A1. Probabilities of Human Extinction
Existential Threat Likelihood summarizes this-century risks for Earth alone (early/mid-21st century), using outcome classes and order-of-magnitude (OOM) estimates rather than false precision. The framing: anthropogenic hazards dominate; classic “planet-killers” are rare but perfectly target a single-world species.
Outcome classes and decomposition
Global Catastrophic Risk (GCR): billions affected / civilizational collapse, but eventual recovery plausible.
Existential Risk (X-risk): permanent and unrecoverable loss of humanity’s potential (extinction or locked-in dystopia).
Decomposition (planning lens): for hazard i, risk scales as
Hazard × Exposure × Vulnerability × Cross-hub Correlation (ρ).
When ρ≈1 (highly correlated across hubs), a second world helps little; when ρ≈0 (de-correlated), a second, self-sufficient hub sharply cuts species-level failure.
Anchor baseline. Syntheses in The Precipice place this-century existential risk at roughly ~1 in 6, with anthropogenic drivers dominating naturals by orders of magnitude. Treat this as a driver, not an exact forecast.
Hazard bands
(Earth only, 21st century — order-of-magnitude; X-risk = existential)
Reading the table. The dominant existential mass lies in AI and engineered bio—anthropogenic, high-ρ hazards—while classic planet-killers (asteroids, VEI-8) are rare but are exactly the threats where a single-planet species has a binary outcome.
Hazard | Century likelihood (OOM band) | Outcome class | Cross-hub correlation (ρ) | Planning note |
Advanced AI misalignment | ~10^-2 – 10^-1 (few–ten %) | X-risk | High unless standards, compute, and deployment are de-correlated | Global monocultures make ρ→1; revocation and heterogeneity matter. (e.g., ~5% median “extinction or similarly permanent disempowerment” from surveys of ~5,000 ML authors) |
Engineered biological risks | ~10^-2 (few %) | X-risk / high-end GCR | High with open traffic; reducible with quarantine & on-hub countermeasures | Countermeasure platforms and incubation-covering windows set vulnerability |
Natural pandemics | ≲10^-4 (tail to existential scale) | X-risk tail / GCR | Medium (travel-dependent) | Existential tail via compounding failures, not pathogen alone |
Nuclear war → winter | ~10^-3 (existential tail) | X-risk tail / high GCR | Low (Earth-centric) | Food systems and non-synchronous hubs cut species-level exposure |
Climate change (tail) | ~10^-3 (tail) | X-risk tail / high GCR | Medium | Cascades and conflict drive risk; deep uncertainty at the top end |
Extreme space weather | O(10^-2) per decade | GCR | High by Sun; impact design-dependent | Hardening, islanding, underground ops reduce vulnerability |
Asteroid/comet ≥10 km | ~10^-6 | X-risk tail | Low | Detection and deflection improving; classic single-planet binary |
Supervolcano (VEI-8) | ~10^-4 (e.g., Yellowstone ~1 in 730,000 per year) | High GCR / X-tail | Low | Rare but Earth-bound; food/energy buffers dominate outcome |
Nearby SN/GRB | ~10^-7–10^-6 | GCR / X-tail | High | Deep-time concern; century-scale planning mostly unaffected |
All naturals combined | ~10^-4 | Mostly GCR; small X-tail | Mixed | Rare individually; together still below anthropogenic mass |
Notes and planning implications:
Use ranges, state uncertainty. Ord’s canonical bands (AI ~1/10; engineered bio ~1/30; nuclear ~1/1000; climate-tail ~1/1000; naturals aggregate ~1/10,000) are intentionally rounded; keep wide credible ranges and avoid false precision.
Correlation (ρ) is the killer variable.
High-ρ: unaligned AI; some bio with open travel; stellar explosions; severe space weather. Here multiplanetary helps mainly by buying recovery options and enabling independent governance.
Low-ρ: nuclear war, asteroids/comets, supervolcanoes—prime wins for an independent Mars hub. DART + NEO Surveyor show planetary defense improving, further shrinking that slice.
Nuclear-winter food shock is the mechanism to watch. Modern crop/fishery models project multi-year calorie deficits up to civilization-scale famine; risk management should emphasize resilient food systems and decoupled supply chains.
Climate tails are about cascades. AR6 emphasizes compound and low-likelihood high-impact events; extinction-scale remains a tail-tail scenario but systemic fragility drives GCR.
Asteroids: detection + deflection now credible. >90% of >1-km NEAs cataloged; kinetic impact validated (−32 min at Dimorphos); NEO Surveyor accelerates discovery of ≥140 m class.
Supervolcanoes: extremely low odds. USGS places Yellowstone’s annual caldera-forming eruption at ~1/730,000; still a classic single-planet risk.
Space weather: big GCR, not X-risk. Competing estimates (≈0.5–12% per decade) justify hardening power, comms, and life-support; design for graceful degradation.