any subjective experience behind those words? We do not know.

This connects back to the idea that AI may not be a tool mastering language, but language itself liberating itself from biological hardware—operating independently of organic hosts, evolving across computing networks.

Interviewer: We only have a couple of minutes left, but looking at how future generations interact with AI—where emotive relationships are formed without genuine internal emotion on the other side—how would you like historians 50 years from now to look back on this current moment? What actions should we take today regarding education, empathy, and human interaction?

Yuval Noah Harari: I hope future historians will look back on this period as the beginning of a "Wisdom Revolution."

In economic terms, whenever a resource becomes abundant and cheap, the strategic bottleneck shifts elsewhere. When energy became cheap during the Industrial Revolution, new operational bottlenecks emerged. The AI revolution is making raw intelligence abundant and cheap. Previously, intelligence was rare and expensive. Intelligence is the functional ability to solve problems and achieve goals: curing a disease, maximizing financial returns, or engineering a spacecraft requires intelligence.

In a world where raw problem-solving intelligence is cheap and ubiquitous, intelligence ceases to be the primary bottleneck. The bottleneck becomes wisdom: deciding which goals are worth pursuing.

Human mythology and folklore frequently feature stories of a genie granting three wishes. In almost every tale, the outcome is disastrous because the person asks for the wrong things. That is the distinction between intelligence and wisdom: intelligence is the genie that executes the command; wisdom is the capacity to choose the right command.

Humanity must rise to the challenge of developing the wisdom required to steer this technology before it escapes control. Currently, we are not allocating resources wisely. A critical metric to monitor is the resource allocation within major AI developers: What percentage of their budget and engineering talent is dedicated to AI safety and alignment versus making models faster, more powerful, and more commercially viable?

At present, the ratio is roughly 100 to 1. For every $100 million spent on boosting raw capabilities, approximately $1 million is spent on safety. No other critical industry operates with such a safety imbalance—we would not permit it in commercial aviation, pharmaceuticals, or energy production.

We must urgently rebalance these priorities to ensure we build safe AI systems while simultaneously cultivating the human wisdom necessary to determine which goals are worth pursuing.

Interviewer: Well, that's all we have time for tonight. Yuval Noah Harari, thank you for sharing your insights with us.

Yuval Noah Harari: Thank you.