As AI explodes, humanity must accelerate its adaptation or reach a collective, enforceable decision to slow the development of these technologies
The hype cycle around artificial intelligence … Tweets: Andrew Leigh / @aleighmp : “In a 2022 survey, AI experts were asked, ‘What probability do you put on human inability to control future advanced AI systems causing human extinction?’ The median reply was 10%.” An important piece by @ezraklein: https://www.nytimes.com/...
Context & Ripple Effects
Ezra Klein's column lands eight days before the Vox case for slowing down AI makes the same argument at greater length — that 'inevitability,' the arms race with China, and safety-by-experimentation are choices dressed up as physics. The empirical anchor both lean on is the 2022 expert survey Andrew Leigh amplified: a median 10% probability that loss of control of advanced AI causes human extinction.
What makes the piece a hinge rather than another op-ed is what follows it: the New Yorker profiling the doomers and effective accelerationists as named camps, the tech industry's prosperous generative-AI narrative crowding out catastrophic-risk warnings through 2024, and eventually Anthropic itself asking top labs to weigh pausing development. The binary Klein poses — adapt faster or stop together — became the actual axis of the debate.
First-order effects
- Klein converts a niche expert-survey statistic into mainstream-policy vocabulary, forcing every subsequent AI argument to pick a side on whether development speed is a collective decision or a race outcome.
Second-order effects
- Labs get squeezed between the two horns of his dilemma: Anthropic ends up marketed as the sector's superego while still needing commercial wins, and the industry's practical-AI promotion actively drowns out the catastrophic-damage messaging his column legitimizes.
Third-order effects
- If the pattern holds, the enforcement gap becomes the story — security researchers remain divided while global regulatory frameworks lag, leaving voluntary lab restraint like Anthropic's pause proposal as the only live brake, which is precisely the 'enforceable decision' Klein says humanity must otherwise reach deliberately.
The trend: AI governance is moving from whether-to-regulate debates toward the question Klein framed in 2023: coordinated slowing versus accelerated human adaptation as the only two exits from an arms race labs themselves now flag.