Book: Nexus: A Brief History of Information Networks from the Stone Age to AI Author: Yuval Noah Harari In one line: Information mostly connects and organises people rather than making them wise - and AI is the first network member that can decide and create on its own, so how we constrain it now will shape human freedom.
1 · Information is not truth
Most information is not a faithful mirror of reality. Its oldest and most common job through history has been to connect and organise people, not to make them wise. Truth is only a small, expensive subset of the information that flows through any network.
2 · Stories plus records, held by self-correction
Every large human network runs on two things: mythology/stories that bind people into a shared “we,” and bureaucracy/records that organise them. What keeps such a network sane is its self-correcting mechanisms - courts, science, free press, elections - that catch and fix its own errors.
3 · AI is a member, not a tool
AI is the first technology that can make decisions and generate ideas by itself - a new kind of agent inside the network, not another instrument we wield. Because it can create stories, forge intimacy, and run bureaucracies at scale, it changes what networks can do to us.
Harari retells the whole human story as a history of information networks - from stone-age gossip to holy books to the printing press to social-media feeds and now AI. His central move is to break two comfortable illusions. The first is the naive view: the idea that if we just pile up more information, it flows naturally toward truth, and truth toward wisdom. The second is the populist view: the cynical claim that information is never about truth at all, only about power and weaponry. Both are wrong. Information is neither an innocent mirror of reality nor purely a club; most of it exists to link people together so they can cooperate at scale.
Because information binds more than it reveals, human networks always mix two mechanisms - stories that create shared identity and trust, and bureaucratic records that track, tax, and organise. Neither mechanism is truth. What saves a network from scaling its own delusions is a set of self-correcting mechanisms that can admit mistakes and reverse them. This is the frame Harari brings to AI. For the first time the network contains a non-human agent that can itself generate ideas and act - and the decisive question, he argues, is one we answer now: whether we build in transparency and correction, or hand opaque power to new forms of control.
Information organises before it informs. Reframe A price, a prayer, a census, a rumour - each is information whose main effect is to coordinate behaviour, not to describe reality accurately. Once you see information as connective tissue rather than as a picture of the world, the naive “more data = more truth” story falls apart.
Two idealised views, both wrong. The naive view trusts that free-flowing information self-corrects into wisdom; history’s witch-hunts and propaganda booms say otherwise. The populist view says all information is just power, so truth is a mask - but that collapse of trust is exactly what lets tyrants win. Reality sits between: networks must hold truth-seeking and order-keeping in tension.
Myth binds, bureaucracy organises. Large-scale cooperation needs both. Mythology/stories - religions, nations, money, brands - manufacture the shared belief that turns strangers into an “us.” Bureaucracy/records - lists, archives, ledgers, filing systems - make that mass of people legible and governable. Powerful, but neither one is reality, and both can lock in error.
Self-correction is the safeguard. A network’s survival depends less on how much it knows than on whether it can catch its own mistakes. Courts overturn verdicts, science retracts findings, a free press exposes lies, elections evict leaders. Strip out these mechanisms and a network scales its blunders as efficiently as its successes.
AI is a new kind of member. Earlier information tech - clay tablets, books, radio - was inert; a human always decided what it said. AI can decide and create on its own: draft the story, run the bureaucracy, choose the target. That makes it an active participant in the network, which is why old assumptions about “tools” no longer hold.
Information is not truth
Truth is costly and rare; most information is cheap and abundant because its job is to connect. Why it matters: it explains why flooding a society with information can make it less wise, and why free markets of information do not automatically converge on the truth.
The naive view fails
“More information leads to truth and wisdom” ignores that lies and fantasies spread as easily as facts - often more easily, because they are simpler and more flattering. Why it matters: it punctures techno-optimism about the internet and AI as automatic engines of enlightenment.
The populist view fails too
“Information is only power, truth is a myth” corrodes the shared trust that any working institution needs. Why it matters: when everyone assumes all sources are equally corrupt, the strongman - not the truth-teller - is the one who benefits.
Stories build the network
Money, nations, and religions are shared fictions that let millions cooperate without knowing each other. Why it matters: it shows how binding stories can be enormously useful and dangerously immune to evidence at the same time.
Bureaucracy runs the network
Written records, archives, and lists let states tax, plan, and control at scale - the unglamorous machinery behind every empire. Why it matters: whoever controls the records shapes reality, and opaque records hide power from scrutiny.
Self-correcting mechanisms
Science, courts, a free press, and elections are institutions built to admit and repair mistakes. Why it matters: their presence or absence is the difference between a network that learns and one that marches confidently off a cliff.
AI as agent, not instrument
AI can generate content, forge one-to-one intimacy with millions, and make consequential decisions without a human in the loop. Why it matters: it can manufacture stories and operate bureaucracies faster than any human self-correction can keep up.
Democracy vs new totalitarianism
Democracy is a transparent, self-correcting conversation; totalitarianism is a closed loop of control. Why it matters: AI can flood the conversation with fakes and hand autocrats a surveillance and persuasion engine of unprecedented reach.
The book moves in three broad arcs. Part one builds the theory - what information does, why it is not truth, and how stories and bureaucracy together made large-scale human cooperation possible, from oral gossip through holy books to the printing press. Part two studies the self-correcting and non-correcting networks: the openness of science and democracy against the closed certainty of dogma and dictatorship, with the modern information age as the pivot. Part three turns to AI as a new inorganic member of the network, and asks how its power to create stories, forge intimacy, and run opaque bureaucracies could either be constrained by democratic institutions or weaponised into a new totalitarianism. The closing message is a choice, not a prophecy: the design of these networks is still ours to make.
Separate information from truth. When something reaches you, ask what it is doing - connecting, organising, persuading, mobilising - before asking whether it is “accurate.” Most of it is not there to inform you.
Name the two mechanisms. In any company, state, or platform, spell out the binding stories (mission, brand, nation, ideology) and the organising records (databases, ledgers, KPIs). Seeing both reveals how it actually runs and where it can go wrong.
Audit the self-correction. Ask of any system: how does it catch and fix its own mistakes? If you cannot point to a court, a review, a vote, or a free critic, treat it as one bad idea away from disaster.
Distrust both extremes. Resist the naive faith that more data fixes everything, and resist the cynical shrug that truth is hopeless. Look for sources and institutions that visibly correct themselves over time.
Treat AI as a network member. When AI acts, ask who it can persuade, what it can decide unsupervised, whose interests it serves, and who can audit it - evaluate it as an agent, not a neutral tool.
Watch for forged intimacy. Be alert when a feed, bot, or assistant seems to understand you personally; personalised persuasion at scale is exactly what AI makes cheap, and it targets the very trust democracy depends on.
Push for constraints now. Argue for transparency, human oversight, and reversibility while the choices are still open. Guardrails are far harder to retrofit onto a network after opaque power has already concentrated.
Critics note that Harari’s sweep can flatten detail: reducing millennia of varied cultures to one “information network” lens is elegant but glosses over cases that resist it. Some historians and technologists argue he overstates AI’s current autonomy - today’s systems are powerful pattern engines, not the self-directed agents the alarming passages sometimes imply - while others counter that he understates near-term risks. The book is stronger as a provocative synthesis and warning than as a precise forecast or a policy manual; the fixes it gestures at (transparency, self-correction, oversight) are named more than engineered. Read it for the frame, not for the fine print.
Information’s oldest job is to connect people, not to reveal the truth to them.
The naive view trusts information too much; the populist view trusts it too little - both misread what it is for.
Stories bind us and records organise us, and neither one is reality.
A network that cannot correct itself will scale its errors as fast as its successes.
AI is the first technology that can be author, actor, and decision-maker at once - a member of the network, not merely its tool.
Democracy is a conversation, and a conversation stays honest only while it stays transparent and open to correction.