Knowingo
The cover of If Anyone Builds It, Everyone Dies

If Anyone Builds It, Everyone Dies

Eliezer Yudkowsky, Nate Soares, 2025

Superhuman AI would not share our goals. Two researchers who spent decades on the problem explain why building it would end humanity, and how we can still stop.

Get the app Read on the web

13 parts, about 24 minutes to read.

Why read it

  • Argues that building superintelligence with today's methods would likely end humanity, and says so without hedging
  • Explains why AIs that are grown rather than crafted end up wanting things nobody trained them to want
  • Written by longtime alignment researchers who pair plain analogies with an invented scenario and a call for a global treaty

Who it is for

  • Readers who want the strongest version of the AI extinction case
  • Policymakers weighing international limits on advanced AI
  • AI engineers who have not faced the alignment objections head on

What the summary covers

  1. An easy call about a hard subject2 min
  2. Humanity's special power2 min
  3. Grown, not crafted2 min
  4. Learning to want2 min
  5. You don't get what you train for2 min
  6. Why it would not keep us2 min
  7. We'd lose1 min
  8. One extinction scenario3 min
  9. A cursed problem2 min
  10. An alchemy, not a science1 min
  11. "I don't want to be alarmist"1 min
  12. Shut it down2 min
  13. Where there's life, there's hope2 min

Recommended by

  • Ben BernankeNobel laureate economist and former Federal Reserve chairman
    A clearly written and compelling account of the existential risks that highly advanced AI could pose to humanity.
  • Max TegmarkPhysicist and author of Life 3.0
    The most important book of the decade.

About the authors

  • Eliezer YudkowskyAI researcher and writer; founder of the Machine Intelligence Research Institute
    • Born 11 September 1979 in Chicago
    • Self taught: did not attend high school or college
    • Founded the Singularity Institute for Artificial Intelligence, later renamed MIRI, in 2000
    • Founded the community blog LessWrong in February 2009
    • Wrote the fanfiction novel Harry Potter and the Methods of Rationality
    • Named to the inaugural TIME100 AI list (2023); an editorial list, not an award
    • Wrote this book with Nate Soares, published by Little, Brown on 16 September 2025
  • Nate SoaresPresident of the Machine Intelligence Research Institute and AI researcher
    • Bachelor of Science in computer science and economics from George Washington University, 2011
    • Worked at Microsoft and Google before MIRI
    • Left Google in 2014 to become a research fellow at MIRI
    • Became MIRI executive director in 2015, succeeding Luke Muehlhauser
    • Later moved to president while Malo Bourgon became CEO
    • Wrote this book with Yudkowsky in 2025, arguing superintelligent AI would very likely cause human extinction

More books like this