Anthropic researcher says more than 10% chance AI "could kill all humans"
A former Anthropic researcher’s claim that there is “more than a 10% chance” advanced AI could kill all humans has triggered sharp disagreement over how seriously such extinction risk estimates should be taken. Some see the rhetoric as sincere alarm from safety-focused insiders, while others dismiss it as cynical “scaremarketing” aimed at regulatory capture, investment, or post-resignation personal branding. Commenters debate how anyone can quantify such probabilities, whether AI is more likely to be an existential threat or a source of broad prosperity, and how its risks compare to long‑standing dangers like nuclear war and climate change.
Skepticism about the “>10% chance of human extinction”
- Many see the number as arbitrary or “ass‑pulled”; they ask why 10% rather than 1% or 50% and suspect there is no rigorous process behind it.
- Some argue that if people truly believed such odds, they would take far more drastic action than giving interviews or resigning, so the rhetoric feels like “cheap talk.”
- Others counter that beliefs don’t imply terrorism or sabotage; quitting a well‑paid job itself is already significant action.
Marketing, hype, and regulatory capture
- A large contingent views extinction talk as a marketing or lobbying strategy:
- “Our product is so powerful it might end civilization” → hype for investors.
- Doom framing as a way to justify heavy regulation that locks in current leaders.
- Others push back that normal companies minimize risk, not loudly advertise it, and note that multiple labs and researchers have been consistently worried for years, suggesting sincerity rather than pure PR.
How could AI cause catastrophe?
- Some say they haven’t yet seen a clear, rational path from super‑LLMs to extinction.
- Proposed routes in the thread:
- AI‑enabled hacking of critical systems (nukes, bio labs), or helping a rogue actor design a lethal pathogen.
- Misaligned optimization (paperclip‑style) where an advanced system pursues a goal with total disregard for human values, outcompeting humans for resources.
- AI‑driven misinformation, social disruption, and economic collapse leading to mass death.
- Others think the bigger risk is still humans using AI as a weapon or putting it in charge of infrastructure it shouldn’t control.
Comparisons to nuclear risk and other existential threats
- Some compare AI to the nuclear arms race but note key differences: uranium is naturally gated; AI is easily proliferated.
- Several argue humans, climate change, pandemics, and nukes are more likely extinction vectors than AI itself.
Optimism and mixed views
- A few place their own “P(doom)” around 10% but also assign high probability to AI greatly improving life and reducing other existential risks.
- Others see current doomerism as overblown, emphasizing humanity’s historical trend of improving living standards and expecting regulation to eventually emerge, albeit slowly.