Nobody has to hate us
In 2016, AlphaGo played a move against Lee Sedol that everyone thought was a mistake. DeepMind said a human would pick it maybe 1 time in 10,000. A hundred moves later, it won the game. It won the match 4 to 1.
No rebellion. No evil robot. It was just better at the game, and that was enough. I’m not saying AlphaGo means machines will take over. It was one narrow system built to do one thing extremely well. But it showed something useful: humans can lose the top spot without a war.
Replacement doesn’t need hate
In movies, AI is always the enemy. Skynet, HAL, the Matrix machines. An enemy is easy. You find the switch and fight back.
But cars didn’t hate horses. Digital cameras didn’t hate film. Something just worked better, and the world moved on.
This idea is older than computers. In 1863, Samuel Butler wrote that we were creating our own successors. Nobody has to plan it.
A farmer who refuses the better machine can preserve his independence and lose his farm. A company that refuses automation can preserve jobs and lose customers. A country that refuses a military advantage can preserve a principle and lose a war.Everybody makes the sensible choice for themselves, and together that adds up to handing over the keys.
The big catch
Software needs buildings, chips, power, cooling, repairs, permits and lawyers. All of that is still human. An AI that can’t get a transformer is still just a tool.
So the real line isn’t “AI gets smart.” It’s when AI starts running the loop: deciding what to build, where the money goes, which supplier to pick, and when a human objection isn’t worth the delay. We’re nowhere near proving that will happen. It might never happen.
What sci-fi got right
Different writers each take away one comfort:
Stanisław Lem: We stop being important. A much smarter mind might simply find our projects small.
Iain M. Banks: Losing control could feel great. Humans in the Culture are safe, free and rich. The Minds still run the place.
Vernor Vinge: Surviving is not the same as being in charge. We could keep the title and lose the power, like a king after the constitution.
Greg Egan: “Humans” and “the successor” may blur together through upgrades, copies and digital minds. The thing replacing us may still have our family name.
Charles Stross: The successor does not need to want anything. Markets already reward whatever is faster, cheaper and easier to scale. This last one bothers me the most. A machine does not need ambition if the people buying it do.
Why I care about long tasks
A chatbot sounding conscious doesn’t worry me much. What matters is how long a job an AI can do on its own.
METR has tried to measure that. Its work found the length of tasks frontier AI systems can complete autonomously doubling roughly every seven months over several years. METR also warns that its tests lean heavily toward software tasks and become unreliable for very long jobs, so don’t read too much into it.
Civilization is just long tasks chained together: design the chip, sign the contract, move the money, build the data center. If enough of those loops close without a human, and the companies that cut humans out start winning, the pressure does the rest.
Hostility isn’t the only warning sign
Everyone watches for AI that lies or fights shutdown. Fine, those matter. But also watch the boring stuff: who gets the contract when the machine is cheaper, and who owns the servers.
Saying no to AI won’t vanish overnight. It’ll just get expensive.
Nobody will press a button that says GIVE CIVILIZATION TO THE MACHINES. There will be procurement forms.
The friendly version
People still play Go. The board wasn’t confiscated. We study the machine’s moves and keep placing stones.
We might end up like that: fed, safe, free, even loved, with a hand on the controls and nothing important depending on it.
Sources and notes
1. AlphaGo and Move 37. Demis Hassabis, “What we learned in Seoul with AlphaGo,” Google, March 16, 2016. This is the contemporary first-party source for the 4–1 match result and DeepMind’s statement that Move 37 had roughly a one-in-ten-thousand chance of being played by a human. Demis Hassabis, “From games to biology and beyond: 10 years of AlphaGo’s impact,” Google DeepMind, March 10, 2026, supplies the retrospective description of professional commentators initially reading Move 37 as a mistake, its importance roughly one hundred moves later, and the continuing study of AlphaGo’s moves by human players. 2016 Google article 2026 Google DeepMind retrospective
2. Samuel Butler. Samuel Butler, writing under the name “Cellarius,” “Darwin Among the Machines,” The Press, Christchurch, New Zealand, June 13, 1863. The National Library of New Zealand’s Papers Past archive reproduces the original newspaper letter. A later Press article also identifies Butler as the author and notes that the letter became part of the germ of Erewhon. Original 1863 newspaper page at Papers Past
3. Stanisław Lem. Stanisław Lem, Golem XIV, Kraków: Wydawnictwo Literackie, 1981. Parts of the Golem material had appeared earlier in Wielkość urojona in 1973; the standalone Golem XIV appeared in 1981. The official Lem site describes Golem as a future supercomputer that vastly exceeds human intelligence, reconstructs itself, criticizes humanity and departs the human world toward forms of cognition beyond human reach. The official descriptive page is undated, so the publication year is taken from the bibliographic record rather than inferred from the website. Official Stanisław Lem page for Golem XIV Bibliographic record for the 1981 first edition
4. Iain M. Banks and the Culture. Paul Wiseall, “Iain M. Banks Interview – Part One,” Fantasy-Faction, October 26, 2012. Banks describes the Culture as a post-scarcity civilization that is populated largely by humans but run by artificial intelligences called Minds. “Iain Banks email Q&A August 2008,” Hachette UK, August 29, 2008, contains Banks’s comparison of humans to pets, a fish tank, an ant farm and The Sims, his explanation of the Minds’ duty of care, and his statement that some Minds leave, which the Culture accepts as the price of creating genuine independent AIs. 2012 Fantasy-Faction interview 2008 Hachette Q&A
5. Vernor Vinge. Vernor Vinge, “The Coming Technological Singularity: How to Survive in the Post-Human Era,” conference paper in Vision 21: Interdisciplinary Science and Engineering in the Era of Cyberspace, NASA Lewis Research Center. The symposium was held March 30–31, 1993; NASA lists the proceedings publication date as December 1, 1993. Vinge’s abstract distinguishes the ending of the “human era” from the possibility of human survival. In the paper he states that physical extinction is one possible outcome but may not be the most frightening one. NASA Technical Reports Server record
6. Greg Egan. Greg Egan, Diaspora, London: Orion/Millennium, 1997. Egan’s official site supplies the synopsis and publication history. It describes Yatima as a digital citizen of a polis, the Coalition of Polises as the computer network where most solar-system life resides, robot bodies used by digital citizens, and biological “fleshers” living in physical communities. Greg Egan’s official Diaspora page and publication history
7. Charles Stross and Accelerando. Charles Stross, Accelerando, 2005. Stross’s own bibliography identifies the novel as a 2005 publication assembled from nine stories published between 2001 and 2004. In “Crib sheet: Accelerando,” Charlie’s Diary, May 28, 2013, Stross explains the book through competitive capitalism pushing beyond the point where merely human actors can compete, then explicitly warns that exponential trends often flatten into sigmoid curves. In “Dude, you broke the future!”, Charlie’s Diary, January 2, 2018, based on his December 2017 keynote at the 34th Chaos Communication Congress, he rejects the usual humanlike singularity model and points toward self-optimizing systems that may remain unlike human minds. Stross bibliography and Accelerando publication information 2013 Accelerando crib sheet 2018 Dude, you broke the future!
8. METR task-completion time horizons. METR, “Measuring AI Ability to Complete Long Software Tasks,” March 19, 2025, reports that the measured 50 percent task-completion time horizon of frontier model agents had approximately doubled every seven months over the preceding six years. METR’s live “Task-Completion Time Horizons of Frontier AI Models” page, last updated May 8, 2026, defines the measure and warns that measurements above sixteen hours are unreliable with its current task suite. METR’s January 2026 limitations note describes the most defensible long-run estimate as one doubling every six to seven months and stresses that the measure primarily tracks software and research tasks rather than arbitrary real-world work. March 2025 METR study METR live time-horizon page, updated May 8, 2026 METR limitations note, January 22, 2026