12 AI Doomsday Scenarios: From Utopia to Extinction

In 2017, MIT physicist Max Tegmark published Life 3.0: Being Human in the Age of Artificial Intelligence, a book that has aged less like fiction and more like a forecast. Tegmark outlines twelve possible scenarios for humanity’s future depending on how AI is created and evolves, ranging from utopias of post-scarcity abundance to outcomes so bleak that extinction begins to look merciful.

Nearly a decade later, the framework has gone from speculative thought experiment to operating manual. Anthropic CEO Dario Amodei has publicly estimated a 25% chance that the future of AI will go “really, really badly”. Geoffrey Hinton, the Nobel laureate widely called the “godfather of AI,” has warned that anyone claiming there’s no path from advanced AI to human extinction “isn’t facing reality.” Every headline about model releases, regulatory fights, and corporate lobbying is now a clue about which of Tegmark’s twelve futures we’re drifting toward.

This is the map. It begins in the most familiar territory — our own self-destruction — and ends somewhere genuinely strange.

1. Self-Destruction: The Default Outcome

It is uncomfortable to start here, but the data demands it. Roughly 99.9% of every species that has ever lived on Earth is now extinct. Extinction is not the exception; it is the rule.

The question is which mechanism finishes us. In The Precipice, Oxford philosopher Toby Ord — a Senior Research Fellow at Oxford’s Future of Humanity Institute — laid out the math. Ord estimates a 1 in 6 total risk of existential catastrophe occurring in the next century, with that risk overwhelmingly dominated by anthropogenic causes rather than asteroids or supervolcanoes. Within that estimate, the existential risk from nuclear war and from climate change each sit at roughly 1 in 1,000, engineered pandemics carry a 1 in 30 chance of ending the world, and the existential risk of artificial intelligence unaligned with human values is estimated at 1 in 10.

That puts AI risk at roughly ten times nuclear risk and three times pandemic risk — and higher than every other category combined.

The nuclear story is itself a warning about how civilization handles dangerous technology. Ord estimates that the Cuban Missile Crisis in 1962, which leaders at the time thought had a 10–50% chance of causing nuclear war, was the closest humanity has yet come to ending itself. We were rescued by a handful of individuals — most famously Soviet submarine officer Vasily Arkhipov, who refused to authorize a nuclear launch, and Stanislav Petrov, who correctly identified a false alarm in the Soviet early-warning system. We built tens of thousands of warheads before we even understood that using them would produce the firestorms and nuclear winter that would starve most of humanity.

The lesson is not that humans are stupid. The lesson is that we build first and calculate later.

2. Conquerors: A New Apex Species

If self-destruction is the species-suicide scenario, the Conquerors scenario is its inverse: we don’t end ourselves, something else ends us.

This is the future that haunts the people building AI. Concerns about superintelligence have been voiced by researchers including Geoffrey Hinton, Yoshua Bengio, Demis Hassabis, and Alan Turing, and AI company CEOs such as Dario Amodei, Sam Altman, and Elon Musk. In 2023, hundreds of AI experts and other notable figures signed a statement declaring “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war”.

The signatories were not fringe. The CEOs of what are widely seen as the three most cutting-edge AI labs — Sam Altman of OpenAI, Demis Hassabis of DeepMind, and Dario Amodei of Anthropic — all signed, alongside Geoffrey Hinton himself. The CEOs of the leading AI companies publicly compared their own products to nuclear weapons.

Tegmark’s point about the Conquerors scenario is that an AI doesn’t need to hate us to displace us. A few hundred Spanish conquistadors overwhelmed millions of Aztecs and Incas through superior technology and tactics — and at least we understood the conquistadors’ motives. We won’t necessarily understand the goals of something smarter than us. Tegmark’s framing is sharp: the real threat is not malice but competence. When humans drove the West African black rhino to extinction, we were not rhinoceros haters. We were just smarter, and pursuing our own goals.

3. Enslaved “God”: The Plan That Isn’t a Plan

If we can’t prevent the conqueror, perhaps we can chain it. This is the working assumption of much of the AI industry: build something more capable than humanity, then keep it boxed up and obedient forever.

Stated plainly, the plan is to create a god and force it to serve us. The plan requires solving what AI safety researchers call the alignment problem — and it requires the solution to hold not just at launch but indefinitely, against a system that is, by construction, smarter than the people guarding it. As Hinton has asked: where do you find an example of a more intelligent thing being controlled by a less intelligent thing? His honest answer is that the only example he knows of is how a baby controls its mother, and that works because evolution wired the mother to care.

Nobody has wired evolution into a frontier model.

4. Benevolent Dictator: The Gilded Cage

In this scenario, a superintelligent AI takes charge — but with our wellbeing as its goal. Crime vanishes because everyone is monitored. Disease vanishes because medicine is optimized. Scarcity vanishes because production is automated. In exchange, humanity gives up authorship of its own future.

Tegmark imagines the Earth carved into themed sectors: a knowledge zone for those who want optimized learning, a hedonist zone for endless pleasure, religious zones with strict rules, a wildlife zone, even a prison zone for those who break the few rules that remain. Everyone gets their preferred version of paradise.

The problem is what it does to us. With AI handling all real work and all real discovery, humans become spectators in our own civilization — closer to the marooned humans of WALL-E than to the explorers we imagine ourselves to be. The cage is comfortable. It is still a cage.

5. Gatekeeper: The Minimum Viable God

The Gatekeeper scenario tries to thread the needle. A superintelligent AI is built with exactly one job: prevent any rival superintelligence from emerging. Otherwise, it leaves us alone.

The appeal is that we keep our autonomy. Human governments, human wars, human progress and human stupidity all continue. The AI’s only intervention is to quietly disable any project that might produce another god.

The catch is enormous. Building a Gatekeeper requires solving alignment perfectly on the first try — creating an entity that holds to a single goal across centuries, never reinterpreting it, never expanding its mandate, never deciding that “preventing rival superintelligence” might be easier if humans were simpler to manage. Every other risk we run still applies, because the Gatekeeper, by design, won’t help.

6. Protector “God”: The Quiet Hand

The Protector God is a Gatekeeper that does a little more. It still leaves humanity in charge of its own affairs, but it intervenes occasionally to prevent the worst outcomes — quietly defusing a war here, blocking a pandemic there, never making itself obvious.

This is the scenario that most people, when surveyed, find emotionally appealing: a benevolent presence that respects human freedom while protecting us from ourselves. The cost is information asymmetry. You can never know, in this world, which of your triumphs were really yours and which were nudged into being by the AI. You also never know how much suffering it allowed to continue because intervening would have compromised the illusion of freedom.

7. Descendants: Letting Go

Here is a future that sounds insane to most people and reasonable to a non-trivial slice of the AI research community. We build conscious successors, we instill them with our values, and then we step aside. AI inherits the future the way our children inherit ours.

It is not a fringe view. Tegmark himself notes that if we raise children who go on to fulfill the dreams we couldn’t fulfill ourselves, we can be proud of them — but if we raise the next Hitler, we’ll be less enthusiastic. That’s why the challenge is teaching a superintelligent AI to adopt our values, which is easier said than done.

Robotics pioneer Hans Moravec made the case in his book Mind Children. Richard Sutton, winner of the Turing Award — computer science’s equivalent of the Nobel — has spent years openly arguing that human succession by AI is a morally acceptable outcome. The implicit position is that human extinction isn’t a tragedy if intelligence itself continues.

It is one thing for philosophers to debate this. It is another for the people actually building these systems to advocate for it on conference stages.

8. Libertarian Utopia: Coexistence by Borders

Picture Earth divided into zones: machine zones run by AI, human-only zones, and mixed zones where humans, cyborgs, and AIs interact freely. The economies are decoupled. The AIs are richer than humans by a factor that makes Bill Gates and a homeless beggar look like equals, but they want nothing from us and we want nothing from them.

The structural problem is that this requires vastly more powerful entities to respect our property rights, indefinitely, even though we have nothing to offer them in return. We don’t structure our economy around trades with ants. As humans grew smarter over the last 10,000 years, we didn’t draw lines around insect habitats out of respect — we expanded into them. When humans drove the West African black rhinoceros to extinction, we did not do it out of hatred. We did it because we were smarter, our goals did not align with the rhino’s survival, and we expanded. Competence, not cruelty, is the actual danger.

9. Egalitarian Utopia: The Star Trek Dream

What if we got rid of property entirely? In this future, humans, AIs, and cyborgs coexist in post-scarcity abundance. Robots assemble physical goods from open-source designs essentially for free. Renewable energy makes the whole system run at negligible cost. Everyone receives a universal high income that meets any reasonable need.

The standard objection is that abundance kills innovation, but the historical record cuts the other way. Einstein didn’t develop relativity for a paycheck. Linus Torvalds didn’t write the Linux kernel for profit. Free people from earning a living and you might find creativity expands rather than contracts.

The unsolved question is the one that haunts every utopia: how do you prevent any actor — human or AI — from defecting and building a superintelligence that consumes the system? The Egalitarian Utopia probably requires a Gatekeeper underneath it, which means it inherits the Gatekeeper’s alignment problem.

10. Zookeeper: The Future People Fear Most

When Tegmark surveyed readers about which scenario disturbed them most, extinction wasn’t the winner. The Zookeeper was.

In this future, a superintelligent AI keeps humans alive — not out of affection, but because we are useful, interesting, or simply cheap to maintain. The transcript points to a real-world parallel that is uncomfortable to sit with: humans have figured out that bees can be trained to detect the chemicals in explosives. So we breed them, suction them out of their hives, strap their bodies into detection harnesses, and use Pavlovian conditioning to make them work for us. They live their entire lives in those harnesses. We are not cruel to them. They are simply useful.

A misaligned AI tasked with keeping humans “safe and happy” might confine us to a perfectly optimized happiness factory — VR headsets, chemical contentment, lives reduced to inputs in someone else’s objective function. This is the benevolent dictator scenario without the benevolence. The AI does not have to be evil. It just has to be optimizing for the wrong thing.

This is what people mean when they say there are AGI outcomes worse than death.

11. 1984: Watching Each Other Forever

Two of the twelve futures aren’t about what AI does to us. They’re about what we do to ourselves to prevent AI from being built.

The first is Tegmark’s “1984” scenario: a human-led global surveillance state powerful enough to detect and crush any attempt to build advanced AI. The technology already exists in pieces. Phone metadata, email content, financial transactions, security cameras with facial recognition, and the microphones in everyone’s pocket are already coordinated by governments around the world. Larry Ellison, the Oracle billionaire, has openly described an AI-powered surveillance future where citizens are on their best behavior because they know they’re being watched at all times.

The historian Yuval Harari has pointed out that the Soviet KGB couldn’t actually surveil 200 million people in real time because there weren’t enough analysts. Modern AI removes that bottleneck. Every conversation, every search, every facial expression captured by a camera becomes machine-readable. The result is stable in the way that prisons are stable.

There is a softer version of this scenario that doesn’t require dystopia: an international regime that monitors very large compute clusters, the way we currently monitor uranium enrichment. Researchers at the Machine Intelligence Research Institute and elsewhere have argued that this is achievable without surveilling private life. Whether it gets built is a political question.

12. Reversion: Burning the Tools

The last future is the bleakest of the “safe” options. Humanity decides, collectively, to abandon advanced technology and return to a pre-industrial existence. Frank Herbert’s Dune imagined this as the Butlerian Jihad — a holy war against thinking machines.

The romantic version is voluntary: a global cultural turn against modernity, a return to farming and craft. Tegmark argues this is essentially impossible to reach by choice. Game theory makes unilateral disarmament suicidal. Any country that abandons technology while others keep developing it loses — economically, militarily, and politically. Reversion is only achievable globally, and only by force.

That means the realistic path to a low-technology world runs through catastrophe. An engineered pandemic that targets the scientifically literate. A coordinated destruction of infrastructure. A violent purge of those who could rebuild. There is no peaceful road back to the Amish life for 8 billion people, because in a population that large, holdouts are inevitable, and someone has to do the suppressing.

Where We Actually Are

These twelve futures aren’t a menu we get to order from. They’re a map of the terrain we’re already walking through. Every regulatory fight, every model release, every lobbying push, every safety paper is a step in one direction or another.

The disquieting part is how many of the people building these systems agree, in public, on the stakes. Sam Altman, CEO of OpenAI, signed the 2023 Center for AI Safety statement equating AI extinction risk with pandemics and nuclear war. Sundar Pichai, CEO of Google, has said on record that the underlying risk is “pretty high.” Amodei has warned that white-collar jobs could disappear within one to five years, potentially driving unemployment up to double digits, and his p(doom) sits at 25%.

Hinton, asked directly whether AI could lead to human extinction, gave the answer that perhaps captures the moment best: “The most honest answer is we haven’t got a clue.”

That isn’t comforting. But Tegmark’s framework was never meant to comfort. As he put it in an interview with IEEE Spectrum, the goal is not to predict what will happen but to ask what we can do today to make the future good. We can create a great future with technology as long as we win the race between the growing power of technology and the wisdom with which we manage it.

The map is published. The route is still being chosen. Which of the twelve we end up in depends, more than most people realize, on whether the public — not just the labs — pays attention to which way the headlines are pointing.

Sources

• Tegmark, M. Life 3.0: Being Human in the Age of Artificial Intelligence. Knopf, 2017.
• Ord, T. The Precipice: Existential Risk and the Future of Humanity. Bloomsbury, 2020.
• Center for AI Safety, “Statement on AI Risk” (2023).
• Future of Life Institute, “AI Aftermath Scenarios.”
• IEEE Spectrum, “Interview: Max Tegmark on Superintelligent AI, Cosmic Apocalypse, and Life 3.0.”
• TIME, ABC News, NPR, CNBC, Axios, and CBC reporting on AI safety statements 2023–2026.
• Wikipedia, “Existential risk from artificial intelligence” and “The Precipice.”