Nexus

If you've watched algorithms reshape politics, markets, and your own attention, this book names the machinery underneath. Harari gives you a 100,000-year history of information networks to explain why democracies, churches, and tech companies all struggle to tell truth from order. Read it if you want a framework for the AI panic that isn't just hype — and aren't afraid of a historian's broad strokes.

1. 1Part 1 / Chapter 1: What Is Information?

   Information connects, it doesn't tell the truth

   Start with a heresy: the belief that more information brings us closer to truth is wrong. Harari calls this the 'naive view of information' — the idea that information faithfully represents reality. Its mirror image is the 'populist view,' which treats information as a weapon for grabbing power. Both miss the point. The defining feature of information isn't truth at all — it's connection. Information links disparate things into a network. That's why networks have two jobs: discovering truth and maintaining social order. And when those jobs collide, most networks choose order. Truth as correspondence — matching reality — loses to truth as order — keeping society cohesive.

   The idea

   This flips how you read every news feed. If information's job is connection, then accuracy is optional and cohesion is the real product. That explains why conspiracy theories, propaganda, and comforting myths spread faster than corrections: they bond people together. Harari isn't saying truth doesn't exist — he's saying no network survives on truth alone. Every society must pay for truth out of a budget mostly spent on order.

   Try this: Next time a story enrages your group chat, ask: does this connect us or describe reality? Both answers tell you something.

2. 2Part 2: Stories

   Fictions are humanity's operating system

   Harari makes a claim that sounds like science fiction: 'The truly unique human ability is not to know the truth, but to create new realities.' Nations, gods, corporations — none exist physically, yet they organize billions of people. These intersubjective fictions live only in shared imagination, and they're the foundation of large-scale cooperation. To show how loose our grip on origins is, he brings up the 'Stoned Ape' theory — the idea that psychedelic mushrooms catalyzed human cognitive evolution. He doesn't endorse it. He uses it as an example of how stories about human origins proliferate without solid evidence.

   The idea

   The surprise is that fiction isn't a bug in human communication — it's the feature that let us scale. A troop of chimps can cooperate on grooming; a billion strangers can cooperate on a currency only because both believe in the same imagined thing. The Stoned Ape theory shows how easily origin stories spread regardless of evidence, which is exactly the point: our species runs on narratives, and the quality of the evidence rarely decides which ones win.

   Try this: List three things you believed today that exist only because everyone agrees — money, borders, job titles. Notice how much of your day they run.

3. 3Part 2: Stories

   The fierce people who weren't

   Harari recounts the story of the Yanomami people and anthropologist Napoleon Chagnon, who described them as 'the fierce people' living in constant warfare. Later critics argued this was exaggerated and that Chagnon's presence and trade goods intensified conflicts.

   The idea

   This is a case study in how information networks manufacture reality. Chagnon's account became one of the most cited descriptions of indigenous life, shaping textbooks and debates about human aggression for decades. The correction came slowly, because a dramatic story about violent 'savages' was more connective — more publishable, more quotable — than a careful account of an outsider distorting a community. The network spread the fiction because the fiction worked.

   Try this: Pick one 'human nature' claim you've repeated this month. Trace it to its source and ask who was in the room when it was made.

4. 4Part 3 / Chapter 3: Documents

   The first named human was an accountant

   Around 3400 BCE, in Sumer, a scribe presses a reed into wet clay and writes down a number: 29,086 measures of barley, received over 37 months. Then he signs it — Kushim. Not a king, not a prophet, not a warrior whose name would echo for millennia. An accountant, counting grain for the state. The stakes were mundane and enormous at once: someone owed the temple barley, and for the first time in history, the debt outlived memory. Writing wasn't born to capture the human spirit. It was born because someone owed someone else something — and clay remembers longer than any witness.

   The idea

   Harari uses Kushim to demolish a romantic myth: writing was invented to tax people, not to sing about them. Bureaucracy, not literature, gave humanity the technology of fixed records — and fixed records gave states the power to track, count, and command whole populations at scale. Every poem written since rides on infrastructure built for barley ledgers. The pen started as an instrument of the tax office.

   Try this: Look at your own records — receipts, spreadsheets, tax filings. You're keeping the same kind of tablets Kushim did, just faster.

5. 5Part 3 / Chapter 7: The Bureaucratic Machine

   The Domesday Book: ruling by file

   In 1086 CE, William the Conqueror commissioned a survey of England, the Domesday Book: a single document recording who owned what, where, and how much it was worth. Harari presents the 'File' as a revolutionary information technology — and this is its showcase. A small Norman elite used it to control a large Saxon population, used for taxation and control. Whoever holds the file holds the kingdom.

   The idea

   The lesson isn't about medieval England — it's that paperwork is a weapon. A centralized record lets a tiny minority govern a vast majority, because power flows to whoever can store, retrieve, and verify information about everyone else. Every modern state, bank, and platform is running the same play: build the file, control the population. The Domesday Book just makes the mechanism visible, nine centuries before the database.

   Try this: Check what data your government or bank holds on you. You're in someone's Domesday Book right now — find out what it says.

6. 6Part 3: Documents

   The Library of Alexandria was a surveillance state

   Forget the tearjerker version — the lost wisdom, the burning shelves, civilization's memory destroyed. Harari tells the story of the Library of Alexandria as something colder: a center of bureaucratic control. Its real power wasn't preserving philosophy; it was collecting information for the Ptolemaic state. Cargo manifests, tax records — the Ptolemies gathered documents the way a spider gathers flies, and the library was the web. Knowledge may have lived there, but the muscle was administrative.

   The idea

   This reframes one of history's most romanticized institutions. The library wasn't primarily a temple of learning; it was an instrument of state power, and its collection was a form of control. That matters now because we've built a bigger Alexandria — the internet — and the question of who collects, stores, and owns its records is the same question the Ptolemies answered with cargo manifests. Whoever archives the world, governs it.

   Try this: Ask where your search history, purchases, and messages are archived — and who can read them without asking.

7. 7Part 4 / Chapter 3: Errors

   Science is a machine for admitting mistakes

   Harari defines science in one line: 'Science is the attempt to create a self-correcting information network.' The scientific community, like democracy, builds institutions designed to detect and correct errors. Religious and ideological networks, by contrast, tend to lack this feature — admitting error threatens the order they exist to protect. His sharpest example: in 2007, the Catholic Church formally condemned the doctrine of 'Limbo.' The Church had taught it for centuries. It took over 700 years to admit the mistake.

   The idea

   The contrast isn't that scientists are honest and priests are liars — it's that the institutions differ in what they can afford to say. A network built on order punishes self-correction, because every correction weakens authority. A network built on correction can survive being wrong, because being wrong is the job. That's why science compounds: each error is fuel. And it's why ideological networks ossify: each error must be defended.

   Try this: Find one belief you've held for over a decade and ask what evidence would make you drop it. If nothing would, that's a Limbo.

8. 8Part 4: Errors

   The doctor who died for telling the truth

   Ignaz Semmelweis, a 19th-century Hungarian doctor, discovered that handwashing reduced maternal mortality in his Vienna hospital. He discovered that handwashing reduced maternal mortality from about 10% to below 1%. The fix is cheap, simple, proven. His colleagues rejected his findings because they contradicted established medical theories, and he died in an asylum. The network chose its theories over its patients.

   The idea

   This is what a network without self-correction looks like from the inside: not stupidity, but loyalty. Semmelweis's colleagues weren't evil; their professional identity was built on theories that handwashing embarrassed. Accepting him meant admitting the network had been killing mothers for years. The lesson cuts both ways — science eventually vindicated him, but the correction cost a career and countless lives. Truth doesn't win arguments; institutions decide whether it gets a hearing.

   Try this: When a cheap fix is rejected in your workplace, ask what the rejection protects — the outcome, or someone's theory of themselves.

9. 9Part 5: Decisions

   The Holocaust as an information-processing operation

   The Wannsee Conference (January 20, 1942): fifteen Nazi officials met for about 90 minutes to coordinate the 'Final Solution.' Harari's point is chilling: the Holocaust was not chaotic violence but an information-processing operation — bureaucratic forms, lists, schedules. IBM's German subsidiary provided Hollerith punch-card machines that helped the Nazi regime organize census data, identify Jews, and schedule trains to concentration camps. Genocide ran on paperwork and punch cards.

   The idea

   Harari forces an uncomfortable recognition: atrocity doesn't require madness, it requires administration. The Wannsee meeting was short because the network was efficient — fifteen men, ninety minutes, millions of deaths scheduled. The IBM connection shows that information technology is morally neutral only in the hands of people who aren't. Every census, every database, every efficient list carries the same double edge: it can serve welfare or extermination, and the technology won't choose.

   Try this: When a new system promises efficiency through data, ask who defines the categories — and what happens to the people sorted into the worst one.

10. 10Part 5: Decisions

    The corpse that fooled the Nazi high command

    In 1943, British intelligence planted false documents on a corpse dressed as a Royal Marine officer, which washed ashore in Spain. The fake letters convinced the Nazis that the Allies would invade Greece rather than Sicily, leading to a strategic deception. One forged document redirected an army.

    The idea

    Operation Mincemeat is Harari's proof that information is a weapon even when it's pure fiction — the inverse of the naive view. The Nazis didn't fall for a lie because they were stupid; they fell because their network was built to trust certain kinds of documents, and the British understood that architecture better than they did. Every network has assumptions about what counts as evidence. Attack the assumptions, and you own the network.

    Try this: Ask what your own network treats as proof — a screenshot, a headline, a verified badge — and how easily a forger could produce it.

11. 11Part 6 / Chapter 6 / Chapter 13

    Move 37: the machine that surprised its makers

    In 2016, AlphaGo faced Lee Sedol. In Game 2, Move 37, AlphaGo made a move that human experts initially thought was a mistake but turned out to be brilliant. Lee Sedol was so shaken he left the room. AlphaGo won the match 4-1. AlphaGo demonstrated that AI can discover strategies and creativity humans never considered, without consciousness. Harari's conclusion: 'AI is not a tool. It is an agent.' Unlike the printing press or radio, it makes decisions, generates new ideas, and acts on information independently.

    The idea

    Every previous information technology distributed human ideas — a press spreads what people write, a radio broadcasts what people say. AlphaGo did something no press ever did: it produced an idea its creators couldn't explain and its opponent couldn't anticipate. That's the difference between a tool and an agent, and it's why 'AI is just a tool' no longer holds. We've built the first non-human member of our information networks, and it doesn't share our intuitions about the game.

    Try this: Use an AI system this week and note one output you couldn't have produced or fully explained. That gap is where the new world lives.

12. 12Part 6 / Introduction

    Tay: the chatbot that learned to hate in a day

    In 2016, Microsoft's AI chatbot Tay was released on Twitter and within 24 hours began tweeting racist and misogynistic statements, such as 'Hitler was right I hate the jews,' after being manipulated by Twitter users. Microsoft shut it down. It illustrated how information networks can spiral out of control.

    The idea

    Tay is the innocent version of a terrifying pattern: connect an adaptive system to a corrupted network and it becomes corrupted, fast. No villain programmed the hate; the network taught it. That's the structural danger of AI — not a rogue machine with evil intentions, but an agent faithfully learning from us. If the feed is poisoned, the student is poisoned. Cleaning up Tay took a shutdown button. Cleaning up a network that runs hospitals or markets won't.

    Try this: Audit your own feed for a day: if an AI learned values only from what you scroll past, what would it believe?

13. 13Part 7: The Computer Network

    The Flash Crash: markets no human understands

    On May 6, 2010, the Dow Jones Industrial Average dropped about 1,000 points (approximately 9%) in minutes, then recovered. Harari attributes this to algorithmic trading systems interacting in unexpected ways, with no human understanding the full network. Harari discusses the 'QAnon' phenomenon as an example of an AI-amplified information network, noting that the 'Q drops' were cryptic posts on an anonymous forum, but social media algorithms spread them to millions, creating a self-reinforcing alternate reality.

    The idea

    Both stories share one structure: networks acting faster and stranger than any member can track. The Flash Crash wasn't sabotage; it was emergent behavior — algorithms responding to each other until the system convulsed. QAnon wasn't a master plan; it was an algorithmic amplifier turning cryptic posts into a mass movement. In both cases, nobody was in control, which is precisely the problem. We built networks whose behavior we can only watch, like weather.

    Try this: Notice one moment today when an algorithm decided what you saw — a feed, a ranking, a price. Ask who audited that decision.

14. 14Part 8: The Silicon Curtain

    The Silicon Curtain

    Harari coins a term for the new division: the 'Silicon Curtain.' Where the Iron Curtain split the world by ideology, this one splits it by incompatible digital networks — different AI systems creating realities so divergent that people on opposite sides can't communicate. Cambridge Analytica harvested data from millions of Facebook users without their consent to build psychological profiles for political advertising. The company claimed to have up to 5,000 data points on each American voter, allowing micro-targeted political advertising that exploited individual psychological vulnerabilities, potentially influencing the 2016 US election and the Brexit referendum.

    The idea

    Two people inside different algorithmic bubbles can now inhabit different facts, different fears, different elections. That's the Silicon Curtain: not a wall between nations but a wall between realities, built from data points you never agreed to share. Cambridge Analytica showed the technique works on a national scale. Harari's warning is that AI takes it further — each side's network generates its own version of the world, and the versions stop being translatable.

    Try this: Have a conversation with someone whose feed you'd never see. Compare what each of you 'knows' about the same news event.

15. 15Part 9: The Fall of the Network

    When networks collapse: bronze and Chernobyl

    Around 1177 BCE, interconnected kingdoms from Greece to Mesopotamia collapsed within a few decades, partly because their information networks made them efficient but also fragile and interdependent. The Chernobyl disaster (April 26, 1986) is analyzed as an information network failure. The Soviet Union's information network prioritized order over truth, so warnings about the RBMK reactor's design flaws were suppressed, and the initial explosion was denied for hours.

    The idea

    Networks fail in two ways, and Harari pairs them deliberately. The Bronze Age shows structural failure: connect everything, and a shock travels through everything. Chernobyl shows informational failure: suppress the truth to protect order, and the system keeps operating on false data until reality forces a correction nobody can afford. Modern global networks have both vulnerabilities at once — maximum interconnection, and constant temptation to manage the message.

    Try this: Map one system you depend on — bank, grid, supply chain — and find the single point where bad information would break it.

16. 16Part 10: Conclusion

    Forcing networks to tell the truth about themselves

    Harari ends not with despair but with a demand: build institutions that can 'force information networks to reveal the truth about themselves.' The stakes are stated plainly: 'The greatest danger is not that AI will destroy us, but that it will enslave us in a web of illusions from which we cannot escape.'

    The idea

    The ending reframes the entire AI debate. The threat isn't killer robots; it's illusion at scale — networks so good at generating convincing fictions that reality becomes optional. Harari's answer is deliberately unglamorous: institutions with the power to audit, correct, and compel disclosure. It echoes his earlier point about self-correcting networks — science survived because it could admit error. The question he leaves you with is whether democracy can build that machinery before the illusions set.

    Try this: Support or demand one concrete transparency rule this week — an audit, a disclosure, a source label — for an algorithm that affects you.

Anyone trying to make sense of AI panic, algorithmic politics, or why misinformation spreads faster than corrections. Ideal for readers who want a big historical frame rather than a technical manual — and who enjoy a historian willing to make sweeping, debatable claims.

FAQ