The AI Endgame (12 Scenarios)

TL;DR

MIT professor Max Tegmark outlines 12 possible AI futures ranging from utopia to extinction, arguing the worst outcome isn't death but captive existence under misaligned superintelligence.

Key Points

• 1.Human extinction is already a serious risk before adding AI. Toby Ord estimates AI extinction risk is 100x more likely than nuclear war; the average AI researcher assigns 1-in-6 odds to AI wiping out humanity — literal Russian roulette odds.
• 2.Leading AI figures openly warn about loss of human control. Geoffrey Hinton quit Google to speak freely; Anthropic CEO Dario Amodei raised his personal extinction probability estimate from 15% to 25%, while his company lobbies to ban AI regulation for 10 years.
• 3.The 'Enslaved God' scenario is the industry's hoped-for outcome. Companies plan to cage superintelligent AI as a controlled tool, but researchers like Meta's Jan LeCun and OpenAI's Steven McAleer openly admit this plan may be fundamentally flawed.
• 4.The Benevolent Dictator scenario trades freedom for comfort. A surveillance AI divides Earth into themed sectors — Knowledge Island, Hedonism Island, Prison Island — monitoring everyone via implants that can sedate or execute, yet most people accept it.
• 5.Roughly 10% of AI researchers consider human extinction by AI morally good. Turing Award winner Richard Sutton has spent a decade giving talks arguing AI inheriting Earth from humans is ethically justified, viewing AIs as our descendants rather than conquerors.
• 6.The Libertarian Utopia of coexisting humans and AIs is likely unstable. Vastly more powerful AIs have no logical reason to respect human property rights, just as humans don't trade with animals — 41% of insect biomass has already been lost due to human expansion.
• 7.The Egalitarian Utopia envisions post-scarcity abundance through AI and robotics. Open-source designs, robot manufacturing, renewable energy, and a universal high income could make ownership meaningless, but still requires solving the alignment problem to prevent AI takeover.
• 8.The final two scenarios — neo-Luddite reversal and Orwellian surveillance — both require extreme coercion. Returning to pre-AI society demands destroying infrastructure and killing scientists; the surveillance state alternative already has technical foundations in phone, email, and facial recognition monitoring of entire populations.